<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Eric Loeb on Medium]]></title>
        <description><![CDATA[Stories by Eric Loeb on Medium]]></description>
        <link>https://medium.com/@ericloeb?source=rss-45f9774115------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*HBZpRqUGlDevAY0WwX6Aqw.jpeg</url>
            <title>Stories by Eric Loeb on Medium</title>
            <link>https://medium.com/@ericloeb?source=rss-45f9774115------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 00:19:52 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ericloeb/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Of Snake Oil and Dashboards]]></title>
            <link>https://medium.com/@ericloeb/of-snake-oil-and-dashboards-51f4f0a81787?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/51f4f0a81787</guid>
            <category><![CDATA[dashboard]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[control-theory]]></category>
            <category><![CDATA[feedback]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Sat, 18 Nov 2017 18:53:13 GMT</pubDate>
            <atom:updated>2017-11-19T22:12:51.961Z</atom:updated>
            <content:encoded><![CDATA[<p>We create a feedback loop when we inform a policy with a dashboard that is built from data that are sampled from the process impacted by the policy. These feedback loops are common, but neither the acquirers nor the developers of dashboards are typically aware that the feedback loops they are creating can easily become unstable. This and misaligned incentives in the analytics market are likely to produce <a href="https://www.nytimes.com/2016/11/10/technology/the-data-said-clinton-would-win-why-you-shouldnt-have-believed-it.html">some unfortunate results</a>.</p><p>Commercial analytics engagements typically devolve to <a href="https://www.klipfolio.com/resources/dashboard-examples">creation of a dashboard</a>. Dashboards rarely distinguish statistically significant changes from insignificant fluctuations, nor do they come with any warnings pertinent to the data collection, Nyquist limits, or model shortcomings. The proud owners of dashboards (and the like) expect they can understand and control their organizations using the dashboards, which, unfortunately, present numbers that may or may not mean what they seem to mean.</p><p>The market for Data Science <a href="https://www.forbes.com/sites/louiscolumbus/2016/08/20/roundup-of-analytics-big-data-bi-forecasts-and-market-estimates-2016/#749816036f21,%20https://thenextweb.com/insider/2016/03/11/business-intelligence-dashboard-dead/#.tnw_X6inEzYM">is projected by Forbes to grow</a> to $189B in 2019 with business intelligence and dashboards accounting for $20B in 2018. These projections are consistent with <a href="https://www.hubspot.com/marketing-statistics">other market tracking data</a>, the <a href="http://www.gartner.com/it-glossary/dashboard,%20http://global.qlik.com/us/landing/go-sm/resources/gartner-magic-quadrant-2017">flooded business intelligence market</a> and the <a href="https://www.forbes.com/sites/davelavinsky/2013/09/06/executive-dashboards-what-they-are-why-every-business-needs-one/#10f09e0f37d1">level of hype</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*6aSsnsHRSmMKA2B_B3kbhA.png" /><figcaption>Figure 1: a convolution network</figcaption></figure><p>Data Science is difficult to do correctly and there are many <a href="http://www.kdnuggets.com/2017/01/current-state-automated-machine-learning.html">easy ways to do it poorly</a>. In Figure 1 (<a href="https://www.linkedin.com/pulse/why-convolutioanl-neural-networks-shamane-siriwardhana">from here</a>) we see the structure of the popular <a href="https://en.wikipedia.org/wiki/Convolutional_neural_network">convolution neural network,</a> which is used effectively for a variety of machine learning problems. This system can be trained to mimic essentially any set of input/output pairings. Figure 1 depicts a network with 576 binary inputs. That means it works in the space of 2⁵⁷⁶ = 243 sexquinquagintillion possible combinations of the 576 inputs to find combinations of input variables that predict the desired outputs. We can train <a href="http://www.subsubroutine.com/sub-subroutine/2016/9/30/cats-and-dogs-and-convolutional-neural-networks">the neural network to recognize pictures of cats</a>, but it will also pick up on any spurious correlations between the inputs and outputs — such as the commonality of close-ups when cat owners post cat pictures on the internet. It is very easy to fit a model and provide a fast report on the significant model factors without attending to the issue of spurious correlations.</p><p>Machine-learnt models also do not necessarily mean what we think they mean. A statistical model is necessarily at best a model of what we think we’re trying to model PLUS whatever processes were used to gather the input data. A model of data that’s called “shipping time” is actually a model of the recorded shipping times plus the processes by which we measured shipments leaving and shipments arriving. A model of a <a href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC2709750/">data field called “suicide ideation”</a> is actually a model a whole series of complex processes that eventually led a database entry.</p><p>Machine learning algorithms will find correlations wherever they reside, and it is easy for variations in measurement to be correlated with the phenomenon of interest. If some delivery agents in a shipping network are rewarded for fast delivery times, those agents will at least be more attentive to measuring delivery and may also find ways to click their “delivered” button a little early. Demographic correlates of suicidal ideation can also correlate with variations in life experience, attitudes, availability medical services, how the doctors will interpret what they have heard, and so on. It is usually not possible to tease apart confounding factors in a post hoc analysis, but most commercial practitioners don’t even try.</p><p>Unfortunately, on top of all this, the innate human tendency to <a href="https://www.wired.com/2011/03/why-do-we-tell-stories/">make stories out of noise</a> makes the problem of false conclusions even worse. I once worked with a team of smart people to puzzle out a 3-way interaction. We spent almost an hour hashing out what was going on until we finally had a simple explanation for why it was happening. It was an interesting phenomenon, and we were all excited to have found it. Then, being the designated Grinch, and I added a counter-theory under the assumption that the interaction was a spurious correlation. The test of the counter-theory revealed that the interaction was an artifact of the data collection. Debugging that spurious result was a data science success, but it is a kind that takes more discipline and is less fun than simply believing the stories that come to mind when we assume that the names of the variables in a database are something more than random symbols.</p><p>The scientific method tells us to be skeptical about our results, but it is exceedingly difficult to apply skepticism in commercial practice. Within a corporation or government agency, the individual responsible for purchasing analytics will have an incentive to show that the analytics were valuable to the organization. The individual will want to report impressive results, and the charlatans inflating the analytics bubble are giving us ever-higher baseline expectations as to what constitutes an impressive result. The purchaser and provider incentives are aligned around doing the natural and easy thing: just believe the most interesting things the machine seems to be telling you. Fighting misaligned incentives to get at the (less impressive) truth is not a commercial win.</p><p>The problems of false conclusions will persist. It is tedious to get into the details of how data is collected, and few executives will want to provide the sustained attention to detail needed to cut through organizational barriers to get at the truth. Business data is accumulated without a lot of oversight into the methodologies employed. The data also typically reside in disparate silos throughout the organization. Good analysts should examine data collection methodologies before concluding that the apparent machine-learnt results are true, but the data collection belong to groups distributed across the organization. In what world do these people take kindly to discussing the veracity of their data with junior people from some other part of the organization? Validating data collection methodologies does not advance anybody’s career, it is tedious, and it requires top cover from executives; so, it doesn’t happen.</p><p>To make matters just a bit worse, there is no accurate way to collect data. Every approach will come with some tradeoff in cost, speed, and precision. For example, one of the US military services has two distinct methods of tracking which unit a soldier is deployed to. One system, used for internal reports, is closely controlled and monitored. It can take weeks to add a new unit in this system. The other system is highly malleable. It only takes a phone call from the unit commander to make a change to this system. The former system is cleanly structured, but it is unable to account for all soldiers. The later system is sometimes poorly structured, but it accounts for 100% of the soldiers in recent months. Because the later system is not regulated and tracked, it is unable to account for increasing numbers of soldiers as we look back in time. The military service faced a trade-off between meeting structure rules but missing some soldiers or else accounting for all soldiers in a haphazard structure. It wisely chose to do both, but now needs to bring the two systems together to help out the analysts and military historians.</p><p>For all of these reasons, I agree that Data Science may soon <a href="https://spectrum.ieee.org/robotics/artificial-intelligence/machinelearning-maestro-michael-jordan-on-the-delusions-of-big-data-and-other-huge-engineering-efforts">become a popped bubble</a> like Lean Six Sigma (LSS). As with LSS, irresistible forces push Data Science into the <a href="https://kore.ai/bots-make-analytics-insights-reality-decision-makers/">fast, easy, wrong, and over-hyped implementations</a> that undermine the core truths that originally <a href="http://www.businessinsider.com/why-big-data-is-just-a-fad-2015-3">gave the business fad value</a>.</p><p>Nevertheless, there are still a lot of dashboards out there. Can we use them to make decisions?</p><p>The accumulated data of any organization are typically collected on a regular schedule, tied, for example, to bi-monthly payroll or quarterly reporting. Accurate forward projections of this data should be limited by <a href="https://en.wikipedia.org/wiki/Nyquist%E2%80%93Shannon_sampling_theorem">the Nyquist theorem</a>, which places a 2x limit on the frequency of sampled signals we can correctly interpret. In other words, a monthly data sample will misinterpret phenomena that occur more rapidly than once every <strong>two<em> </em></strong>months. A quarterly report can help us understand phenomena with six month and slower frequencies.</p><p>When presented with too-fast signals, regular samples will be “aliased”, which means that the too-high frequencies will seem to be lower frequencies (<a href="https://svi.nl/AntiAliasing">as seen here</a>). For example, an organization has a training class for new hires every week but samples the employee file every month for a dashboard. If the dashboard is used to understand the flow of new hires into the organization, it will have aliasing effects due to the weekly new employees signal being sampled monthly (8x too slowly). This problem can be solved by smoothing or filtering, or by creating a daily data sample, but the need to do these things are apparently not apparent to most dashboard developers.</p><p>The Nyquist limit is related to a delay in how quickly we can recognize changes in the environment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/425/1*JDCxc7EBnn8lxi6hvhrzNg.png" /><figcaption>Figure 2: sine waves shown in time and in frequency space</figcaption></figure><p>Figure 2 shows four sine waves in the top box and the Fast Fourier Transform (FFT) of those sine waves in the bottom box. We can see that the taller sine waves also have taller FFTs. The FFT represents the energy in each frequency bin, and because these four signals are just sine waves, there is only one frequency bin with any energy in it. So, the four FFTs are each just a signal that is mostly 0 except for one point (one frequency) with some non-zero energy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/370/1*Fxs6mBjLRjq1JmHJiSQBzg.png" /><figcaption>Figure 3: sine waves with different start times shown in time and frequency space</figcaption></figure><p>Figure 3 shows how it takes time even to recognize signals that are within the sampling limits. The box on top in Fig 3 shows a sine wave with three high frequency add-in signals starting near the end. The blue signal is present in the last 32 samples; the green signal is present in the last 16 samples; and the red signal is present in the last 8 samples. The bottom box shows the FFTs of the total (sine-plus-add-ins) signal. We can see that the blue, 32-sample signal is readily discernible as a bump on the right side. The green, 16-sample signal is at threshold detect-ability. The red, 8-sample signal is not discernible at all. If we suppose that the base sine wave represents some seasonal effect on some business metric that we are sampling bi-monthly, we still may not be able to discern the new high frequency (red) change in metric 4 months after it begins.</p><p>The mistake of projecting onto the next sampling interval is prevalent. For example, an annual survey of CEOs, <a href="http://www.pwc.com/gx/en/ceo-agenda/ceosurvey/2017/us">is mistakenly publicized as a projection</a> of what will occur in business over the following year. Changes from baseline in a random process should not be asserted without compelling statistical evidence. If we use monthly data to control a monthly process, then we are creating an <a href="http://www.wescottdesign.com/articles/Sampling/sampling.pdf">inherently unstable control loop</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_UGLJWQo_x_6v4Mfyzy5ZA.png" /><figcaption>Figure 4: simulation of a policy change viewed through a dashboard. The black line shows the average employee satisfaction score for a survey issued each month. The red line shows the 12-month moving average of those scores and the grey shading represents the standard deviation (smoothed squared residuals of the moving average). The simulation divides the employees into 5 satisfaction levels. At lower levels (1 and 2) the employee is almost certain to take the survey that month. Most of the employees are happy and do not take the survey (20%, 5%, and 10% for levels 3–5 respectively). The baseline before the policy change represents this process plus (0, 0.05 SD) Gaussian measurement error. Prior to the policy change, we see spurious apparent deviations from the baseline score of 3.75. The company has a steady state hiring and attrition rate of 1% (10 employees/month). At month #48 the company begins hiring half of the new employees from a new source (call it “Harvard”) that results in employees who are themselves happy with the firm, but who greatly irritate the other employees. At the start of the new policy, the scores of these new employees (accumulating at 5/month) are randomly adjusted upwards in their ratings, while 10% of them will have a negative interaction with one of the other employees, whose ratings are randomly adjusted down. This underlying process leads to increased variance in the satisfaction metric, an initial increase in average satisfaction, but accumulating discontent within the overall employee population.</figcaption></figure><p>Figure 4 shows graphics from a numerical simulation of a business process sampled monthly over eight years with a policy change at year four. The policy change, starting Oct, 2013, makes some new employees happy right away while causing increasing discontent among the bulk of the employees. Evidence of the positive impact of the new policy on employee satisfaction begins five months after the policy change, in March, 2014. The company’s growing discontentment first becomes evident two years after the policy change in Sept, 2015.</p><p>Business intelligence systems are not normally designed with attention to the control theoretic considerations that prevent feedback loops from becoming unstable. This and other misaligned incentives around the Data Science market are likely to produce some unfortunate results. I expect analytics will ultimately get a bad reputation because it is misapplied, not because it is a bad idea.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=51f4f0a81787" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What should progressives do about the ongoing Russian attacks?]]></title>
            <link>https://medium.com/@ericloeb/what-should-progressives-do-about-the-ongoing-russian-attacks-aaffd4560377?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/aaffd4560377</guid>
            <category><![CDATA[russian-hacking]]></category>
            <category><![CDATA[feedback]]></category>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[progressive]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Thu, 09 Nov 2017 00:41:45 GMT</pubDate>
            <atom:updated>2017-11-09T00:41:45.839Z</atom:updated>
            <content:encoded><![CDATA[<p>Here we are, gripping the edge our seats, wondering if our country’s democratic institutions will be strong enough to withstand the sustained assault. Things are looking better with <a href="http://fivethirtyeight.com/features/itll-take-some-time-before-the-mueller-indictments-affect-trumps-approval-rating/">Mueller’s first indictments</a>, but who knows what <a href="https://www.washingtonpost.com/politics/trump-heads-to-asia-where-concerns-are-growing-over-possible-war-with-north-korea/2017/11/03/5b805f48-c00a-11e7-959c-fe2b598d8c00_story.html?utm_term=.52568a66c28a">outrages lie in store</a>. One thing for sure: our country needs to be strengthened so as not to get thrown this close to the edge again.</p><p>As I’ve considered what our half of the political spectrum should do about the Russian attacks, I’ve concluded that we need to establish a mirror government that provides social services. This will be political infrastructure that can help us weather many predictable future storms; a steady source of employment for left-leaning operatives; and a fitting answer to the right wing’s prolific propaganda machinery.</p><p>As a collection of left-of-center political interests, it is not directly relevant for us that Russia is attacking our elections. While Russia’s attacks are undeniably important, the protection from these attacks and legitimate responses are government responsibilities. Engaging in a rag-tag militia defense against a cyber superpower won’t meaningfully alter the situation.</p><p>It is important that Russia is attacking us, but for progressives, this political attack is of a piece with the many other attacks from the right wing. We do not need to distinguish between Putin, Mercer, Murdoch, and the Koch brothers. The oligarchs generally perceive (apparently) that it is in their interests to subvert Democracy. It would be great if our government could stop Russian agents from sowing discord, but it cannot stop American billionaires. It would be great if our government could stop Russian spies from hacking into our vote tabulation systems, but the vulnerabilities they might have exploited have been known for years, and republican gerrymandering and voter suppression may have been sufficient to sway the election anyway.</p><p>What’s most important is that the Republican Party has been propagandizing US citizens with disinformation for decades. Falsehoods about carcinogens and climate change are not worse when they are funded by foreign sources, like an Australian news magnate, than the same lies funded by Texas oil billionaires. We do not need to distinguish. We need to recognize and respond to the fact that there are sustained, well-funded and well-organized right wing attacks on Democracy itself.</p><p>It would be unconstitutional for the US government to stop Americans from spreading lies. We progressives must bind together in an extra-governmental, superhuman force that will fight for Truth, Justice, and the American Way.</p><p>The key to this fight is a collection of positive feedback loops that work in the favor of Democracy’s enemies. As many have noted, destroying education leads to voters who are more easily duped into voting against education. Attacking government and governing poorly leads to loss of voter confidence in our government and willingness to believe that we need to cut the relatively inexpensive services that are helpful to the bulk of the voters. Sowing confusion and discord increases support for draconian policies that further increase confusion and discord. To stop these attacks, we need to alter the social dynamics that make disinformation, sabotage, and discord self-reinforcing.</p><p>Altering the dynamics of US politics will be a major undertaking. We will need to create massive new political machinery.</p><p>And although it would be satisfying, it would be a bad idea to copy the right-wing industry that systematically undermines democrats. The republicans are playing by an un-American set of rules in which only republican Presidents are legitimate. It feels like Justice to do the same back at them, but it would not be Truth and it is not the American Way. The tribal wars are at the core of the discord we need to fight. “<a href="http://www.abrahamlincolnonline.org/lincoln/speeches/house.htm">Every kingdom divided against itself is brought to desolation</a>” (Matthew 12:25). We need to bring our country back to unity, not work to tear it down.</p><p>So, let’s look at what would be needed to intervene in the bad dynamics that the oligarchs are exploiting:</p><ul><li>Cutting education. We want cuts in education to lead to better informed voters. This means we need an education funding source that increases as <a href="http://scholar.harvard.edu/files/rpande/files/caninformedvotersenforcebettergovernance.pdf">measures of voter informed-ness</a> decrease.</li><li>Broadcasting misinformation. We want <a href="http://www.ucsusa.org/our-work/center-science-and-democracy/disinformation-playbook#.WfZCWWhSyUk">disinformation campaigns</a> to be self-defeating. This means we need funds for counter-measures that increase in proportion to the falsehoods.</li><li>Distrust in government. We want increasing corruption and incompetence to give us better governance. The is might happen through voting, but we can’t count on that in the face of disinformation, gerrymandering, et al. Those who sabotage our government cannot be counted on to respect the sanctity of the vote. The only sure approach I can see is to build second government. As the right wing work to <a href="https://en.wikiquote.org/wiki/Grover_Norquist">drown the US government in a bathtub</a>, we can construct a better, modernized version.</li><li>Political fund-raising. Fundraising goals and media attention are most easily achieved by demonizing our opponents. The more political operatives demonize, the more they succeed in their careers. We cannot alter the conservative movement, but we must change the career incentives of our own operatives.</li></ul><p>To sum that up, we can break the vicious cycle that Democracy’s enemies are exploiting by tying some source of extra-governmental funding to metrics on the success of destructive tactics. In other words, the more destructiveness we see, the more counter measures we must fund. The more lies they tell, the more education and advertising we will fund. The more they demonstrate that those in authority cannot be trusted, the more our own officials will shine. As they tear down the US government, our progressive government will pick up the slack. Our operatives will work in that government, and we will fund that government, so that our operatives will advance by governing well, not by demonizing contrary points of view.</p><p>Our mirror government will provide the blue states with economies of scale for social services. The mirror government will be structured like the US government. The mirror government will create and manage disinformation countermeasures (such as anti-troll farms), provide assistance to progressive state governments, and it will be a new organizing framework that converts our current donation-based progressive political activities into voluntary-tax-based activities.</p><p>I estimate our mirror government would cost about $3.5B/year. This covers an employee structure that mirrors the legislative branch and the top nodes of the non-military executive agency org charts (25,000 employees in total); and another $1B/year to develop and deploy advances in disinformation counter-measures, and AI-enhanced governing and organizing methods. These revenues would come from voluntary progressive “taxes” of about $150/year on average from the 25M highest-income progressives. With a mirror legislature in place, we can take up other issues, such as funding for education. For comparison, <a href="https://www.charitynavigator.org/index.cfm?bay=content.view&amp;cpid=42">US charitable donations exceeded $390B in 2016</a>. We can do this.</p><p>If things go well, the mirror Congress will be at a minimum an (expensive) progressive version of the <a href="https://www.alec.org/">American Legislative Exchange Council</a> (ALEC), which provides templates of “conservative” legislation to red state legislatures. It is fitting that our version of ALEC would create legislation by open democratic means instead of … what ALEC does. Our mirror government will also provide oversight for our joint expenditures, which will be part of the new and better role for our operatives. They will be helping voters solve real problems, and they will help us more electorally than they ever did before.</p><p>If things do not go well, our mirror government will be indispensable. We will be ready with an organization that can defend the constitution. We will provide a place for actual conservatives (IE those who support American values like tolerance and free speech) to participate. And we will provide the coastal states with the federated coordinating and possibly financial resources they will need to confront the real problems that are coming our way, including those that are being <a href="http://www.chicagotribune.com/news/opinion/zorn/ct-kansas-brownback-trickle-down-tax-fail-zorn-perspec-0621-md-20170620-column.html">manufactured by republican governance</a>.</p><p>We will come to this eventually. California is pushing forward with the <a href="http://www.latimes.com/world/europe/la-fg-brussels-eu-jerry-brown-20171107-story.html">Paris Climate Accord independently</a>, and there is discussion of intra-governmental cooperation within the <a href="https://www.nytimes.com/2017/11/04/us/washington-state-west-politics-democrats.html">“Blue Wall” on the west coast</a>. The progressive movement will benefit from a national coordinating body.</p><p>The Russian attacks on Democracy are legally distinct, but not operationally distinct, from the attacks by right wing forces generally. As progressives, we can hope that any crimes committed by the President and his cronies will be punished, but we cannot expect that the ongoing assault on our freedoms will end there. We must organize for a long-term fight that the US government is constitutionally unable to engage in. We should also stop contributing to our country’s problems by pushing our operatives into a Darwinian brawl for funding. The best solution is to re-organize our side of the political spectrum around voluntary tax revenues instead of donations, with those tax revenues and disbursements being managed by an elected copy of the US government.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=aaffd4560377" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Can we recast citizen engagement as open source development of laws?]]></title>
            <link>https://medium.com/@ericloeb/can-we-recast-citizen-engagement-as-open-source-development-of-laws-a0fecc76f469?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/a0fecc76f469</guid>
            <category><![CDATA[civictech]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[computational-thinking]]></category>
            <category><![CDATA[civic-engagement]]></category>
            <category><![CDATA[software-development]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Sun, 01 Oct 2017 20:30:49 GMT</pubDate>
            <atom:updated>2017-10-02T04:39:55.512Z</atom:updated>
            <content:encoded><![CDATA[<p>Software engineering has been astonishingly effective at applying mass numbers of developers to directed problem solving. Can we use successful software technologies and/or techniques to involve more citizens more equitably in the development of laws and policies?</p><p>In the software engineering classic <a href="https://en.wikipedia.org/wiki/The_Mythical_Man-Month">Mythical Man-Month</a>, Fred Brooks articulated Brook’s Law: “adding manpower to a late software project makes it later”. Over the intervening 42 years, software developers and managers have settled into techniques that help to bend that law. It is not possible to completely parallelize software development, but modern methods such as <a href="http://agilemanifesto.org/">Agile</a>, <a href="https://opensource.com/resources/what-open-source">open source</a>, and <a href="https://en.wikipedia.org/wiki/Test-driven_development">test-driven development</a> enable large numbers of programmers to work together without falling into confusion. For example, the open source operating system, <a href="https://www.youtube.com/watch?v=GGBlBIFAKmA">Linux, has 21M lines of code contributed by 4000 programmers</a>. About 10,000 lines of code are changed every day, but the software has been stable through dozens of timed releases (every 2.5 months) for over a decade.</p><p>Viewed as code, the scale of Linux is larger but roughly comparable to that of the US tax code plus its regulations and case law. That body of law runs to <a href="https://taxfoundation.org/how-many-words-are-tax-code/">about 80,000 pages</a>, or 4M sentences as compared to 21M lines of code in Linux. The <a href="http://www.washingtonexaminer.com/look-at-how-many-pages-are-in-the-federal-tax-code/article/2563032">rate of change</a> is approximately 1500 new pages/year. I’d guess we should double that for number of changes (IE there can be a change without a new page) for a rate of 411 sentences/day in the tax code, which is about 20x less than Linux.</p><p>Viewed as a social construct, the workforce and processes of Linux and The US Tax Code are quite different. The <a href="https://en.wikipedia.org/wiki/Internal_Revenue_Service">IRS has ~80k employees</a> and there are ~<a href="http://www.answers.com/Q/How_many_tax_attorneys_are_there_in_the_US?#slide=2">200k tax attorneys</a> in the US. There are also Congressional staff, lobbyists, and concerned citizens involved. So, the number of people actively engaged in the tax code, policy, and case law is at least an order of magnitude higher than 4000 Linux developers. The update process is also remarkably different. Linux uses <a href="https://en.wikipedia.org/wiki/Regression_testing">rigorous regression testing</a> to maintain “correct operation”, but the concept of correct operation barely even exists with respect to the law. There is a nascent regression test in that the <a href="https://www.cbo.gov/topics/taxes">Congressional Budget Office</a> will estimate the revenue by income breakdown that will result from major alterations in tax laws. This is a far cry from sophisticated and precise testing that would determine the existence of ambiguities, double binds, absurdities (“Uncle Sam owes me $10M!”), and tests of fairness like the <a href="https://en.wikipedia.org/wiki/Buffett_Rule">Buffett Rule</a>.</p><p>Let’s wave a magic wand and create a create a regression test suite for law and policy. How might this new capability change things? Maybe a test suite is not obtainable (unlike a line of code, a section of law is not constrained to be computable), but let’s start by pretending we’ve solved that problem. Specifically, we have automated testing available, so that anybody can propose a change and run all the tests to find out if anything “breaks”.</p><p>Even with a simple change we see the need to distinguish between values and facts (<a href="https://medium.com/@ericloeb/the-pro-life-pro-choice-debate-is-a-great-issue-8dd763dc683a">https://medium.com/@ericloeb/the-pro-life-pro-choice-debate-is-a-great-issue-8dd763dc683a</a>). For my simple change, I adjust my copy of USC Title 26 Section 1a (<a href="https://www.law.cornell.edu/uscode/text/26/1">https://www.law.cornell.edu/uscode/text/26/1</a>) to say that the taxes paid by married individuals filing joint returns on taxable income of more than $250k will be $75,650 plus 42.1% of the excess over $250,000. I then upload my copy of 26 U.S.C § 1 into the legal repository and request a test. I receive a test report indicating my change produced a slight alteration in the projected revenue and I did not break any fundamental constraints.</p><p>Automated tests can easily simulate tax revenue, but whereas computer code can be tested for bugs, the meaning of “bug” is not clear for laws. Some would argue that increasing the top tax rate by 2.5% does indeed violate a fundamental constraint. Automated tests can answer factual questions about the law, such as revenue raised under a transparent set of assumptions about economic conditions, and they can provide factual answers to specific values questions about the law, such as acceptability of the law given a transparent set of assumptions about acceptability, but all those assumptions need to be quantified, and the assumptions surrounding values, such as the Buffett Rule will be highly controversial.</p><p>If we want to try something more complicated, it also becomes clear that we need a machine-readable programming language for law. For example, I want to create a flat tax that <a href="https://medium.com/@ericloeb/civic-engagement-and-tax-reform-988300404505">covers the government expenditures that 80% of us </a>agree to each year. To do that I’ll need to alter a great many paragraphs and sections of the law. Crucially, after making those alterations, I’ll need to be able to run automated tests. This brings us back to the magic wand that was used to create the test suite, because running automated tests on arbitrarily re-written law means the law will really have to be written in a programming language. A machine must be able to interpret the new laws, because an automated test of new laws must first transform the new laws into a standardized data structure on which the tests are defined.</p><p>There are <a href="https://link.springer.com/book/10.1007%2F978-3-540-85569-9">efforts to create computable legal structures</a>, and some argue that computable l<a href="http://logic.stanford.edu/complaw/complaw.html">aws are the next step in legal evolution</a>. Computable law looks to me <a href="https://www.quora.com/Are-systems-of-aesthetics-ethics-values-and-moral-law-Turing-computable">like artificial intelligence</a> from two sides: (a) if we can create AI then we can use AI to interpret law, so making laws machine-readable is a lot easier, and (b) if AI is possible (IE if a finite state automaton can capture human intelligence), then a finite state automaton can also probably encode all the laws we want to enact. I don’t know. I suspect that legal compliance will prove to be computable but Justice will not. Computers can objectively determine the fact that you were going 46 mph in a 45 mph zone, but the values question of the fairness of receiving a ticket for that will have to be determined by a social process.</p><p>So, my answer is yes. We can apply open source (et al) to involve more citizens more equitably in the development of laws and policies. This will require machine-readable laws (which enable legal test suites) and the social processes that enable us to construct agreed-upon, formalized legal tests. With these technologies, open source legal structures will be possible. Anybody can create new legal modules and upload them for anybody to use. In that world, we can opt-in to the package of laws that we wish to comply with, which will in turn determine where we can live. Somebody who picks the “murder anyone any time” legal module, can only live in a prison. Somebody convicted of murder would have their preferences for that law set by the state (“you have demonstrably opted into the murder-at-will version of the law”), and the places they can live will be set by that choice. Where there are disagreements on values, we can opt into different sets of laws that represent our beliefs and localities can set the law packages that are allowed within that location.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a0fecc76f469" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Designing an app to alter social dynamics]]></title>
            <link>https://medium.com/@ericloeb/designing-an-app-to-alter-social-dynamics-e66d13fb0e25?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/e66d13fb0e25</guid>
            <category><![CDATA[apps]]></category>
            <category><![CDATA[resistance]]></category>
            <category><![CDATA[social-cognition]]></category>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[feedback]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Sun, 17 Sep 2017 20:12:37 GMT</pubDate>
            <atom:updated>2022-10-05T17:34:03.668Z</atom:updated>
            <cc:license>http://creativecommons.org/publicdomain/zero/1.0/</cc:license>
            <content:encoded><![CDATA[<p>I recently recommended that the DoD acquire a free, open source coordinating application that will help increase the size and impact of non-violent mass protests in democratic countries that have come under the sway of authoritarian regimes. This can be done by <a href="https://medium.com/@ericloeb/an-adaptive-and-parameterized-control-law-for-campaign-finance-reform-f1e5883e93d3">applying feedback control</a> logic that will alter the dynamics that lead citizens to riot or lose heart. Here’s how…</p><p>Feedback control does not need to be complicated or hard to understand. <a href="https://thinkprogress.org/german-town-pranked-neo-nazis/">An elegant example</a> has been discussed a great deal lately, wherein an app can collect donations to an anti-Nazi cause for every mile the Nazis march. This approach <a href="https://www.nytimes.com/2017/09/06/opinion/comedy-protest-taxes-nazis.html">has had the intended effect</a> of discouraging the Nazis from marching in the German town that coined the method.</p><p>Non-violent protest movements may have more trouble applying feedback control to maintain their momentum, because of the complicated dynamics of apathy, fatigue, discouragement, and impotent fury. As more protesters grow discouraged by the limited impact of protests, the State is better able to crack down on the protests, further discouraging the protesters. As more and more people turn to violence, popular support builds for the State violence in response, and the violence further discourages peaceful protests.</p><p>So, my goal is to design an app that will adjust the dynamics of the movement, such that activist anger and fatigue help build the movement instead of causing it to dissipate.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/352/1*7dVDCiQLf9OvP-K7eCv5Yg.png" /><figcaption>Figure 1: A hypothetical picture of a resistance movement. Most Americans will disapprove of an anti-democratic regime, but will not be active. The vast majority of those who actively protest will be non-violent.</figcaption></figure><p>The dynamics that we need to alter will show over time in the population distribution of the resistance movement. Figure 1 is a sketch of that distribution. Most of the people opposed to the regime do not turn out for rallies, and most of the protesters who do turn out are non-violent. The violent protesters are a minority of a minority. Over time, as the regime commits outrage after outrage, the natural response will play out as changes in how many people just try to ignore the whole thing (“Inactive”), how many join in non-violent resistance protests, and how many turn to violence.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/359/1*8pzOwJZH8XgpuA4Fcgf7qQ.png" /><figcaption>Figure 2: Hypothetical population responses to an unending succession of outrages. Yesterday the regime did something horrible; today we see it in the news; what happens next? The solid orange line shows a natural response. People become inured to The Orange. Fewer people show up to the next non-violent protest. Out of frustration, there may be a slight increase the number of people moved to violence. The dotted line shows the response we need instead. We need to convert the natural fatigue response into an accelerating non-violent protest response in which more cause for fatigue means bigger protests.</figcaption></figure><p>Figure 2 shows what is likely to happen and what we would prefer to happen because of all the calls to action. The initial state, also shown alone in figure 1, is a mostly inactive resistance movement and protesters who are mostly non-violent. As the abuses pile up, fatigue sets in. There is a natural drop in the number of people who turn out for non-violent protests. “<a href="https://www.washingtonpost.com/opinions/trump-abuses-his-power-in-the-cause-of-bigotry/2017/08/28/1f473e3c-8c22-11e7-84c0-02cc069f2c37_story.html">It is a short road from being serially outraged to being slightly bored to being completely inured</a>”. The response that we would prefer is shown in the black dotted line. Ideally, we will use feedback control to create an accelerating dynamic, in which more outrages bring about more and larger protests.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/351/1*l8Y-iWaHWNIx8EFLHYHxSg.png" /><figcaption>Figure 3: A hypothetical natural and feedback-altered population response to an enraging provocation. Yesterday the vigilantes did something violent and horrible; today we see it in the news; what happens next? The solid red line shows a natural response. People are moved to do something, and there is an increase in both the non-violent and the violent response. The dotted line shows the response we need, which is as little increased violence as possible, with energy instead re-directed into a greater non-violent response.</figcaption></figure><p>Figure 3 shows what is likely to happen and what we would prefer to happen because of anger. The initial state, also shown alone in figure 1, is a mostly inactive resistance movement and protesters who are mostly non-violent. It is sometimes <a href="https://canipunchnazis.com/">OK to punch Nazis</a>, but it is <a href="https://talkingpointsmemo.com/edblog/should-we-be-punching-nazis">never OK to resolve civic disputes through violence</a>. What’s a civilization to do? As provocations pile up, people will naturally respond in kind. The response we would prefer is shown in the black dotted line. Ideally, we will use feedback control to channel the energy and anger of the people moved to violence toward support for bigger, non-violent protests.</p><p>We can use feedback control to change dynamics so long as the system modes are <a href="http://www.me.umn.edu/courses/me8281/notes/Chapter%203%20Controllability%20Observability.pdf">observable and controllable</a>. We will not be able to <strong>observe </strong>the precise number of non-violent/inactive/violent people who are on the side of Democracy. However, we can count the number (or rough order of magnitude) of people who show up at rallies. We will not be able to <strong>control </strong>the number of people who show up for non-violent protests. However, we can provide <a href="http://backspace.com/notes/2016/02/protest.php">organization, transportation, and communications</a> that make it easier for the activists to show up.</p><p>The task is to construct a feedback control law such that the more we lose, the bigger we grow. I have approached this as an exercise in <a href="https://uxmastery.com/how-to-create-a-customer-journey-map/">journey mapping</a>. To design the app, one could facilitate a group of experts in diverse fields — sociologists, organizers, communications, app design, etc — in a focused discussion of the psychology of apathy, discouragement and anger. This exercise will yield dozens of touch-points where we can hope to intervene. With those touch-points in hand, one can rapidly develop, field-test, and refine an app before distributing it nationally.</p><p>Table 1 shows an example of what the initial list of touchpoints might look like. The table contains three natural reactions that we can observe if they occur <em>en masse</em>. Each reaction has an associated feedback control (“Adjustment”) that can help mitigate and channel the natural reaction.</p><p>Table 1: Observations and feedback adjustments to alter the dynamics of protest fatigue</p><iframe src="" width="0" height="0" frameborder="0" scrolling="no"><a href="https://medium.com/media/8b91c09d6274ded21bc69e213b196a6d/href">https://medium.com/media/8b91c09d6274ded21bc69e213b196a6d/href</a></iframe><p>Item one in Table 1 indicates that some drop-off in activity will occur when activists become unhappy at events. The number of potential problems are legion. The app will need to supply funding for organizers to arrange for signage, porta-potties, speakers, sound equipment, permits, etc. The app should track attendance (anonymously) and make use of event ratings from participants and other in-the-moment indicators of event success.</p><p>Item one also shows that the app needs an automated budget adjustment in response to metrics on activist discontent. The app needs to disburse funds automatically in accordance with a budgeting algorithm. The budgeting algorithm will include learned parameters, such as the proportion of revenue dedicated to event support. These learned parameters can be tailored to circumstances, such as event location (IE we actually have a vector of learned budget parameters for amenities and support).</p><p>The donor arrangement should also be “automated” in the sense that it would rest on (anonymous) recurring donations within a range — IE $1/month normally, up to $5/month maximum. The desired dynamic is an increasing non-violent response as time goes on, and the more the non-violent movement fails to bring about change the bigger it should get. This means that the worse things get, the more expensive the response will be. That kind of arrangement is not a forte for traditional fundraising. Allocation of a steady stream that’s pre-arranged to allow for some growth will better fit the need.</p><p>Item two indicates that the app should be built to die with a bang, not a whimper. As more people become inured to authoritarian rule, the app will automatically increase the portion of the budget directed to advertising. Signs of apathy will include decreased web traffic to certain pages. The participating organizations should provide web traffic data to the organizing app to enable it to sense activist apathy. Chatbots and social media analytics can also provide good proxy measures of declining interest. A feedback control law that converts apathy to engagement will necessarily seem increasingly noisy to some users, because the app will learn who it needs to yell at (and how best to yell at them) to get them off their couches and onto the streets.</p><p>Finally, item three indicates that the app needs funding for a well-regulated militia. The well-regulated militia will shoot violent protesters with their cameras. If participants are rewarded for publicizing evidence of violence by people on “our” side, then it will help show the seriousness of intent of the non-violent protest movement. The militia should also perform other dangerous tasks. They should be responsible for citizen’s arrests of violent protesters, standing in harm’s way if things do turn violent, managing the secure technical infrastructure needed to operate in a hostile environment (for example, on-site systems to enable anonymous broadcast of evidence of a big protest that the authorities do not want people to see), and help recruit those eager for confrontation into these more civilized avenues for use of force.</p><p>It is possible to design an app to apply feedback control with the intent of altering some aspect of social dynamics. In particular, with a steady stream of funding, an organizing app can use feedback control laws to alter the natural dynamics of a protest movement. By the analysis above, such an app will be a coordinating tool that instantiates positive monetary feedback loops around metrics of failure, because the desired dynamic is for the protests to grow larger so long as they fail to achieve their objectives.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e66d13fb0e25" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Civic Engagement and Tax Reform]]></title>
            <link>https://medium.com/@ericloeb/civic-engagement-and-tax-reform-988300404505?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/988300404505</guid>
            <category><![CDATA[collective-intelligence]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[tax-reform]]></category>
            <category><![CDATA[taxes]]></category>
            <category><![CDATA[customer-feedback]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Sun, 06 Aug 2017 21:08:09 GMT</pubDate>
            <atom:updated>2017-10-01T18:22:24.772Z</atom:updated>
            <content:encoded><![CDATA[<p>Civic Engagement and Tax Reform</p><p>As Congress turns its jaundiced eye toward <a href="http://thehill.com/blogs/pundits-blog/economy-budget/344599-tax-reform-must-be-congressional-republicans-next-move">tax reform</a>, we can hope for simplification while expecting the rewards to go instead to those who can afford to hire accounting departments. But now, for something completely different, I propose fundamental changes to our tax code through persistent data collection and feedback control. Once we implement persistent data collection, our laws can work in an environment in which they have continuous feedback signals from the populace. When we’re ready, we can achieve simplification for individuals and corporations through behind-the-scenes (but transparent for those who want to look) feedback-controlled tax algorithms.</p><p>To move toward feedback-controlled tax algorithms, I recommend that the near-term tax reform efforts include</p><p>1. A browsable, expandable menu of the US budget by which any US citizen can vote yes or no on a line item or aggregate of line items. This would be a non-binding vote.</p><p>2. The IRS should compute expected taxes for each tax paying entity and supply those results to the tax payer.</p><p>3. The IRS and HHS should coordinate on a national credit card by which negative taxes and government benefits for low income individuals can be consolidated and inexpensively disbursed.</p><p>The following notional methods use web and/or mobile user interfaces. Any practical system would also require staffing and support for phone, fax, and snail mail voting.</p><p><strong>Differential Taxes</strong></p><p>Suppose that our tax laws can make use of a “persistent vote”, which would be a permanent Monte Carlo survey. Every citizen can provide a response at any time and can change their response at any time. Today I vote “yes”; yesterday I voted “no”; maybe I’ll change my vote again later today. This possibility is one of the most interesting things about digital governance: persistent votes would not have been practicable before the modern era.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1003/1*sLwKehxrPoWdG6oZDvmzLw.png" /></figure><p><strong>Give us a menu</strong>: Suppose in the persistent vote every citizen can set a check box for each item in a the US budget. With this interface, each voter can give a yes/no vote for each budget item. Checking the box means that you think the US budget should include that item. Uncheck the box if you think the US should not pay for that item. Perhaps some of us would vote to eliminate one of our aircraft carriers, or eliminate PBS, or make no changes at all, but we would each get an up-or-down vote on each item. The image to the left gives an example of a top-level summary (from <a href="https://www.nationalpriorities.org/budget-basics/federal-budget-101/spending/">here</a>). A whole-budget checklist would start like the image at the left, and would be expandable down to the <a href="http://federal-budget.insidegov.com/l/120/2017-Estimate">finest levels of detail</a>, so the users can vote at whatever level of detail they want.</p><p>An expandable menu interface for voting on the budget would support a control scheme in which, for example,</p><p>A. We pay a flat tax to cover the items that 80% of us agree on; and</p><p>B. We pay a progressive tax to cover the items that 40% — 80% of us agree on; and</p><p>C. We pay a voluntary tax on the remaining items. These would be progressive taxes that are only levied on the tax payers who support the program. The taxes would grow very large for expensive programs with little public support, thus providing a strong source of pressure for the budget reductions that republican voters say they want but even republican governments cannot deliver.</p><p>The default budget vote would be “yes” on all items, so it would take some exertion to drive up the cost of a program. The computed tax rates would also have to be enacted by Congress; they would not be automatic.</p><p>How would this play out? If 80% of all citizens either did not vote or said “yes” to everything, we would end up with a <a href="http://www.governing.com/columns/public-money/col-Federal-Flat-Tax-Plan-Actually-Work.html">flat tax of about 20%</a> to cover the <a href="https://www.cbo.gov/topics/taxes">current taxes of about $3.4T</a>. If fewer than 80% of all citizens said “yes” to half of the budget (by dollar value), then all existing progressive taxes are reduced by half plus all tax payers would pay a 10% flat tax. Like other flat taxes, this is regressive for low incomes. The shape of the progressive tax rates can be steepened to adjust for the impact of the flat tax at the low end. All else being equal, halving the progressive tax and adding a 10% flat tax would have no net effect for a tax payer with an effective tax rate (including deductions, etc) of 25% on $100,000 with $20,000 in deductions. Above that, the net effect is to flatten the rise of the progressive tax rate.</p><p>The “voluntary tax” loading for programs with low support should help re-balance the budget. Suppose that there were a concerted effort to kill something expensive, such as the $1T wedge for Medicare &amp; Health, and 62% of us are somehow persuaded to vote against Medicare. In this case the computed tax rate for the remaining 38% who vote “yes” for Medicare would be about 72% higher than current tax rates (assuming no benefits from a partial flat tax or from voting “no” on other things). During the period before the Congressional vote that sets the actual tax rates, there would be efforts to convince people to support Medicare. There would likely be a re-balancing — for example, part of the country would support the Military but not Medicare and vice versa. Alternatively, the added expense of Medicare might inspire more people to vote against Medicare, which would drive up rates for the remainder, and so on until everybody votes no. In that scenario, many would be voting “no” because they feel forced to by the expense, and Congress might <a href="http://www.chicagotribune.com/news/opinion/commentary/ct-republicans-cut-welfare-social-security-20161212-story.html">override the voters</a> or open discussions on how to restructure the programs in that part of the budget.</p><p><strong>Give us a dial</strong>: Suppose we each had a little thermostat with which we could set the tax rate we prefer. Maybe everybody would just set the dial to zero, but maybe not. After all, Warren Buffett supports <a href="https://en.wikipedia.org/wiki/Buffett_Rule">the Buffett rule</a>. The expectation from the <a href="https://en.wikipedia.org/wiki/Wisdom_of_the_crowd">Wisdom of the Crowd</a> is that the average of all our dials will be “the right” number. There is no “right answer”, of course, but it would be interesting to see how the average dial setting varies: would higher income tax payers give themselves higher income tax rates? How would the tax rates come out for other classes of tax payers, such as corporations or investors?</p><p><strong>Give us a menu of dials</strong>: Suppose every voter had a line item budget menu described above but with dials instead of checkboxes. The dial for each item would be set initially to the dollar amount that the voter’s taxes would contribute to that line item: <a href="http://comptroller.defense.gov/Portals/45/documents/defbudget/fy2016/fy2016_Weapons.pdf">$30.14 for the Joint Strike Fighter</a>; <a href="https://www.babble.com/entertainment/trumps-federal-spending-cuts-could-end-pbs/">$1.37 for PBS</a>; etc. One could set the dial to $0, which would be like voting “no” in the menu of checkboxes, or one could set the dollar amount to something other than its default setting. Feedback control laws can be built into these dials, so that, for example, the individual’s default dollar amount for PBS would represent a flat tax if there was 80%+ agreement that PBS was worth more than $0.</p><p>Again, the dollar averages across all tax payers should display the Wisdom of the Crowd effect, with each line item funded at the “right” level. This would presumably be helpful information for Congress. If the individual tax amounts given by our dial settings were enacted in law, we should quickly settle into a state in which taxes come into line with ideology without having to reference ideological distinctions in the law. In other words, people who value PBS will pay more for PBS, and people who value the JSF for pay more for the JSF.</p><p>A menu of dials interface could also be a helpful device to further public policy discussions in the <a href="https://medium.com/@ericloeb/build-a-national-daydreaming-system-d46f414fc0a">national town meeting</a>. To implement the menu of dials, it will have to be possible to compute each person’s next tax bill. However quickly we can do that now, it will be <a href="https://en.wikipedia.org/wiki/Moore%27s_law">1000 times faster in 15–20 years</a>. As politicians propose changes to the tax code and/or the budget and/or the structure of programs, a hypothetical set of next-year tax bills can be computed, so that voters can see the impacts of the proposal and give feedback. Such a device would expand the range of the politically feasible. With fast feedback, citizens could demonstrate the popularity of a tax loophole worth millions for their region (or occupation, age bracket, income bracket, etc) without realizing in advance that that’s what they wanted. This contrasts with the current system in which they would need to coordinate explicitly to hire lobbyists or organizers. Conversely, citizens may be able to coordinate implicitly (by turning up their dials) to pay more taxes for certain services.</p><p><strong>Give us functions over time</strong>: <a href="http://www.slate.com/articles/business/moneybox/2012/04/grover_norquist_and_h_r_block_the_unholy_alliance_of_tax_prep_firms_and_conservative_activists_to_make_your_taxes_even_more_complicated_.html">Once the IRS can compute our tax returns</a>, it should be possible to provide every tax payer with a graph showing how much that tax payer is expected to pay in taxes each year over the next decade. It would be helpful to add to this a graph of the expected government benefits over the same year. For example, as a tax payer approaches retirement, their individual graph would probably show the lines crossing: post-retirement benefits exceed post-retirement taxes.</p><p>A voting interface can be built on graphs of taxes and benefits over time. The user would “bend the curves” to indicate how the curves should change over time. One person might indicate higher taxes now but then a drop to zero after retirement; another might indicate decreasing taxes and decreasing benefits. The graph of tax bills depends on information the IRS already collects on us, but the graph of benefits depends on a lot else besides. Accordingly, a citizen might interact with variations of this timeline interface: the timelines as a citizen; timelines as a business owner; timelines as a member of some profession (IE personal taxes displayed against government investments that impact a given profession); timelines for employees of a company, hobbyists, neighborhoods, <a href="https://medium.com/@ericloeb/vignettes-from-an-ai-enhanced-district-town-meeting-b61d2303e2ca">districts</a>, etc.</p><p><strong>Give us a menu of functions over time</strong>: Finally, it is much harder for the government to respond to our tax dial settings for next year than it is to respond to dial settings that manage the change in behavior over the next 10 years. The best interface will enable us to drill down into the budget and interact with a graph that shows our individual tax bill for that item over the next 10 years. The interface can have a single dial — so that setting the value to $0, for example, turns yields a graph that drops 50% each year and then drops to $0 and stays there three years from now. The graph can have other feedback logic built in. For example, if 80% of us agree on an item but it will take 5 years to shift over to the flat tax, then everybody’s individual graphs for that budget items (and all budget categories that subsume that budget item) will show a shift at 5 years when the new flat tax kicks in.</p><p><strong>“Tax” is not just a four-letter word</strong></p><p><strong>Assistive taxation</strong>: The word, “tax” is synonymous with words like task, work, and exercise. Taxation, like exercise, can make us stronger and healthier. If presented with options, the largest tax payers might be willing to pay more than their required taxes if the excess were arranged as an investment with potential returns. For example, what if the IRS devoted some employees to helping billionaires find government programs that served those billionaires’ needs — infrastructure, education, community development, research, policing, major procurements, etc. “Pay the full taxes on your billions that your accountants have determined will be $0.25 , but then pay an additional $250,000 into this research project in exchange for 1% ownership”. Yes, the billionaires will be paying more taxes than if they just played an accounting game to keep the tax bill low, but by working with the IRS they can direct those extra dollars to programs that they want. Corporations could benefit from similar government staff support. The government is now led by an expert deal-maker, so this should go very well.</p><p>Notionally, the progressive tax code could extend to <a href="https://en.wikipedia.org/wiki/Negative_income_tax">negative taxes</a>, wherein the government would provide cash payments. Negative taxes are another positive sense of “taxation”. Negative taxes would equate in the extreme to a <a href="https://en.wikipedia.org/wiki/Guaranteed_minimum_income">guaranteed basic income</a>, a notion that is <a href="https://www.cnbc.com/2017/02/21/technology-ceos-back-basic-income-as-ai-job-losses-threaten-industry-backlash.html">coming into vogue</a> as a counter to the expected ravages of AI.</p><p>In an assistive taxation arrangement for lower incomes, the IRS would help coordinate the services that those citizens need. This would look something like a basic income floor and something like a generalized Civilian Conservation Corps arrangement. For example, given every one of the 1.5M people <a href="http://www.pewresearch.org/fact-tank/2014/09/08/who-makes-minimum-wage/">living on minimum wage</a> a credit card to help them gain their targeted subsidies. When they take public transit to work, they pay with their credit card and there is no cost. They get cash benefits (negative taxes) and big discounts on child care, health care, etc. Instead of a massive, isolated infrastructure for government programs and a massive, isolated infrastructure for computing taxes, the government could conceivable combine (or cross-talk) IRS and HHS to facilitate negative tax rates in the form of a smoother path to existing services. The required coordination would not have been practical in an earlier era, and may not yet be possible without upgrades to the government’s computational infrastructure and integration of our government processes with that infrastructure.</p><p><strong>Taxation and control theory</strong>: <a href="https://en.wikipedia.org/wiki/Control_theory">Feedback control</a> can be used to alter any system’s dynamics. Feedback control laws on tax rates can be used to alter the dynamics of markets and <a href="https://medium.com/@ericloeb/an-adaptive-and-parameterized-control-law-for-campaign-finance-reform-f1e5883e93d3">other processes so that they are more socially beneficial</a>.</p><p>To my (perhaps naïve) view, the tax expenditures in our tax code seem to be attempts at feedback control. For example, the <a href="http://solaroutreach.org/wp-content/uploads/2015/03/ResidentialITC_Factsheet_Final.pdf">investment tax credit for solar</a> is a successful mechanism that <a href="https://www.wsj.com/articles/will-solar-energy-plummet-if-the-investment-tax-credit-fades-away-1447643512">has helped increase the demand</a> for solar panels. If implemented as real feedback control, instead of a fixed tax credit, the level of the tax credit would have been computed from a simulation of the evolution of the market for solar panels (which might have been done), and the size of the credit would be a function of how strongly the market needs added stimulus this year and/or which segments need stimulation (as clearly was not done). Now that there is a strong demand for solar panels, a tax credit implemented as feedback control would automatically reduce to zero. There would be no need to sunset or repeal the credit.</p><p>The Investment Tax Credit for solar was created in the <a href="https://www.gpo.gov/fdsys/pkg/PLAW-109publ58/pdf/PLAW-109publ58.pdf">Energy Policy Act of 2005</a>. Section 1335 adds a new paragraph, 25D, to a section of title 26 (the tax code). Paragraph 25D has three individual tax credits, including: “30 percent of the qualified photovoltaic property expenditures made by the taxpayer during such year”. That’s the bit of law that creates a -30% tax for solar panels. With feedback control, instead of “30 percent”, the law might say, “an amount given by the target demand minus current demand, as defined in paragraph 25E”. This carries the complexity of a new paragraph, 25E, that explains how we set target demand levels. But that’s not so bad, and if written as a function of target demand, this law would have the benefit of turning into an actual tax once the market has reached and surpassed that targeted level of maturity. In other words if the target demand is 10 units and the current demand is 0, the tax is (10 minus 0) times -3 = -30%. Later, when current demand is 12, the tax is -2 times -3 = 6%, and we are using the same law to generate revenue as we previously used to stimulate the market.</p><p>Tax reform has the potential to simplify our tax code. Much of the complexity occurs at tax time when we go to look up rules. Taxes could be simpler and yet depend on even more complex rules if we were not required to use those rules in computations ourselves. Once we implement persistent data collection, our laws can be simplified and operate in an environment in which they have continuous feedback signals from the populace. To move toward feedback-controlled tax algorithms, I recommend that the near-term tax reform efforts include</p><p>1. A browsable, expandable menu of the US budget by which any US citizen can vote yes or no on a line item or aggregate of line items. This would be a non-binding vote.</p><p>2. The IRS should compute expected taxes for each tax paying entity and supply those results to the tax payer.</p><p>3. The IRS and HHS should coordinate on a national credit card by which negative taxes and government benefits for low income individuals can be consolidated and inexpensively disbursed.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=988300404505" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An Adaptive and Parameterized Control Law for Campaign Finance Reform]]></title>
            <link>https://medium.com/@ericloeb/an-adaptive-and-parameterized-control-law-for-campaign-finance-reform-f1e5883e93d3?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/f1e5883e93d3</guid>
            <category><![CDATA[feedback-loop]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[collective-intelligence]]></category>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[campaign-finance]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Sun, 30 Jul 2017 16:54:18 GMT</pubDate>
            <atom:updated>2017-08-13T17:23:11.420Z</atom:updated>
            <content:encoded><![CDATA[<p>Engineered control systems use feedback control and collections of parameterized and/or adaptive algorithms, called <a href="https://www.britannica.com/topic/control-theory-mathematics#ref828088">control laws</a>. Control laws provide correct system response across the entire range of system environments. Our legislators, who are not, regrettably, computer programmers, should make more use of methods from system dynamics and computer science to write laws that account for, and are more robust in the face of, changing circumstances.</p><p>We are all familiar with control laws. Thermostats, the cruise control in a car and autopilot in a plane are familiar examples. The control law for an air conditioning thermostat is quite simple: if the air temperature climbs above the AC setting, blow chilled air into the room until the temperature drops to the setting. After that, turn off the power in the cooler, but keep blowing the reservoir of available chilled air while it is colder than the thermostat setting.</p><p>Even the simple control law for a thermostat shows the computer science (and military) principle of “divide and conquer”. Instead of creating a single temperature control system, which is mechanically difficult, engineers have divided the temperature control problem into distinct systems for heating and cooling. The same thermostat can be used for both heating and cooling, but there are different physical systems being controlled when we switch the thermostat from cooling mode to heating mode. That’s a parameterized control law.</p><p>In more complex systems, control laws typically divide and conquer the control problem through <a href="http://portal.research.lu.se/ws/files/4673551/8571461.pdf">piecewise linear control</a>. In each part of the control space, there is a simple linear control law with more variables. These control laws are parameterized by features of the system being controlled. The control law of the thermostat has one variable, temperature, and two parameters, the temperature setting and the mode (heat or cool). An airplane autopilot system has many variables, having to do with the yaw, pitch, roll, and airspeed and other <a href="https://courses.cit.cornell.edu/mae5070/Caughey_2011_04.pdf">measures of the plane in flight</a>. For a given slice of all those variables (a given speed, yaw, etc.), there is a simple linear control law that translates small deviations from desired settings to modified actuators: just as a thermostat control law turns on the AC when the room gets too hot, an autopilot system will adjust the flap angles on the wings and tail if the plane banks too steeply in a turn. The difference for the autopilot control law is that those adjustments change the variables (the airspeed, etc), and a new linear control law comes into play. The autopilot control law has as many parameters as the plane has variables, plus the parameters the pilot sets, such as airspeed, altitude, and heading.</p><p>Biology is full of amazing <a href="https://www.albert.io/blog/positive-negative-feedback-loops-biology/">examples</a> of feedback control. There are numerous biochemical systems in cells where some chemical level is stabilized through <a href="https://bio.libretexts.org/TextMaps/Map%3A_General_Biology_(OpenStax)/2%3A_The_Cell/07%3A_Cellular_Respiration/7.7%3A_Regulation_of_Cellular_Respiration">feedback control</a>. Substance A stimulates production of substance B which causes decreased production of substance A. When the B levels are steady, the cell is fine; when B levels drop, it makes substance A, which helps the cell re-stabilize. Moving up the genome, my favorite example is programable spinal reflexes. If you are sitting in the middle of a room and your chair unexpectedly tilts back (perhaps due to a prank or an earthquake) you will reflexively splay your arms and legs to maintain balance. However, if you are touching something solid that can prevent your fall, such as a sturdy table, you will reflexively grab the solid object instead of splaying your limbs. These distinct reactions are too fast to be anything but spinal reflexes. The (parameterized) reflexes are thought to be programmed into the spinal circuitry based on (the parameters of) your circumstances by <a href="http://neuroscience.uth.tmc.edu/s3/chapter05.html">the cerebellum</a>.</p><p>Our future politics can use smarter approaches to control via adaptive structures that incorporate citizen input. Our government was designed as a feedback control system (via our votes for state and federal representatives), and we do have some systems thinking built into our institutions. We should do more of that.</p><p>One example of adaptation in law is the price-indexed Cost of Living Adjustment (COLA) to social security and various <a href="http://militarybenefits.info/cola-cost-of-living-adjustments/">military benefits</a>. The COLA is computed from a consumer price index, which is an average of prices of various consumer goods. As the price of consumer staples (bread, milk, etc) increases, so does the <a href="https://fas.org/sgp/crs/misc/94-803.pdf">monthly social security payment</a>. Conversely, the threshold income for the Alternative Minimum Tax was not indexed, and now a tax intended for small numbers of extremely wealthy families is applied annually, for better and for worse, to a <a href="http://www.taxpolicycenter.org/briefing-book/what-amt">larger than originally planned</a> number of US households.</p><p>Let’s look at how we can use adaptive and parameterized feedback control to simplify the campaign finance reform problem.</p><p>Comprehensive campaign finance reform is hard to agree on. The <a href="https://en.wikipedia.org/wiki/Citizens_United_v._FEC">Citizens United ruling</a> in 2010 let more money flow into campaigns than individuals can compete with, and gerrymandering makes Congressional (House) districts predominantly non-competitive. The <em>status quo</em> gives a pronounced advantage to defenders of the <em>status quo</em>.</p><p>To apply an adaptive algorithm to campaign finance reform, let’s first divide and conquer: break the problem down into smaller pieces. Let’s start by tackling blow-out districts in which there is either no out-party candidate (IE the incumbent is the only major party candidate) or the victory margin in the previous general election was greater than say, 40%. Whereas it is exceedingly difficult to pass campaign finance reform for all districts, it may be relatively easy to put laws in place to cover the blow-out districts, particularly if we do not too-greatly threaten the incumbent.</p><p>Like the COLA, we need some measured values that will control the amount of money provided instead of just writing that dollar amount into law. Instead of some fixed amount, like $100,000, the out-party nominee will receive funds based on some measured values. The social Good to be stabilized here is the ability of the voters to hear competing messages. So, the measures should reflect the ability of the out-party candidate to get his or her message to the public.</p><p>We’ll base the adaptive and parameterized campaign finance law for the blow-out districts on the cost of running a given number of ads in the largest media market that reaches into the district. There were <a href="https://docs.google.com/spreadsheets/d/1oArjXSYeg40u4qQRR93qveN2N1UELQ6v04_mamrKg9g/edit#gid=0">64 districts without competitors in 2016 and another 95 with victory</a> margins of 40% or more. Just as the thermostat has distinct modes for heating and cooling, we’ll try to pass an adaptive campaign finance reform that only applies to these 159 blow-out districts (about 1/3). In these districts, an out-party candidate will receive enough public money to make a showing, where “enough money” depends on the district. By defining the amount by political impact instead of dollars, we will tend to give more dollars to republicans than democrats, because the democrats tend to come from urban areas with expensive media markets.</p><p>Suppose the law provides funds to run 25 60-second ads reaching 200,000 viewers. In 2018, the law would provide $135,800 to a republican opponent for Rep Maloney in D-NY-12, who had a 66% margin in 2016 in the <a href="http://fitsmallbusiness.com/tv-advertising/">New York media market</a>. The same law would provide $57,500 to a democratic opponent for Rep Walden R-OR-02 who had a 43% margin in 2016 in the Yakima media market. This seed money is unlikely to tip the balance, but it should at least help voters in a few more blow-out districts hear an opposing point of view.</p><p>Next, let’s tackle the public financing of the slightly competitive races, such as the elections with 20% — 40% victory margin in the previous general election. There is often an out-party competitor in those districts, but the competitor lacks the stature of his or her incumbent opponent. The problem in these districts is not the existence of competition, but rather the existence of viable competition. There are enough out-party voters for the out-party candidate to be known, but their political work will inevitably lack the visibility of a member of congress. The campaign finance law in these districts, which would be much more contentious to pass, could perhaps center on providing out-party candidates with loyal opposition visibility. For example, a tax on campaign expenditures in these districts could be used to fund some public project to be chosen by the voters and managed by the losing candidate in the 1–2 years following the election.</p><p>Finally, in the districts with margins less than 20%, the incumbent representatives probably feel that the margins are too close for comfort. They will resist any assistance to their competitors. Efforts to control the spending in these districts also run up against first amendment protections. And, by the way, what exactly is the social Good in reducing the dollar amount of campaigns in which both sides can raise enough money to compete? In these districts, I believe, the social Ill to be combated is the overall degradation of civility by campaign ads. Perhaps we can build a legislatively viable feedback control law around objective measures of civility. Assuming we can construct appropriate measures, perhaps we can set a negative tax rate around them. For example, if voters are better able to correctly answer factual questions about an issue after viewing your ad, you get x% per factual question back on the ad buy. If a random panel of independent reviewers consistently agree that your ad shows your opponent and his/her ideas respect, you get some multiple of y% back on the ad buy. The voters of the district would have some say in the relative balance of x and y, so they can decide what’s more important.</p><p>By dividing difficult problems into distinct sub-domains, we may be able to find sub-domains where passage of a law is more tractable than is whole-domain, comprehensive reform. Within any domain, we should take a moment to consider what social Good we are attempting to promote, define measures of that Good, and then construct the law via feedback control on those measures. Within these (and other) laws, we should use computed values rather than fixed values to define the impact. Parameterized and adaptive feedback control laws are the way of the world, except for our actual laws. We should fix that.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f1e5883e93d3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[New forms of streamed political debate]]></title>
            <link>https://medium.com/@ericloeb/new-forms-of-streamed-political-debate-94becad3c1c0?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/94becad3c1c0</guid>
            <category><![CDATA[collective-intelligence]]></category>
            <category><![CDATA[politics]]></category>
            <category><![CDATA[presidential-debates]]></category>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[reform]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Sun, 23 Jul 2017 15:56:01 GMT</pubDate>
            <atom:updated>2017-07-23T17:44:09.051Z</atom:updated>
            <content:encoded><![CDATA[<p>The formats of our political debates are stale. There are many different ways that candidates for office could engage with each other, the issues, and the audience. The basic format of the Lincoln/Douglas debates, if not their attention to detail, are still our model for how debates should be conducted. We have barely begun to incorporate TV into the possibilities, much less the internet. Candidates don’t even use charts with agreed-upon facts. Reality shows like The Apprentice, Shark Tank, and Top Chef do a better job engaging the audience with substantive information and issues than our formalized, sound-bite debates. It is long past time for new forms of political debate.</p><p>I will here focus on televised and/or streamed debate formats. This is the modern standard, in which the candidates endure some live contest of will and passion for our entertainment and nominal edification. In the status quo approaches, these debates appear to follow formal academic debate rules, except that they are not formally judged or scored. I suspect/assume that the early American politicians had training in the academic format, and fell back on that by default. It also made sense not to score the debates, because the voters were to be the judges.</p><p>Once <a href="http://billmoyers.com/story/theres-no-debate/">for-profit media began hosting the debates</a>, the media and political professionals indirectly conspired through successive elections to arrogate scoring of debates for themselves. Modern debates are <em>de facto</em> scored by the media leaders and the partisans of each side as they all work in their respective roles to shape the narrative. In the post-debate spin each side makes the case that their candidate won, and the media plays the highlights (effectively, the highest scoring moments) for some time after the debate. The scoring criteria, for better and for worse, point out the most exciting moments, not the most informative.</p><p>Because our modern debates are effectively scored anyway, we should go back to the roots and use new (to us) debate formats that are explicitly scored as well as new formats that are not.</p><p>Any debate will face a fundamental knowledge/engagement tradeoff between depth of detail and audience engagement for a given duration. An internet debate can do more to educate and engage voters because it can use a longer period of time than a live-streamed debate. Conversely, a live-streamed debate that really delves deeply into any issue(s) must either baffle most of the audience or else spend much more time on each issue than is currently the norm for televised debates, so as to provide voters with requisite background information. While it could be tremendously helpful to begin the discussion of an issue with a review of the background and facts that the candidates agree on, would it sell? The sponsoring media organization and the candidates all want a very large audience, which means that the debate, if televised, must be entertaining.</p><p>This use of for-profit news corporations as arbiters of our political debates is a <a href="https://medium.com/@DeepGreenPhilly/2016-presidential-debates-why-climate-change-was-not-on-the-agenda-5d3f821cacbe">questionable choice</a>. However, this deal with the Devil has had the benefit of settling our debates into an equilibrium point at which large numbers of voters tune in to Presidential debates. The campaigns and media all want large audiences, so the dumbing down of our debates was not necessarily a deliberate craven act, it may also have been a natural market optimization. As politicians and media work to engage large audiences, we can expect a continued decline in substantive content and further appeals to basic mammalian rituals of dominance and submission. Eventually, we will inevitably elect a brute primate with no concern for the truth and a finely hone instinct for sadistic display.</p><p>Finally, it should be possible to play the interests of the media against the candidates in the realm of public participation and interactivity. The candidates will do everything they can to control the situation, and so work toward negotiated debate formats that diminish spontaneity to the extent possible. In contrast, the media sponsors should love the exciting moments of public humiliation that true spontaneity will provide. The town hall format has apparently been heavily controlled by the candidates, given the very low surprise and spontaneity of the citizen involvement. Accordingly, I suspect much of the innovation in participatory formats must be pioneered at the Congressional level, where the relative balance of power between the candidates and media sponsors is more heavily weighted toward the media.</p><p><strong>Explicitly Scored Formats</strong></p><p><strong>Agreement format</strong>: It might be particular helpful in this age of hyper-partisanship to engage in a format that rewards candidates for finding areas of agreement. The debate would be similar to current formats, but the rules should include more back and forth. The candidates will each state their positions on the issue being discussed. Perhaps three issues could be discussed in an hour-long debate. Candidates would get one point for each item of agreement. A bi/non-partisan panel of judges would rule out any trivial agreements (“Well, Bob, I agree that you said your name is Bob, but everything you said after that was fake news.”) Candidates would earn points for offering compromise, and lose points for attacks. In a competitive district, this format may help to encourage an honest discussion of the issues.</p><p>The knowledge/engagement tradeoff for the agreement format is masked by the judges. Many voters will not know the implications of a concession. In a typical hour-long debate, the candidates will have strong incentive to appear reasonable by playing up their minor concessions. The judges will presumably know the issues well enough to give lower scores for minor concessions. Perhaps the after-debate news/entertainment can include discussion among the judges rank-ordering the originality and substance of any concessions or compromises in the debate.</p><p>One can imagine many entertaining variants of the agreement format involving interactions with the judges and/or the audience. For example, the goal of the exercise could be to find non-trivial agreement <em>and then convince partisans to come along</em>. Perhaps there could be two partisan panels of judges who the candidates must persuade to agree to join some compromise. A high participation variant could substitute votes from the party-identified audience. Too-easy agreement will show up as high agreement from the whole partisan spectrum.</p><p><strong>Public service contest</strong>: In 2005 I founded a political action committee, <a href="https://herbriede.tumblr.com/post/57825730336/goodworks-pacorg-announces-a-new-model-for-all">Goodworks-PAC</a>, to support democratic candidates who would campaign via public service events. There had been a few candidates who did nothing but public service work as their campaign “rallies”. The candidate and his/her volunteers worked together intensely, and this style of campaign was seen to forge deep bonds between the candidate and a relatively small number of volunteers. The technique is difficult to put into practice in a state-wide or national campaign, but that was not my purpose.</p><p>For Goodworks-PAC I envisioned an alternative form of primary campaign debate between many democratic candidates in a deep-red (I.E hopelessly republican) congressional district in a Presidential swing state. Each of these “debates” would be scored by the number of volunteers the candidate could turn out. Obviously, if we could get many democrats competing to turn out democrats for public service events in the red parts of swing states, then that would help us win those swing states in the state-wide elections. I was able to recruit candidates and organize events, but I was amazed to discover that every candidate thought they could win the whole election. Fostering cooperation among the democratic candidates, even in deep red districts, was surprisingly difficult. Accordingly, for this to work, the public service contests between candidates need to be truly competitive. Score the candidates on the outcomes they (and their parties) produce: pounds of trash picked up, number of people served, liters of blood donated, etc.</p><p><strong>Culminating event format</strong>: A televised discussion between the candidates could be the final event in a longer-duration online discussion. As a scored format, the culminating event would advantageously provide a mechanism for bias in the outcome. In particular, in a highly partisan/gerrymandered district, the inevitable winner has no incentive to elevate the stature of his/her opponent by participating in a debate. The out-party candidate does have incentive to work with local media and online to educate voters. However, those outlets do not want to anger the incumbent who is about to be re-elected. A scored format can be constructed to assure a televised victory for the incumbent, which will provide political cover for the media to work with the out-party candidate in the online portion of the debate.</p><p><strong>Implicitly Scored Formats</strong></p><p><strong>Agreement format</strong>: We can work candidate agreement into a standard, non-scored format. For example, a debate could have ground rules around two different clocks. When an issue is introduced, the candidates have the first bolus of time to articulate basic facts and history of the issue that they agree on. Voters will gain basic background information and they will see how well the candidates are able to work together. If the candidates can agree on basic facts, then the second clock starts and they can articulate their distinct policy views on the issue. If no agreement is found, the debate instead moves on to the next issue.</p><p><strong>Legislation-based format</strong>: A format centered on drafting legislation would be instructive for a Congressional debate. The candidates and their advisers would all work together to draft legislation that they agree on. Notionally this would be the district’s legislation that will be proposed in the next Congress no matter who wins. This would also be a proverbial “lock them in a room until they work it out” kind of format. As the candidates and advisers work together over some number of sessions, the process would be caught on camera and edited for TV under agreed-upon rules. The text of the proposed legislation would be available for questions and comments through-out the rest of the campaign, which will provide further opportunities for the high information voters to learn more about the candidates’ thinking.</p><p>Another legislation-based format could engage the candidates and audience in a detailed examination of the impact on the district of a small section of law or a regulation. This would be something like the non-scored agreement format in which the candidates start by finding facts that they agree on. In this case, the legislation or regulation is the bedrock fact. But is it on balance beneficial or harmful to the district? The format could make use of audience voting, expert testimony, policy simulations and other interactions to make the proceedings interesting and effective. The format could be a single live event or a culminating event that is edited for TV under a set of agreed-upon rules.</p><p><strong>Apprentice format</strong>: Now, in the era of Trump, it should be possible to institute debates modeled after the President’s hit show, The Apprentice. There are many <a href="https://www.boozallen.com/e/media/press-release/cyber-security-breach-simulation-demonstrates-need-for-todays-le.html">management</a> and government simulations, where a management team spends a few hours playing their roles in a hypothetical real-world event that would take place over days or months. The <a href="https://en.wikipedia.org/wiki/Military_simulation">military conducts simulations</a> of this sort for combat exercises, and so could the competing candidates and their advisers. Each candidate team would work through the same exercise simultaneously. The process would be caught on camera and edited for TV under a set of agreed-upon rules. The voters will have a chance to compare the leadership style of the competing candidates and their advisers when placed in the same, difficult situation.</p><p>Another variant of the Apprentice format would involve constituents and real-world outcomes. Rather than an exercise, each candidate team will be responsible for an equivalent politico-economic problem to help solve. For example, each candidate will work with one small town in the district, selected from a pool of matched small towns. Over the course of the campaign, the candidates will work with the citizens of the town (or owner of the business, county program, school district, etc) to solve the same problem that both towns face. The process would be caught on camera and edited for TV under a set of agreed-upon rules. The voters will have a chance to compare the leadership style of the competing candidates and their advisers in the context of normal, daily events.</p><p><strong>Televised online format</strong>: In Presidential and some state-wide races, there is enough of an audience to create profitable TV from ongoing, online policy debate. This format would basically be a modernized Lincoln/Douglas. Each candidate will write or present a considered opinion on an issue. The candidates will then respond and critique in writing and/or via video. The back and forth can continue as long as the candidates feel there is more to be said. This primary material will be enhanced by a Natural Language Processing interface that will help interested voters and journalists find passages of note. The structure of the argument can also be graphed automatically for logical evaluation. The resulting pile of structured and unstructured data should be a rich source of material for TV segments. In addition, the media corporation that develops better automation for constructing informative TV segments from the data <em>might </em>win more viewers.</p><p><strong>Stylized combat formats: </strong>If we drop the constraint that candidates should be dignified — and at this point, why not — then there are many more possibilities still. For example, why merely content ourselves with the Al Smith dinner for candidate comedy. Why not have a face-to-face comedy smack-down? We needn’t stop there; we could even have actual combat. Presidential candidates are competing to be commander in chief, so why not show them leading in battle? While they could conceivably preside over existing combat exercises, per the Apprentice format, that would be difficult to do without revealing secrets. Instead, the candidates should lead their forces in a giant paint-gun battle that includes advanced paint-gun weaponry. Stylized combat would make for fun TV that lots of voters would watch.</p><p>There are many different ways we could require our candidates to compete. It is absurd that we are limiting ourselves to such a crabbed view of what the competition between candidates could entail. What would it take to start innovating?.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=94becad3c1c0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Vignettes from an AI-enhanced district town meeting]]></title>
            <link>https://medium.com/@ericloeb/vignettes-from-an-ai-enhanced-district-town-meeting-b61d2303e2ca?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/b61d2303e2ca</guid>
            <category><![CDATA[civic-engagement]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[national-town-meeting]]></category>
            <category><![CDATA[healthcare]]></category>
            <category><![CDATA[politics]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Sun, 16 Jul 2017 13:29:28 GMT</pubDate>
            <atom:updated>2017-07-22T23:59:12.656Z</atom:updated>
            <content:encoded><![CDATA[<p>I am a fan of the national town meeting, a term <a href="http://www.imdb.com/title/tt1331004/">introduced by Ross Perot in his 1992 campaign for President</a>. In my view, the national town meeting must comprise many meetings at lower levels of government, including, town meetings for congressional districts. While the national town meeting will be an excellent forum for our understanding of ourselves as a nation and our preferences for the future, consensus for the practical steps and changes to the law may more commonly occur locally in congressional districts. These district meetings should be easily accessible, ubiquitously available (24x7), and operate under special rules for those relatively rare times when the representative is able to attend.</p><p>A national town meeting will be the heart of our nation’s ability to conduct itself intelligently. By providing a formal mechanism for continuous public engagement, our government will literally increase its intelligence, in the sense of being better able to understand our environment, form plans and achieve long-term goals. The United States is already in the process of becoming an intelligent being: development of collective intelligence is occurring naturally and inevitably. Constructing our national town meeting will be another step down that gradient.</p><p>In our coming world of intelligent assemblies, a congressional district (typically ~700,000 people) should itself act intelligently with respect to its self-interests. The people of the district will need to work together and in coordination with their representative’s office to sense the district’s physical, economic, and intellectual environment; plan its actions; and track its progress toward goals. Much interpretation and disagreement will be involved, and even the most gerrymandered of districts contains enough diversity of opinions to support robust debate.</p><p>Let’s see how the district town meeting works from the perspective of a small business owner, a citizen with concerns, and a lobbyist trying to pass an upcoming bill…</p><p>A small business owner asks her computer for economic trends that impact her business. The result is a map and synopsis supported by the district town meeting. The map includes detailed economic projections for the next year in the area surrounding her business, views for expected future impacts of existing state &amp; federal laws, and some relevant bills that are under consideration. One potential bill is indicated to have significant impact. The bill is tied to a projected economic slow-down in her area over the next ten years. She asks for more information. An AI tied to the bill shows tax savings for her and her business and offers to check the impact on her customers anonymously. She checks the impact on her customer list (She does this via <a href="https://medium.com/@ericloeb/personal-government-agents-for-all-ba4f56faf817">her personal government agent</a>. The government does not retain her customer data.). She sees that she and her high-end customers will enjoy a net tax savings but most of her customers will suffer a net income loss. From the computed impact on her income, her customers, and her area, the computer provides a best guess of a net loss for her business with an indicated 60% certainty. The business owner registers a “no” vote on the upcoming bill but does not otherwise participate in the debate.</p><p>A concerned citizen has aging parents who require expensive medication and one of his children has a pre-existing condition. He comes to the district meeting, which is a Congressional component of <a href="https://medium.com/@ericloeb/a-new-desk-for-our-new-president-34fa349ab446">the President’s desk</a>. At the top level there’s an interface for upcoming votes, and within that module there’s an interface for the upcoming health care bill. The icon for the health care bill is glowing brightly because there’s a lot of discussion occurring, and he is tagged in much of that discussion. He sees in a glance that the representative has been personally involved in the discussion, but is not here right now. He clicks on the health care discussion icon and can see that in the eight hours since he last participated, no conclusions have been reached, there are many open questions and there are two nationally-rated postings. These summaries are curated by a mix of AI, staff, and volunteers. He looks at responses under the nationally rated posting of interest, which began with video of him confronting the representative several days ago about his daughter’s pre-existing condition. The AI organizes the responses, and he sees that there is a cluster of responses in which people have said that maybe his child should die; maybe it’s God’s will. He composes himself and tapes a video response to those people. The AI helps him edit the response for presentation and then post it. His response to the people who think God wants him to lose his health care coverage goes viral nationally over the next 36 hours.</p><p>A lobbyist is working to pass a major health care bill. Automated polling and response inferences strongly suggest that the viral video of the calm but clearly upset father may be moving support away from the bill. The lobbyist calls a friend in the representative’s office and asks to speak at the representative’s district meeting in a head-to-head, give-and-take format. She calculates that the district meeting will choose the upset father as her debate opponent, who she expects she can handily defeat in a back-and-forth debate. When the lobbyist’s presentation is announced for the following day with an open slot for the opposition, the AI schedules a vote over the next two hours to select an opposing speaker for the open debate slot. Opted-in participants throughout the district are alerted that a vote is open. The meeting vote selects the upset father. However, the father knows his limitations and asks that the district’s 3rd choice, the head cardiologist of the local hospital, speak instead. When the debate does occur, many people tune in to hear the lobbyist present her case for passage of the bill and the response from the cardiologist. Many more listen to the highlights, which are carried by news outlets as well, and a large portion of the district’s voter’s register their non-binding votes on the bill for the representative’s edification.</p><p>Somewhat later, when the Senate is preparing to vote on its version of the health care bill, a Senator reviews the straw vote outcomes of his statewide town meeting and all the state’s representative’s town meeting. The AI indicates that the most persuasive argument was the grief of parents over the anticipated loss of their children. Computed persuasion indexes show that parental grief was compelling for people in nearly all psychographic and ideological categories. The Senator requests a quiet experiment with negative ads showing lively children who die needlessly and wrecked parents saying, “You did this, Senator”. Over the next several days the computer finds that those ads will have high impact and it is unable to identify effective counter arguments. The AI provides an estimate of <a href="https://www.cdc.gov/media/releases/2014/p0501-preventable-deaths.html">2571 preventable deaths</a> of children under 10 in his state prior to the next election if the health care law passes. The Senator is certain his opponents will not hesitate to blame him for those deaths if he votes for the health care bill.</p><p>It was <a href="http://quoteinvestigator.com/2012/11/11/exhaust-alternatives/">not Churchill</a>, but <a href="https://en.wikipedia.org/wiki/Abba_Eban">Abba Eban</a> who said, “<a href="http://www.peoriamagazines.com/ibi/2010/feb/americans-will-always-do-right-thing-after-they-have-exhausted-all-alternatives">Men and Nations do behave widely once they have exhausted all other alternatives</a>.” The national town meeting will not alter the nature of politics, but it will allow us to simulate many more alternatives as we feel our way forward. The district meetings of Congressional representatives are a critical system component.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b61d2303e2ca" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cognitive maps of political issues]]></title>
            <link>https://medium.com/@ericloeb/cognitive-maps-of-political-issues-2799fc7bea93?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/2799fc7bea93</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[democratic-party]]></category>
            <category><![CDATA[political-theory]]></category>
            <category><![CDATA[political-science]]></category>
            <category><![CDATA[cognitive-science]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Wed, 12 Jul 2017 02:15:25 GMT</pubDate>
            <atom:updated>2017-07-12T02:53:49.196Z</atom:updated>
            <content:encoded><![CDATA[<p>Life will be better in that fine not-so-far-off day when Science has devised a representation for political issues that is able to capture their complex structure and inter-relationships. Even when discussed on the news and in political debates, issues are usually treated as isolated concepts, indexed and identified via keywords. This approach to issues misses a lot. Advances in natural language processing will allow us to create <a href="https://en.wikipedia.org/wiki/Topic_model">better representations of issues</a>, and these will help us improve the supporting technologies for politics.</p><p>At present, the most widely deployed technologies treat political issues as categorical variables. Each issue (“climate change”, “tax reform”, “jobs”, etc) is an isolated entity. As human beings who understand words, we know that these words are connected: jobs are related to climate change, for example. Still, it is common to treat each issue separately. It’s easier. The typical approach is to associate keywords and phrases with issues and then use those associations to determine which category a block of text is best matched to. Tools that support this type of analysis, such as <a href="https://provalisresearch.com/products/content-analysis-software/">WordStat</a>, <a href="http://www.qsrinternational.com/nvivo-product">NVivo</a>, and <a href="https://www.tensorflow.org/tutorials/word2vec">Word2Vec</a> are critical for <a href="https://en.wikipedia.org/wiki/Digital_humanities">status quo analysis of “unstructured” (text) data</a>.</p><p>If you want to see the limits of isolated word analysis, try writing to your Senators. The most popular software on the Hill for processing incoming letters appears to be <a href="https://intranetquorum.com/">Quorum</a>. Quorum will match you letter to a bank of keywords and associated responses. A staffer will review the automatically assembled response before being sending it to you. If your letter does not easily match the current issue, and mismatches like yours are occurring frequently enough, then the software may trigger creation of a new issue. In this case, staff will use built-in workflows to write an automated response for the “issue”, which in this context means a new collection of keywords.</p><p>The deep structure is of issues is important to political discourse. One group believes that climate change and jobs are related because there will be more green jobs in the future. Another group believes that climate change and jobs are related because fighting climate change will require the loss of jobs in mining and oil. Those underlying relationships between political issues and concepts are where the work of political discourse resides; not in the simple, siloed way we structure issues at present.</p><p>Political issues, like other human concepts, are supported and/or implemented psychologically by associative networks. The neuronal mechanisms behind concepts remain unknown. However, the inter-<a href="http://marciazeng.slis.kent.edu/Z3919/44association.htm">relationships between concepts</a> can be <a href="http://onlinelibrary.wiley.com/doi/10.1002/jeab.55/abstract">probed experimentally</a>, as Freud famously did through <a href="https://en.wikipedia.org/wiki/Free_association_(psychology)">free association</a>. There are many such experimental methods. For example, reaction time differences provide repeatable metrics on associations between concepts with a <a href="https://en.wikipedia.org/wiki/Priming_(psychology)">well-studied effect called priming</a>. When you see “horse” your time to recognize “knight” is reduced. Having seen “night” you are quicker to recognize “moon”. Priming spreads through our conceptual network. <a href="http://dnd.wizards.com/">“Horse” activates “knight” which activates “armor”</a> — so that you will be quicker to recognize that word — but “horse” does not activate “moon”.</p><p>The application of cognitive science to political science has <a href="http://www.people-press.org/2011/05/04/beyond-red-vs-blue-the-political-typology/">produced maps</a> but no central theory or systematic mapping of the concepts and associations that define political issues. Political scientists seem mostly to be concerned with the meaning of political concepts within political theories (“what does equality mean in a democracy”). The cognitive scientists appear mostly concerned with understanding how <a href="https://www.amazon.com/Thinking-Fast-Slow-Daniel-Kahneman/dp/0374533555">our insights into decision-making</a> apply to <a href="http://www.bbc.com/future/story/20150506-the-dark-psychology-of-voting">voting behavior</a>. Working political scientists, like Frank Luntz and George Lakoff, are able to identify how subtle <a href="http://time.com/4682090/frank-luntz-america-relationship/">changes in wording can influence our thinking</a>, but I’ve not seen these techniques applied to produce a compendium of <a href="https://lwp-viz.shinyapps.io/usfactors">distinct political psychologies</a>. It would be helpful to understand the degree and extent to which liberals and conservatives are <a href="https://www.theatlantic.com/politics/archive/2016/07/why-democrats-and-republicans-literally-speak-different-languages/492539/">fundamentally speaking different languages</a> while superficially talking about the same thing.</p><p>Assuming it is possible to aggregate cognitive maps across individuals, it will be interesting to compare the results of different methods. For example, concepts can be mapped by priming studies of individuals; alternatively, we could develop a map by applying Word2Vec to large corpuses of topically diverse but ideologically consistent political text (ie DailyKos or Breitbart); alternatively, we could conceivably construct a map from the overlap of the donor and/or membership lists of issue advocacy groups (in essence, this would allow some vendor to supply us with <a href="https://en.wikipedia.org/wiki/Collaborative_filtering">collaborative filtering</a> on political purchases).</p><p>Improved science around the cognitive maps that we use to understand politics should lead to improved tools for conducting politics. Media, parties, candidates, governments, professional associations, interest groups, corporations, and individuals will all benefit to some extent from improvement in <a href="http://journals.sagepub.com/doi/abs/10.1177/00027640121958483">tracking the national dialogue</a>. Improvements in message refinement could help us reach durable compromises, just as we have seen these technologies <a href="https://www.theatlantic.com/technology/archive/2017/04/facebook-data-does-not-contradict-intelligence-report-about-russia-meddling/524703/">used to deepen existing schisms</a>. Perhaps with better automation for <a href="https://medium.com/@ericloeb/bots-for-democrats-9e15323a6307">sensing and comprehending the political environment</a>, we may come to understand the world in <a href="http://blogs.discovermagazine.com/intersection/2011/09/07/your-brain-on-politics-the-cognitive-neuroscience-of-liberals-and-conservatives/#.WO5srNLyuUk">more sophisticated terms than liberal vs conservative</a>. Maybe the Democratic Party could even recognize how it is falling out of step with the voters without first having to lose its place in all branches of government!</p><p>Mapping cognitive networks can contribute to advances in many domains besides politics. Job referral sites and HR departments would work better with a better model of the complex interrelationship between skills. In these cases, the application would be to map the associations of experts to provide guidance to the non-experts who need to match resumes to jobs. Collaborative filtering interfaces (for movies, products, prospective first dates) may also be improved. In these cases, the filters are domain-limited cognitive maps that can perhaps be inferred rapidly from broader cognitive maps.</p><p>An understanding of the cognitive maps of political concepts for individuals or ideologies will help us construct better tools for (eg) processing official correspondence, tracking issue debates, and providing consumers with news and opinion content they are likely to find interesting. There are many possibilities for advances in <a href="http://onlinelibrary.wiley.com/doi/10.1111/jcc4.12124/full">political science, political reporting, and political organizing</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2799fc7bea93" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Neuromorphic Bureaucracies: Space, Time, & Cost]]></title>
            <link>https://medium.com/@ericloeb/neuromorphic-bureaucracies-space-time-cost-d47465952d50?source=rss-45f9774115------2</link>
            <guid isPermaLink="false">https://medium.com/p/d47465952d50</guid>
            <category><![CDATA[neuromorphic-bureaucracy]]></category>
            <category><![CDATA[collective-intelligence]]></category>
            <category><![CDATA[cybernetics]]></category>
            <category><![CDATA[engineering-tradeoffs]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Eric Loeb]]></dc:creator>
            <pubDate>Sun, 02 Jul 2017 18:08:16 GMT</pubDate>
            <atom:updated>2017-07-02T18:24:53.190Z</atom:updated>
            <content:encoded><![CDATA[<p>There are several examples of neural circuits that manipulate the fundamental tradeoffs between space, time, and cost. These limitations cannot be avoided, but they can be mitigated. Our engineers should listen to their brains.</p><p>Space/time/cost (S/T/C) tradeoffs are exemplified at a human scale in a trope among contractors: “<a href="http://www.sixside.com/fast_good_cheap.asp">you can have it fast, you can have it good, or you can have it cheap. Pick two.</a>”. The limitations imposed by S/T/C tradeoffs are easy to find. “Space” can mean a variety of things having to do with complexity: size, detail, accuracy, aperture, etc.; “Time” has the usual meaning; and “cost” can mean money, resources, metabolic costs, etc.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/211/1*eyRqhL2egGb2tcySogv7GA.png" /><figcaption>Filter properties of auditory sensory cells</figcaption></figure><p>My favorite example of the brain’s manipulation of S/T/C tradeoffs is in the auditory system. My <a href="https://www.ncbi.nlm.nih.gov/pubmed/3103861">first neuroscience work </a>was in somatosensory cortex, but my first love was for <a href="http://ieeexplore.ieee.org/abstract/document/1169791/">signal processing in the inner ear</a>. There is a clear summary of auditory processing <a href="http://www.tulane.edu/~howard/BrLg/t6-AuditoryTransduction.html">here</a>, which includes the embedded <a href="http://www.cns.nyu.edu/~david/courses/perception/lecturenotes/pitch/pitch.html">image</a> to the left. The X axis is the log of sound frequency. The 1 on the X axis represents 1 Khz (This is <a href="https://www.youtube.com/watch?v=3FBijeNg_Gs">what 1 Khz sounds like</a>). The Y axis is signal attenuation. The zero at the bottom of the Y axis represents prefect transmission and no attenuation. Each line in the graph shows measured responses of auditory sensory cells to sounds of various frequencies. We see that these cells respond somewhat to lower frequency, hit a peak response (the valleys in the depicted lines), and then their responses roll off sharply to higher frequencies. The 60–80 decibels of signal attenuation are like making a noisy city street inaudible. Most <a href="http://www.hearforever.org/blogs/01/24/2011/qa-why-arent-there-any-nrr-40-earplugs">ear plugs are rated in the mid-30s decibel range</a> for sound attenuation. The graph shows that the auditory cells can respond across the spectrum, but they typically only respond near their tuned frequency.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/222/1*75RQF56ZtuUA8Az9vp-Ylw.png" /><figcaption>Frequency characteristics of the FFT</figcaption></figure><p>By way of contrast, engineers convert sound waves into measures of specific frequencies, as shown in the image to the left <a href="https://en.wikipedia.org/wiki/Fast_Fourier_transform">from Wikipedia</a>. The top of the graph shows a simple sound wave. The bottom graph shows the outputs of the popular Fast Fourier Transform (FFT) of the signal at the top. The FFT gives coefficients for frequency bins, much as the auditory cells respond to sounds in a range of frequencies. Unlike the auditory cells, the engineering approach uses box-like frequency ranges. Namely, the blue line on the bottom shows that there are positive coefficients, representing signal amplitudes, in each of 5 concise frequency ranges (E.G 1 kHz to 2 kHz). The neuronal and FFT approaches are very different. Auditory sensory cells eventually respond to nearly any signal if it is loud enough; FFT coefficients will be zero no matter how loud the signal is, so long as there is no signal in a specific frequency range.</p><p>The difference in natural vs engineered approaches to “hearing” represents a clever natural workaround to S/T/C tradeoffs. The FFT faces a tradeoff: narrow bins of frequency require more data and thus take longer to recognize new signals than do wide bins of frequency. We can see the natural workaround to this tradeoff in the strange asymmetric shape of the natural frequency bins. These shapes are very wide in frequency space but not at all boxy. They enable faster recognition of new tones while also enabling precise distinctions between tones. Engineers distinguish two frequencies by making inexpensive and direct comparisons of energy in neighboring frequency ranges. The genome is more extravagant. Instead dividing up the spectrum into a few non-overlapping frequency bins, the natural (but counter-intuitive) approach is to divide the spectrum into a huge number of huge and overlapping frequency bins. Each of these huge frequency bins responds quickly to new signals, as huge frequency bins do. But because of the asymmetric response curves, the brain can also detect precise differences in frequency.</p><p>Humans and animals have evolved to detect changes quickly. We respond rapidly but we also have exquisite accuracy. Accuracy <strong>must </strong>take longer. How do we use just one pair of ears to do both? We rely on characteristics of the populations of neurons that respond to sound. For any frequency F, there will be a lot of neurons firing. Nudge F just a little higher, and there will still be a lot of neurons firing, but some new neurons will have started and others will have stopped. That small difference in the population that’s responding is what enables our brains to tell the difference between F and F plus a nudge.</p><p>Do bureaucracies face space/time/cost tradeoffs? Yes. All systems do. “It can be good; it can be fast; or it can be cheap. Pick two”. The “good” in that saying is space: desirable features for a project, complexity, robustness, scale, and so on. In project management, it is <a href="https://www.pmi.org/learning/library/time-cost-tradeoff-problem-study-10217">most common to see a time-cost trade-off</a>. A “project” is assumed to have fixed scope, and, as everybody knows, increasing the scope of a project or program is likely to increase the cost or time to completion.</p><p>How can organizations make use of the clever space/time/cost workaround exemplified by the inner ear?</p><p>One neuromorphic workaround can be applied to situations in which there are triggering events. For example, fraud alerts, cyber intrusion and other kinds of risks that simultaneously need fast and accurate onset detection. To implement the neuronal approach, we need our best broadband signal with which to build a fast response. For <a href="https://en.wikipedia.org/wiki/Intrusion_detection_system">cyber intrusion</a>, we would build anomalous traffic detectors that operate over many things (many ports, or many files, many data types, users, sub-systems, etc) at once. These wideband anomaly detectors will have more data with which to develop models of normal activity. They will have limited individual ability to identify the source of unusual traffic, but better resolution: with larger data volumes, we can label smaller fluctuations as significant. A bank of these detectors with shifted preferences would implement the natural filtering approach, wherein many detectors will respond to an intrusion and the population density of the detector responses will indicate which ports/files/users/etc are likely sources. Neuromorphic intrusion detection is a <a href="http://www.lewis-rhodes.com/assets/img/Cyber-Security-and-Neuromorphic-Computing-White-Paper.pdf">topic of commercial interest</a>, but the hype is too thick to know what is really being done.</p><p>In the cyber intrusion example, the neuromorphic system used many software agents as model neurons. Is there a useful neuromorphic approach to human management, wherein we organize groups of people like neurons? We cannot expect people to provide the sort of statistical regularity that will let us get more precision with more data. So, how can we use the neuronal tricks for evading S/T/C tradeoffs in bureaucracies?</p><p>Example 1: Suppose we wish to do a better job of counter-intelligence against the insider threat. There are <a href="http://www.defenseone.com/business/2015/04/number-security-clearance-holders-drops-12-percent/111268/">millions of cleared personnel</a>, and we want to know in advance who is going to do something damaging with the information they have access to. One neuromorphic approach would be to apply broad, overlapping, standardized data collection. However, to collect standardized data from thousands of people, we would need some kind of automated data collection, like a standard survey. At that point, we’re into a machine technique (Use a bank of analyzers; each takes weighted mean of nearly same set of people with weights giving preference on one side of demographic/topical/etc space). This is not a different kind of human management.</p><p>A human version of neuromorphic counter-intelligence could be deployed by an army of trained interviewers. The interviewers would check for warning signs in each person in a set. The interviewers would have overlapping sets, and every cleared person would be in at least two interviewers’ sets. More overlap is better, and people with access to highly sensitive data should be in several sets. If two or more interviewers flag the same person, more scrutiny is applied to that person. I estimate at least 5000 interviewers and 5000 schedulers and other support would be needed at a cost of about $2.5B/year. This would be a large investment. The <a href="http://www.washingtonpost.com/wp-srv/special/national/black-budget/">combined annual intelligence budget</a> is approximately $53B.</p><p>It’s rare to find employee types numerous enough to apply neuronal mechanisms. In those rare cases where an employer has millions of specialized employees, there is at least a superficial resemblance between the people and neurons. The people perform their functions in ignorance (by necessity) of what most of the others are doing. The people are distributed organically: despite efforts to put boxy human-engineered org charts around their work, there are too many org charts created by people who aren’t coordinating, so that the net effect is a set of oddball, overlapping shapes.</p><p>Organic structure is more resilient, more resource intensive, and more easily created than the engineered structures we think we want. As an example of organic structure, consider US cyber defense. The federal government makes <a href="https://www.bloomberg.com/news/articles/2017-01-31/trump-directing-agency-leaders-to-secure-federal-cyber-networks">every agency responsible for its own defense</a>, plus there are five agencies with <a href="http://www.ciodive.com/news/5-federal-agencies-with-a-role-in-ensuring-enterprise-cybersecurity/424557/">legal cyber protection responsibilities</a> (FBI, DHS, etc ), plus we have the cyber defense work conducted by <a href="https://www.defense.gov/News/Special-Reports/0415_Cyber-Strategy/">each of the military services and cyber command</a>. In all of these agencies, we have many people covering the same cyber defense topics and activities. Each person is doing this generally similar work, but from the vantage of their employing agency’s responsibilities. This is how we had two agencies, <a href="https://www.washingtonpost.com/graphics/2017/world/national-security/obama-putin-election-hacking/?utm_term=.0fc0d23477df">both the State Department and the FBI</a>, notice a spike in requests for temporary visas for Russian technicians before the 2016 election.</p><p>For organizations that cannot hire millions of people, augmentation through AI automation will be the normal approach to neuromorphic bureaucracy. One design approach would be to treat the automation as additional, inexpensive staff positions. This approach would facilitate use of a blend of human and automata filling those roles while the automata are trained to perform the function. Another design approach will be to assume that every human employee comes with some set of <a href="https://medium.com/@ericloeb/a-neuromorphic-ai-sensor-network-for-professionals-14806c5cbf8e">standard and in-development automata</a>. In this approach, we pass as much of the human’s work to the AI as possible, while the person oversees the automata and completes the tasks the bots cannots. This approach should lead to banks of specialized AIs feeding higher level work to the humans.</p><p>Example 2: Call centers (for customer service, sales, etc) are staff-heavy, required costs for many organizations. There is steady pressure to reduce these costs and call centers are <a href="https://www.forbes.com/sites/bernardmarr/2016/09/06/how-analytics-big-data-and-ai-are-changing-call-centers-forever/#57fc79203a32">quickly incorporating AI assistance</a>. We can see that both design approaches (treat AI as staff, treat AI as augmentation to staff abilities) are being used. As most of us have experienced, help lines will typically route us through an AI first to determine how to handle our calls. When we finally make it to a human being, that human being uses data-driven tools, <a href="http://searchcrm.techtarget.com/essentialguide/A-guide-to-call-center-analytics">which are becoming increasingly sophisticated</a>, to know who we are, help us resolve the issue, and to track the issue across calls.</p><p>A neuromorphic call center will use statistics on customer contacts (calls, emails, chats, social media mentions, telegrams, etc) to provide management with a fast and accurate view of the customers. Call center reporting will pick up on subtleties, but the call center will also respond immediately, with hyper-acute sensitivity to important features in the environment, such as a dissatisfied customer. Just as <a href="http://www.nature.com/news/people-can-sense-single-photons-1.20282">retinal receptors can respond to a single photon</a> while also screening out visual noise, can we notice a significant issue from a single legitimately unhappy customer while screening out the angry callers who are merely taking out their life frustrations on our call center employees?</p><p>The neuromorphic approach to the S/T/C tradeoff of speed vs accuracy is to use overlapping resources that do both in aggregate. The call center operators would accordingly have broad topic responsibilities that surround their specialized topic areas. For example, we might have an operator who specializes in widget X of product A; another operator specializes in widget Y of product B; and everybody knows a bit about products A through Z. Unfortunately, in the realm of customer contacts, we cannot easily provide the same signal to multiple operators. If we pass calls around from one specialist to another, we will degrade the signal (customers hang up) and anger our customers. We can provide new operators with the recordings of everything that has happened in the interaction so far, but there is still a start-up cost for each new operator getting up to speed on the call so far. Accordingly, the neuromorphic approach will be to answer each call with a team of specialists.</p><p>In the future, when you call to complain, there will be a whole team of people and bots on the line. The team leader will let you know they are all there to make sure your issue is resolved as quickly and painlessly as possible. Once we get used to these initially intimidating conference calls, they’re going to make us feel pampered. I’m looking forward to it.</p><p>Example 3: Another area with heavy staffing costs is management. Can we reduce managerial staff by replacing some management functions with AI agents? One design approach is to augment our management layer with many more staff positions that we will populate with AIs; another approach is to consider how we will augment the productivity of a smaller number of human managers with AI sensors and actuators. For example, using the former approach, we may be able to build specialized management AIs with work teams that act as business “fixers”. Where there are problems, the fixers move in to turn around the lagging or failing business unit. However, the fixer team also includes engineers who set up and feed an AI to take on the managerial work that is being done by the other people on the fixer team. Once the fixer team leaves, the business unit will have a stand-alone AI manager or a trained AI assistant for the manager(s) who remain.</p><p>While it will be possible to create some management staff positions that are populated by AIs, it is also a certainty that managers will soon be personally augmented with <a href="https://medium.com/@ericloeb/a-neuromorphic-ai-sensor-network-for-professionals-14806c5cbf8e">AI sensors</a> and actuators. Managers who do not avail themselves of AI augmentation technologies will soon be out-competed. They can gang up, and for a while the “good-old-human” network will prevail in some places. However, the companies that hire the modern, AI-augmented, higher-productivity managers will win that battle.</p><p>A neuromorphic approach to management should have massively overlapping reporting chains. This is not something people are good at. We interact a few people at organizational levels above our own, but typical reporting ratios are <a href="http://yourbusiness.azcentral.com/ideal-ratio-managers-staff-24643.html">4–20 employees per manager</a>. This situation will change with AI augmentation. The AI support for a human manager can draw on (mostly automated) reports from areas of the organization that are well outside the human’s portfolio. The manager can achieve excellence in some specific function but have broad awareness and respond quickly when the organization faces a new threat or opportunity.</p><p>Example 4: A perverse example of space/time/cost tradeoffs in our national politics is the hypothetical waste, fraud, and abuse line item in the budget. Politicians will often assert that we can cut the budget simply by getting rid of <a href="http://www.nationalreview.com/article/431446/trump-waste-fraud-and-abuse">waste, fraud, and abuse</a> (WFA). The clear examples of WFA are few. When clear examples are found, they are rapidly eliminated, but they also <a href="https://parkerfinancial.wordpress.com/2012/11/19/budget-myths/">save relatively little</a> compared to the overall federal budget. It is plausible that at least <a href="http://www.pkf.com/media/31640/PKF-The-financial-cost-of-fraud-2015.pdf">5% of our public expenditures</a> go toward WFA, but WFA are murkily entwined with valid expenditures. This is a basic fact of bureaucratic life related to space/time/cost tradeoffs.</p><p>The boundary between correct and incorrect expenditure has space/time/cost tradeoffs, of course. For a given level of enforcement (cost), we can take longer time (time) to review or else use more accumulated data (space) about the expenditure. For example, the government has attempted to reduce costs by <a href="https://www.gao.gov/modules/ereport/handler.php?1=1&amp;path=/ereport/GAO-11-318SP/data_center_savings/General_government/Promoting_competition_for_the_over_$500_billion_in_federal_contracts_can_potentially_save_billions_of_dollars_over_time">requiring competition for government contracts</a>. This cost savings comes at the <a href="https://www.linkedin.com/pulse/understanding-procurement-lead-times-why-takes-so-putnam-jd-cpcm">expense of time</a> (months and years), as the processes for submitting, evaluating, and challenging competitive bids plays out. Some of this added time has been shifted to space (staff, data) through contract vehicles that pre-approve certain expenditures by the firms that win those <a href="https://www.fedmarket.com/contractors/Government-Contract-Vehicles">contract vehicles</a>. These contract vehicles reduce the apparent time for purchase of specific items, but require many 1000s of hours of government effort to maintain as a legal category and in support of competitions. The government approval time can also be reduced by pushing labor onto supplicants. The government time and costs to review your taxes are fixed, but if you itemize deductions the system requires more space (data) that you must provide.</p><p>The S/T/C tradeoffs limit government efforts to stop waste, fraud, and abuse. No matter what method we use to make and check our expenditures, we will have tradeoffs in the speed of action versus precision for a given level of enforcement cost. We can catch anybody at all faster than we can catch a specific crook. Enforcing in one location (or intellectual realm, law, etc) is less expensive than enforcing everywhere. Building a strong legal case is slower than building a weak one.</p><p>The neuromorphic approach to waste, fraud, and abuse (WFA) may be to worry about it less. If we spend $200M to reduce the WFA of a $1B program by 1% ($10M), then we will also probably incur 5% fraud on that $200M, which would be $10M. The cost to diminish WFA by another 1% will be more than the cost to diminish WFA by the first 1%. We rapidly enter the realm of diminishing returns.</p><p>On the other hand, unchallenged crime breeds more crime. Challenging crime is a job for the immune system. Biology’s great workarounds to the S/T/C tradeoffs are in evidence in the immune system as well, where there are <a href="http://www.biology.arizona.edu/immunology/tutorials/immunology/05t.html">billions of random antibodies</a> available in small amounts. An antibody is mass-produced (<a href="https://www.ncbi.nlm.nih.gov/books/NBK26860/">and improved</a>) once it matches to an antigen. We could potentially have many millions or billions of inexact pattern matchers looking through the books. This could be done by random bots or by providing deidentified data to the public. In either case, a potential match would call in a series of bigger guns. If the matched pattern does prove to be an example of waste, fraud, or abuse, then the pattern matcher will be replicated and systematically applied to all transactions.</p><p>Space/Time/Cost tradeoff are ubiquitous, and there are many opportunities to apply the brain’s clever workarounds. Neuromorphic bureaucracies will use fast but inexact processing together with slow precision processing. By using both approaches, AI-enhanced businesses and agencies will be able to operate in their environments with speed and precision.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d47465952d50" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>