<?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 ed fernandez on Medium]]></title>
        <description><![CDATA[Stories by ed fernandez on Medium]]></description>
        <link>https://medium.com/@efernandez?source=rss-809593f1f2ef------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*oWwISAU4rLHl__pM.jpeg</url>
            <title>Stories by ed fernandez on Medium</title>
            <link>https://medium.com/@efernandez?source=rss-809593f1f2ef------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 10:12:42 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@efernandez/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[State of the Machine Learning & AI* Industry]]></title>
            <link>https://medium.com/data-science/state-of-the-machine-learning-ai-industry-9bb477f840c8?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/9bb477f840c8</guid>
            <category><![CDATA[automation]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[api]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[devops]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Mon, 30 Mar 2020 22:14:00 GMT</pubDate>
            <atom:updated>2020-11-28T10:57:11.159Z</atom:updated>
            <cc:license>https://creativecommons.org/licenses/by-nc-sa/4.0/</cc:license>
            <content:encoded><![CDATA[<h3>Adoption of Machine Learning at Scale in the Enterprise: ML Platforms &amp; AutoML <strong>[part I]</strong></h3><blockquote>How many ML Models are too many Models</blockquote><h4><em>TL;DR</em>:</h4><p>Enterprises and Companies in general, even Startups, are struggling to operationalize and <a href="https://towardsdatascience.com/what-does-it-mean-to-productionize-data-science-82e2e78f044c"><em>productionize </em></a>machine learning models.</p><blockquote>ML Models encapsulate patterns, rules, and knowledge enabling automation and optimization of business processes. They are the essential components of data driven applications and AI* services, improving and often replacing legacy rule-based business processes, increasing productivity and results.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CPuA5uU9BUIIySNUo40_ig.gif" /><figcaption>#MLSEV Machine Learning School — slidedeck <a href="https://www.slideshare.net/edslide/machine-learning-platformization-automl-adopting-ml-at-scale-in-the-enterprise">https://www.slideshare.net/edslide/machine-learning-platformization-automl-adopting-ml-at-scale-in-the-enterprise</a></figcaption></figure><p>From 2012 to 2018, blue-chip technology companies implemented custom-built ML platforms for internal use (i.e Facebook’s FBLearning, Uber’s Michelangelo, Twitter’s Cortex, AirBnB’s BigHead), many of these platforms are primarily based on open-source packages and have been deeply tailored for the specific use cases of their respective companies.</p><p>Since then, the industry has seen a strong evolution of Enterprise-grade ML platform solutions, including those from incumbent vendors (e.g. Amazon Sagemaker, Microsoft Azure ML, Google Cloud ML, etc) and the challengers in the space (e.g. DataRobot, H2O, BigML, Dataiku). Incumbent vendors follow an incremental strategy approach, with their ML services offering sitting on top of their existing cloud services as another application layer vs the ML <em>native </em>approach taken by the challengers.</p><p>As adoption of ML increases, many Enterprises are quickly turning to <em>off-the-shelf</em> Data Science &amp; Machine Learning Platforms to accelerate time to market, reduce costs of operationalization and increase success ratio (<em>number of ML models deployed and operationalized</em>).</p><p>Given only a small fraction of current ML projects, PoCs (Proof of Concepts) and models reach production and meaningful ROI, ML platforms and AutoML (<em>Automated Machine Learning</em>) are arising as tools of choice to increase rapid prototyping of ML models and validation of use cases (<em>ROI validation</em>).</p><h4><strong>Conclusions:</strong></h4><ol><li>Despite significant and growing efforts, including corporate investment in ML projects and initiatives for the past few years, <strong>only a fraction of ML models reach production</strong> and deliver tangible results. Those which do have proven to be profitable or very profitable in many cases. At scale, <a href="https://en.wikipedia.org/wiki/Pareto_principle">Pareto’s principle</a> applies with <strong>20% of ML initiatives accounting for 80% of benefits</strong>, whilst many PoC projects are discarded or frozen for later improvement. <br>*[market update: Wired, article - Nov-2020<em> ‘</em><a href="https://www.wired.com/story/companies-rushing-use-ai-few-see-payoff/"><em>Companies are rushing to use AI — but few pay off</em></a><em>’]</em></li><li>Evidence suggests the market is approaching a turning point where more and more Companies are inclined to <strong>buy (vs build) </strong>with regards to ML platforms and tools, considering <strong>off-the-shelf vendor solutions or hybrid approaches vs Open Source</strong>.</li><li><strong>AutoML, as well as APIs,</strong> are integral to ML platforms to increase rapid model prototyping, validation of ML use cases and deployment capabilities. The underlying rationale is to achieve better ROI (Return of Investment) by increasing both the number of ML candidate use cases and the success ratio (ML models deployed in production).</li><li><strong>Data (quantity and quality — ML ready), ML technical debt, MLOps vs DevOps and Enterprise ML processes and skills remain the main barriers to adoption</strong>. Despite that, traditional ML is quickly going beyond the hype cycle peak and mainstream adoption in the Enterprise is expected to be only 2–3 years away.</li><li><strong>Future evolution</strong>: beyond deep learning, in the next decade, ML platforms are expected to gradually incorporate <strong>Knowledge Representation, Reasoning, Planning and Optimization</strong> functionality paving the path towards robust AI.</li></ol><blockquote><strong>*Disclaimer</strong>: The term <strong>AI*</strong> (Artificial Intelligence) in the context of this post refers specifically to <strong>the ability to build machine learning driven applications </strong>which ultimately <strong>automate and/or optimize business processes</strong> and <strong>SHOULD NOT BE CONFUSED with robust or strong Artificial Intelligence</strong> in the formal sense, ‘something not likely to happen for a least this decade and/or next’ (emphasis from the author)</blockquote><h3>Summary</h3><p><a href="https://www.slideshare.net/edslide/machine-learning-platformization-automl-adopting-ml-at-scale-in-the-enterprise">Slides here</a> — — <a href="https://youtu.be/SZWWbi_hq24">Video 45 min here</a></p><p><strong>[1] Definitions &amp; Context [Part I: this post]</strong></p><p>Machine Learning Platforms Definitions •ML models &amp; apps as first-class assets in the Enterprise•Workflow of an ML application•ML Algorithms overview •Architecture of an ML platform•Update on the Hype cycle for ML</p><p><strong>[2] Adopting ML at Scale</strong></p><p>The Problem with Machine Learning • Technical Debt in ML systems • How many models are too many models • The need for ML platforms</p><p><strong>[3] The Market for ML Platforms</strong></p><p>ML platform Market References • early adopters • Custom Build vs Buy: ROI &amp; Technical Debt • ML Platforms • Vendor Landscape</p><p><strong>[4] Custom Built ML Platforms</strong></p><p>ML platform Market References — a closer look • Facebook — FBlearner • Uber — Michelangelo • AirBnB — BigHead • ML Platformization Going Mainstream</p><p><strong>[5] From DevOps to MLOps</strong></p><p>DevOps <strong>&lt;&gt;</strong> ModelOps • The ML platform-driven Organization • Leadership &amp; Accountability</p><p><strong>[6] Automated ML — AutoML</strong></p><p>Scaling ML — Rapid Prototyping &amp; AutoML • Vendor Comparison • AutoML: OptiML</p><p><strong>[7] Future Evolution for ML Platforms</strong></p><p>Beyond Deep Learning: Knowledge Representation, Reasoning, Planning &amp; Optimization</p><h3>1. Definitions &amp; Context</h3><p><a href="https://www.forbes.com/sites/cognitiveworld/2019/12/12/theres-no-such-thing-as-the-machine-learning-platform/#4a9b5eda8dde">There is not something as a unique ML platform</a>, product or service. The industry is very fragmented still and the terms ‘ML Platform’, ‘Data Science Platform’ or ‘AI* platform’ are many times used interchangeably.</p><p>There are, though, commonalities emerging from the different products and services available in the market, in summary:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RYXjMTj6SvkBiEr38EFuRQ.png" /></figure><blockquote>An ML platform offers advanced functionality essential for building ML solutions (primarily predictive and prescriptive models).</blockquote><blockquote>ML platforms support the incorporation of these solutions into business processes, surrounding infrastructure, products and applications.</blockquote><blockquote>It supports variously skilled data scientists (and other stakeholders i.e ML Engineers, Data Analysts &amp; Business Analysts/Experts) in multiple tasks across the data and analytics pipeline, including all of the following areas:</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aIgj1Ee0zbjdls3sD2FUcg.png" /><figcaption>ML Workflow — % allocation of resources</figcaption></figure><blockquote>• Data ingestion • Data preparation &amp; Transformation • Data exploration &amp; Visualization</blockquote><blockquote>• Feature engineering</blockquote><blockquote>• Model Selection, Evaluation &amp; Testing (and AutoML)</blockquote><blockquote>• Deployment</blockquote><blockquote>• Monitoring &amp; Explainability</blockquote><blockquote>• Maintenance &amp; • Collaboration</blockquote><p>ML platforms, as a combination of tools, systems, and services, enable experimentation and rapid prototyping but their ROI materializes in the context of Enterprise operations and business processes, well beyond experimentation.</p><blockquote>References:</blockquote><blockquote>Poul Petersen, Anatomy of an End To End ML Application <a href="https://www.slideshare.net/bigml/mlsev-anatomy-of-an-ml-application">https://www.slideshare.net/bigml/mlsev-anatomy-of-an-ml-application</a></blockquote><blockquote><a href="https://assets.ctfassets.net/nubxhjiwc091/48kgJWJ6s6p7QZWKOsKWhH/c92e2a2a1e85f6ef44d1bd19d012bf8e/Machine_Learnings_Platform_Digital_Catapult_paper.pdf">ML Platform Research Report</a> Papis.io</blockquote><h4>Enterprise Adoption Framework for ML &amp; AI*</h4><p>The ML platform &amp; tools as well as their products, ML models and data-driven applications, can be considered first-class enterprise assets, given their impact and ROI in the Enterprise.</p><p><a href="https://www.mckinsey.com/featured-insights/artificial-intelligence/global-ai-survey-ai-proves-its-worth-but-few-scale-impact">McKinsey’s Global AI Survey</a> provides a current overview of Enterprise self-reporting impact of AI* and its benefits.</p><p>However, it’s not only about technology. Processes, priorities, and people (leadership &amp; accountability) also play an integral role in onboarding and successfully deploying ML-driven applications and services.</p><p>Not only software development is different when it comes to ML apps, deploying and integrating them with existing systems is proving to be a major challenge. DevOps can’t handle MLOps (Machine Learning Operations) adequately, considering the many differences in ML Software development (more details in section 5).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ZyWyQkyydAavk1jw_7Yddg.png" /></figure><h4>ML Algorithms</h4><p>Managing ML algorithms as well as supporting their computational and data transformation needs is arguably the key task ML platforms solve both for Data Scientists and Domain Experts. The ability to quickly experiment and apply different algorithms to a given dataset is key to validate any use case (hypothesis validation).</p><p>Precisely AutoML (Automated Machine Learning) provides further automation for this process, creating many model candidates with different algorithms and evaluating their performance to suggest the best model options to a given prediction objective.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CRY4dQs4Av2p5vx8n_xZ0w.png" /><figcaption>ML Algorithms — Supervised &amp; Unsupervised (#MLSEV Courtesy of BigML Inc)</figcaption></figure><p>In practice, the vast majority of ML use cases in the enterprise (think Marketing, Finance, HR, Manufacturing, etc) do not require Deep Learning or Neural Network algorithms unless there are extraordinary performance requirements, or, in specific use cases where there is a need to deal with unstructured data (e.g. video, images or sound).</p><p>Almost 80–90% of the use cases leverage traditional ML algorithms such as Linear Regression, Logistic Regression, Random Forests or Ensembles which provide solid results and further explainability vs the more opaque Neural Network methods.</p><p><a href="https://www.kaggle.com/kaggle-survey-2019">Kaggle’s survey</a> which nearly comprises 20K respondents (data scientists a significant part) serves as a good proxy to monitor algorithm usage:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*p2db_ncnO9bN3kCfbJV4pw.png" /><figcaption>Blue box: Traditional ML Algorithms — Red Box: CNN, RNN, DNN &amp; Bayesian — Source: Kaggle · The State of Data Science &amp; ML 2019</figcaption></figure><blockquote>References:</blockquote><blockquote>Kaggle’s State of Data Science &amp; ML Survey <a href="https://www.kaggle.com/kaggle-survey-2019">https://www.kaggle.com/kaggle-survey-2019</a></blockquote><blockquote>McKinsey — <a href="https://www.mckinsey.com/~/media/McKinsey/Business%20Functions/McKinsey%20Analytics/Our%20Insights/The%20age%20of%20analytics%20Competing%20in%20a%20data%20driven%20world/MGI-The-Age-of-Analytics-Full-report.ashx">The Age of Analytics</a></blockquote><h4>ML Platforms Architecture</h4><p>For custom-built ML platforms, open-source packages and technologies have been the preferred choice (e.g Uber’s case: <a href="http://hadoop.apache.org/">HDFS</a>, <a href="https://spark.apache.org/">Spark</a>, <a href="http://samza.apache.org/">Samza</a>, <a href="http://cassandra.apache.org/">Cassandra</a>, <a href="https://spark.apache.org/mllib/">MLLib</a>, <a href="https://github.com/dmlc/xgboost">XGBoost</a>, and <a href="https://www.tensorflow.org/">TensorFlow</a>), determining much of the underlying architecture.</p><p>Use cases to be solved and how much of real-time vs static data modeling was required have been the other factors shaping platform architectures. Uber services’ real-time needs (e.g. forecasting ride time) differ from AirBnB’s (e.g. recommender) and are very different from FICO’s (e.g. credit scoring) in terms of access, processing data requirements and expected time to prediction, hence having very distinct architectural needs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_CC2F6WnFhownfHui4myLg.png" /><figcaption>Uber’s ML Platform — Michelangelo, source: Uber engineering <a href="https://eng.uber.com/michelangelo-machine-learning-platform/">https://eng.uber.com/michelangelo-machine-learning-platform/</a></figcaption></figure><p>For example, in the case of Uber, models that are deployed online cannot access data stored in HDFS, making it difficult to retrieve key features from production online databases (e.g. query the UberEATS order service to compute the average meal prep time for a restaurant over a specific period of time).</p><p>Uber’s solution is to pre-compute and store those needed features in Cassandra so latency requirements at prediction time could be met (source: Uber engineering — Michelangelo <a href="https://eng.uber.com/michelangelo-machine-learning-platform/">https://eng.uber.com/michelangelo-machine-learning-platform/</a>).</p><p>Commercial ML Platforms, as opposed to custom-built open-source, are intended to generalize over a vast and diverse array of ML problems and use cases. Their architecture needs to support critical capabilities while allowing sufficient customization via abstraction layers:</p><ol><li>Automatic Deployment and Auto Scaling of Infrastructure (Cloud, Hybrid and On-Premise) — Distributed</li><li><strong>API</strong> — RESTful ML services —Programmatic API-fication and Integration</li><li>Advanced Feature Engineering — <strong>DSL**</strong></li><li>Programmatic use of Algorithms and Models</li><li>Customization with Programmatic Automation for ML Workflows — <strong>DSL</strong></li><li>Front-end: Visualizations and Interface</li><li>Development Tools: Bindings (Python, R, Java, C), Libraries, CLI tools</li></ol><p>** DSL Domain Specific Language</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1012/1*Yh3vQlCwhbx4k9NAzmGV4Q.png" /><figcaption>ML Platform architecture overview (Courtesy of BigML Inc)</figcaption></figure><p>APIs and programmatic automation with DSLs —<strong> Domain Specific Languages </strong>— warrant a spin-off post. AutoML, in particular, will be covered in section 7 of this post series.</p><p>DSLs provide the necessary level of abstraction on top of APIs and the computing infrastructure to deal with complexity, automation, and customization. Examples of DSL approaches are Uber’s DSL for Feature Selection and Transformation (<a href="https://eng.uber.com/michelangelo-machine-learning-platform/">Uber’s ML Platform — Michelangelo</a>) and <a href="https://github.com/whizzml/examples">BigML’s DSL for ML workflow automation</a> (<a href="https://github.com/whizzml/examples">WhizzML</a>).</p><blockquote>References:</blockquote><blockquote>Uber engineering — Michelangelo <a href="https://eng.uber.com/michelangelo-machine-learning-platform/">https://eng.uber.com/michelangelo-machine-learning-platform/</a></blockquote><blockquote>Arxiv: A survey on DSLs for Machine Learning and Big Data <a href="https://arxiv.org/pdf/1602.07637">https://arxiv.org/pdf/1602.07637</a></blockquote><blockquote>PAPIS.io proceedings: The Past, Present, and Future of Machine Learning APIs <a href="http://proceedings.mlr.press/v50/cetinsoy15.pdf">http://proceedings.mlr.press/v50/cetinsoy15.pdf</a></blockquote><blockquote>WhizzML a DSL for ML workflow’s automation <a href="https://bigml.com/whizzml">https://bigml.com/whizzml</a></blockquote><h4>Update on the Hype Cycle for Machine Learning</h4><p>As Enterprises and practitioners go past the learning curve of applying ML in practice, both Machine Learning and Deep Learning as technologies went through and beyond the hype cycle peak, quickly navigating through the ‘Valley of Disillusionment’ (adjustment of expectations). Gartner’s latest Hype Cycle report reflects on this and predicts mainstream adoption to happen in the next 2–3 years.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*tdr5qwvqUo-LcxTzH1xLtQ.png" /><figcaption>Source: Gartner Hype Cycle for DS and ML 2019</figcaption></figure><p>Following suit, new ML platform’s functionalities such as AutoML, MLOps, Explanaible AI or Reinforcement Learning are currently entering the hype curve, quickly becoming overhyped. These new ML technologies, still immature, are not ready for mainstream or massive adoption yet.</p><h4><strong>Built vs Buy</strong></h4><p>For some years now, the market has been divided between open-source and industry solutions.</p><p>From 2012 to 2018, blue chip technology companies implemented custom-built ML platforms for internal use (i.e Facebook’s FBLearning, Uber’s Michelangelo, Twitter’s Cortex, AirBnB’s BigHead, more detailed analysis in Section 4 later on).</p><p>Many of these platforms are primarily based on open-source packages and have been deeply tailored for the specific use cases of those companies.</p><p>Despite the inherent free nature of open source packages (cost benefit on acquisition) and the virtually unlimited capacity for customization, the problem with open source, however, comes with the associated hidden technical debt, and Glue Code in particular (more details in section 2, see <a href="https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems.pdf">Hidden Technical Debt in Machine Learning Systems</a>. <em>D. Sculley et al., Google, NIPS 2015).</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t8Gw1xZALqF0o425pxF_Yw.png" /><figcaption><a href="https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems.pdf"><em>D. Sculley et al., Google, NIPS 2015</em></a></figcaption></figure><p>It’s precisely the cost of configuring, orchestrating, and consolidating the use of different open source packages (Glue Code) what increases costs in the long run, making the build vs buy business case more appealing toward industry solutions.</p><p>The industry has seen a strong evolution of Enterprise-grade ML platform solutions and MLaaS (Machine Learning as a Service or Cloud ML services), with the offering now divided between industry incumbents (e.g. Amazon Sagemaker, Microsoft Azure ML, Google Cloud ML, etc) and challengers (DataRobot, H2O, BigML, Dataiku, etc).</p><p>An important key driver for many Enterprises is <strong>time to market and competitiveness. </strong>As they struggle to scale and get enough ML models successfully to production with a traditional build R&amp;D approach, many are quickly turning to <em>off-the-shelf</em> Data Science &amp; Machine Learning Platforms to accelerate go to market, reduce costs of operationalization and increase success ratio (number of ML models deployed and operationalized).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lzAnmo3t3rHHA37VacYEeA.png" /></figure><blockquote>References:</blockquote><blockquote>Hidden Technical Debt in Machine Learning Systems. <em>D. Sculley et al., Google, NIPS 2015 </em><a href="https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems.pdf">https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems.pdf</a></blockquote><blockquote>Gartner’s Hype Cycle for AI <a href="https://www.gartner.com/smarterwithgartner/top-trends-on-the-gartner-hype-cycle-for-artificial-intelligence-2019/">https://www.gartner.com/smarterwithgartner/top-trends-on-the-gartner-hype-cycle-for-artificial-intelligence-2019/</a></blockquote><p><strong>[end of section 1]</strong></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.slideshare.net%2Fslideshow%2Fembed_code%2Fkey%2Fb2RdHKKRTJLBWm&amp;display_name=SlideShare&amp;url=https%3A%2F%2Fwww.slideshare.net%2Fedslide%2Fmachine-learning-platformization-automl-adopting-ml-at-scale-in-the-enterprise&amp;image=https%3A%2F%2Fcdn.slidesharecdn.com%2Fss_thumbnails%2Fmlplatformizationautoml-edfernandez-mlsev20shortpub-200326123638-thumbnail-4.jpg%3Fcb%3D1585226614&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=slideshare" width="600" height="500" frameborder="0" scrolling="no"><a href="https://medium.com/media/dbfb40a1df8b3f299875d0800b3bd617/href">https://medium.com/media/dbfb40a1df8b3f299875d0800b3bd617/href</a></iframe><p>Companion slides for this post are available at this link [<a href="https://www.slideshare.net/edslide/machine-learning-platformization-automl-adopting-ml-at-scale-in-the-enterprise">presentation at the Machine Learning School #MLSEV — 26th March</a>]</p><h4>About the Author:</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Y-UjBc30kdkXQnZSu1wVGw.png" /><figcaption><a href="https://www.twitter.com/efernandez">Twitter</a></figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9bb477f840c8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/state-of-the-machine-learning-ai-industry-9bb477f840c8">State of the Machine Learning &amp; AI* Industry</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Rise of the Rest: The Unicorn Trend has gone Global]]></title>
            <link>https://medium.com/@efernandez/rise-of-the-rest-the-unicorn-trend-has-gone-global-e67bef1f67c0?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/e67bef1f67c0</guid>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Sun, 25 Dec 2016 07:01:17 GMT</pubDate>
            <atom:updated>2016-12-25T07:01:17.063Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o2X_s4clVk8FaiGLL5-2WQ.jpeg" /></figure><p>Love this pic, had to post it <a href="http://www.linkedin.com/pulse/billion-startup-unicorn-trend-going-global-ed-fernandez?trk=mp-reader-card">again</a>, makes the perfect background for this topic.</p><p>It was Steve Case, former AOL, who has championed the term ‘<em>Rise of the Rest</em>’ on the emergence of tech hubs and VC out of traditional coastal zones in US.</p><p>This is not a US domestic phenomenon, the Unicorn trend (startups valued over $1 Bn) has gone global after being in the works for a few years now. See <a href="http://www.linkedin.com/pulse/billion-startup-unicorn-trend-going-global-ed-fernandez?trk=mp-reader-card">here</a> for educated predictions back in 2015 (somehow neglected at that time, now endorsed by stats).</p><p>Thanks to CB insights who just released their <a href="http://us1.campaign-archive1.com/?u=0c60818e26ecdbe423a10ad2f&amp;id=6a5433117c&amp;e=ea9f215a7f">Christmas newsletter</a> yesterday, we have some closing year stats to share:</p><blockquote>2016 saw close to <strong>300 deals</strong> of VC-backed companies in overlooked regions such as <strong>Latin America, Africa and Oceania</strong>scoring record numbers in terms of # of startup deals.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*9espoYJ5A0qXVQP4.png" /></figure><blockquote>In terms of overall investment in these regions, <strong>this year ended with $1.3 Bn</strong>, with three years in a row now consistently <strong>over $1Bn</strong> in capital deployed.</blockquote><p>More interesting even, and, as anticipated almost two years ago, flashback <a href="http://www.linkedin.com/pulse/billion-startup-unicorn-trend-going-global-ed-fernandez?trk=mp-reader-card">here</a>:</p><blockquote><strong>of the 1.063 deals done in these three regions since 2009, Latin America alone took almost half of them or 47%</strong>.</blockquote><p>The world is truly changing.</p><p>Happy Holidays</p><p>/<a href="http://www.twitter.com/efernandez">ed</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e67bef1f67c0" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Future according to Google Search results]]></title>
            <link>https://medium.com/no-i-wont-fix-your-computer/the-future-according-to-google-search-results-8cdb6c88630b?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/8cdb6c88630b</guid>
            <category><![CDATA[predictions]]></category>
            <category><![CDATA[google]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[future]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Mon, 21 Nov 2016 04:49:23 GMT</pubDate>
            <atom:updated>2016-11-23T02:00:03.679Z</atom:updated>
            <cc:license>https://creativecommons.org/publicdomain/mark/1.0/</cc:license>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*qa3oOrijUu7PSC2V37s3tw.jpeg" /></figure><p>We google everything these days.</p><p>The amount of knowledge out there is immense, and we tap into it everyday through the all mighty search box which, algorithmically, renders most relevant links in the form of a results front page.</p><p>What would be google’s search view on future events, not just historic?. My guess was, given the amount of pages and content referring to future happenings, google’s first page of results could be a pretty good resource for forecast validation.</p><p><a href="http://en.wikipedia.org/wiki/Randall_Munroe"><em>Randall Munroe</em></a> author and creator of <a href="http://www.xkcd.com/">xkcd</a>, the exceptional <em>webcomic </em>(and I quote) <em>about computer science, technology, mathematics, science, philosophy, language, pop culture and romance</em>, did <a href="https://xkcd.com/887/">this back in 2011</a> (hat tip for doing all the work).</p><p>In terms of methodology, these are future events which come up in google search first page after typing ‘by year ####’, ‘in the year ####’ or similar combinations.</p><p>Here is an slideshare presentation with the outcomes, enjoy the trip and don’t forget to go to the last slide to check the ‘<em>other</em>’ predictions, a collection of preposterous and funny fortuneteller bites:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.slideshare.net%2Fslideshow%2Fembed_code%2Fkey%2FewjrW2UJyhhJcR&amp;url=http%3A%2F%2Fwww.slideshare.net%2Fedslide%2Fthe-future-according-to-google-search-xkcd&amp;image=http%3A%2F%2Fcdn.slidesharecdn.com%2Fss_thumbnails%2Fthefutureaccordingtogooglesearch-xkcdanim-161112015744-thumbnail-4.jpg%3Fcb%3D1479589884&amp;key=d04bfffea46d4aeda930ec88cc64b87c&amp;type=text%2Fhtml&amp;schema=slideshare" width="600" height="500" frameborder="0" scrolling="no"><a href="https://medium.com/media/19ad0dc342aabd81b5efa33059609bc8/href">https://medium.com/media/19ad0dc342aabd81b5efa33059609bc8/href</a></iframe><p>An after thought, if we label good predictions (vs bad) we could train a machine learning model to leverage the vast amount of data behind the search box and come up with a reliable future prognosis.</p><p>A nice topic for next post. Stay tuned.</p><p>@<a href="http://www.twitter.com/efernandez"><em>efernandez</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8cdb6c88630b" width="1" height="1" alt=""><hr><p><a href="https://medium.com/no-i-wont-fix-your-computer/the-future-according-to-google-search-results-8cdb6c88630b">The Future according to Google Search results</a> was originally published in <a href="https://medium.com/no-i-wont-fix-your-computer">No, I won’t fix your computer</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Let’s Get Paid Upfront: getting ahead of the carrot-stick game]]></title>
            <link>https://medium.com/@efernandez/lets-get-paid-upfront-getting-over-the-carrot-stick-game-7d7413dbd16e?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/7d7413dbd16e</guid>
            <category><![CDATA[innovation]]></category>
            <category><![CDATA[venture-capital]]></category>
            <category><![CDATA[management]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[growth]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Mon, 11 Jul 2016 08:57:30 GMT</pubDate>
            <atom:updated>2016-07-16T11:26:53.086Z</atom:updated>
            <cc:license>https://creativecommons.org/publicdomain/mark/1.0/</cc:license>
            <content:encoded><![CDATA[<p>In start-up and VC land is all about technology innovation, isn’t it?… or so we tend to think.</p><p>Actually is not, or at least not all, many companies are thriving on simple innovation schemes applied to processes, business models or simple tactics.</p><p>Going further, the concept itself of ‘frugal innovation’, a.k.a creative thinking in the face of constraints, a.k.a doing more with less in the face of critical conditions, shows all the many ways problems can be solved even in the absence of proper resources.</p><p>This wonderful <a href="https://www.ted.com/talks/navi_radjou_creative_problem_solving_in_the_face_of_extreme_limits?language=en#t-67378">TED talk by Navi Radjou</a> is an excellent primer on the topic, covers from fridges running with no electricity in India to advertising billboards producing water literally out of thin air in the rain scarce city of Lima in South America.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*Hq7rVhadNup3QQeUAtURkQ.jpeg" /></figure><p>But today, in this post, we’re going to talk about an even more simplistic yet powerful innovation tactic: incentives.</p><p><strong>First, the theory:</strong></p><p>Incentives and reward schemes are used in companies to drive employee and management behaviors, aligning those toward a set of objectives. Incentives range from simple weekly, monthly or yearly salaries all the way to sophisticated cash bonuses with triggers, accelerators and equity schemes more long term oriented.</p><p>For the vast array of different incentive and compensation mechanisms one thing is true, they all are paid <em>AFTER</em> the activity and expected behavior has happened. Simply put, if you behave and meet your targets, you get paid (carrot), if not, you don’t (stick).</p><p>…There must be a better way. Considering everybody needs to be paid for his/her work, what if we get paid <em>BEFORE?.</em></p><p>With current company incentives lying ahead, in the future, all rewards are perceived as a future ‘gain’ contingent on employee &amp; management behavior and this is an excellent transaction scheme, no doubt. But there is a much bigger motivation factor for humans, actually double at least: <a href="https://en.wikipedia.org/wiki/Loss_aversion"><em>loss aversion</em>.</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/793/1*BdPkSNDs-Ok3Sls-pZU3ZA.png" /><figcaption>Kahneman, D. and Tversky, A. (1984) — ‘Choices, Values and Frames’</figcaption></figure><p>Just look at the grey area in the left (loss quadrant) in comparison with the green area on the right (gain), for the same absolute incentive amount X, you get 2Y motivation for a loss.</p><p>So, what if we re-wire the incentive schemes in a way we can leverage this much bigger motivating factor?</p><p><strong>The Watney rule for Start-ups</strong></p><p>In this <a href="https://medium.com/@firstround/the-watney-rule-for-startups-and-the-return-to-the-old-normal-cba75583365e#.9j585c1ua">letter to their limited partners</a>, First Round venture capital firm set the stage for what is to come this year for start ups.</p><p>Back to the ‘old normal’ and adjusted valuations has put all expectations on start-ups to become deeply conscious on their expenditure and make the most efficient use of capital. Kind of the behavior the <a href="https://en.wikipedia.org/wiki/The_Martian_(Weir_novel)">astronaut Watney showed while stranded on Mars waiting for rescue in the movie The Martian.</a></p><p>Similarly, and while the next ‘rescue’ round of capital comes, Start-ups CEOs need to use their limited resources imaginatively while securing key milestones and monetization in particular happens. So they started paying bonuses in advance.</p><p>Wait… how is that an efficient use of working capital?</p><p>It actually is, and, as <a href="https://en.wikipedia.org/wiki/Irfan_Pardesi"><em>Irfan Pardesi</em></a> told me when we met in San Francisco in May, paying in advance his sales execs was a much better motivator for performance, which grew 15% on average and even 50% in some individual cases.</p><p>Irfan is a serial entrepreneur, founder of Accentuate Capital Markets, a holding company providing financial FX brokerage services from South Africa. Irfan is also a member of <a href="http://www.ypo.org/">YPO, the Young Presidents Organization</a>, a non-profit organization of leaders under 45.</p><p>Paying bonuses and/or the paycheck in advance creates trust and strengthen bonds between the company and its employees/managers, who feel much more confident about the company and react incredibly well to the trust put on them.</p><p>On top of that, think how cool it is to get your money upfront every month, so you can do the things you want now rather than waiting to get that elusive bonus at the end of the year. This is a form of instant gratification in a way, and your company is doing that for you in appreciation of your work.</p><p>There needs to be a catch of course for this model to work and to secure ‘loss aversion’ dynamics work towards objectives. Irfan told me, for this model to work two things must be in place:</p><ol><li>Close management and monitoring of the activity so targets are really realistic and achievable, consistently, each month</li><li>A catch, a future deduction of the paycheck/bonuses if targets are not met, something reasonable and agreed with the employees</li></ol><p>End result? once the process is tuned in, and, securing you build that trusted relationship with your employees, productivity and performance will be boosted significantly (specially at the end of the month when loss aversion feelings kick in). The magic of a 2x motivating factor.</p><p>Actually there is a whole new industry arising around the concept of instant gratification, the precursor to loss aversion.</p><p><a href="https://www.activehours.com/">ActiveHours</a>, a Palo Alto startup offers the possibility to get your monthly salary paid almost in real time, through an app and in cozy hourly installments.</p><p>It’s like everyday is pay-day!</p><p>I’m particularly fond of the trusted relationship model, in my own experience, all stelar teams and epic growth stories come from nimble teams with strong bonds of trust both amongst them and with management, add the right incentives at the right timing, and there you go, the sky is the limit.</p><p><a href="https://medium.com/u/809593f1f2ef">ed fernandez</a></p><p>@efernandez</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7d7413dbd16e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to make machines learn like humans: Brain-like AI & Machine Learning]]></title>
            <link>https://medium.com/no-i-wont-fix-your-computer/how-to-make-machines-learn-like-humans-brain-like-ai-machine-learning-7c07adec3f2e?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/7c07adec3f2e</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[neuroscience]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Fri, 27 May 2016 21:18:48 GMT</pubDate>
            <atom:updated>2018-11-29T04:11:47.736Z</atom:updated>
            <cc:license>http://creativecommons.org/publicdomain/zero/1.0/</cc:license>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/390/1*rLZAaUUAc3QP7q-mnXXeEg.gif" /></figure><p>AI and machine learning are all over us, a simple search on google draws 105 Million entries and counting, google trends shows a growing demand for this search term consistent with the exponential rise of deep learning since 2013, more or less when Google’s X Lab developed a machine learning algorithm able to autonomously browse YouTube to identify the videos that contained cats.</p><p>In 1959, <a href="https://en.wikipedia.org/wiki/Arthur_Samuel"><em>Arthur Samuel</em></a> defined machine learning as a</p><blockquote><em>“Field of study that gives computers the ability to learn without being explicitly programmed”</em></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lBEpfgwOnpaD8ZS98S-7FA.jpeg" /><figcaption>Source: Wikipedia</figcaption></figure><p>AI and machine learning changes the software paradigm computers have been based on for many decades.</p><p>In the traditional computing domain, providing an input, we feed it into an algorithm to produce the desired output. This is the <strong>rule-based framework</strong> the majority of the systems around us still work with.</p><p>We set up our thermostat to a desire temperature (<em>input</em>) and a rule based programming (<em>algorithm</em>) will take care of reading a sensor and activating heating or AC machines to get to the room temperature we want (<em>output</em>).</p><p>The industry has been working relentlessly for many years developing better hardware, software and apps to solve a gazillion problems and use cases around us with programmable solutions. But still, every new functionality or <em>feature</em>, every single new ‘<em>learning</em>’ has still to come via an update of the software (or the firmware itself in hardware).</p><p>Machine learning puts head over heels the rule-based paradigm.</p><p>Given a dataset (input) and a known expected set of outcomes (output), machine learning will figure out the optimal matching algorithm so that, after trained (learning), it can autonomously predict the output corresponding to new inputs.</p><p>The new AI and machine learning paradigm opens up the promise land of <strong>‘self-programming’ machines</strong>, capable of finding the right algorithms to be used in any occasion, this is, providing availability of sufficient input training data, a bottleneck today.</p><p>However, and despite all the incredible progress made in this field, including breakthroughs around deep learning in recent years, <strong>machines are far from matching human ability to learn new patterns</strong>, and worse, <strong>we don’t know how they learned what they learned</strong> nor how they come up with a decision or an specific (wrong) output. We just feed them with big data and ‘<em>tweak</em>’ the machine learning process till we get them to work and deliver the desired outputs within acceptable thresholds of accuracy, but the whole thing remains a <strong><em>‘black box’ </em></strong>(<em>Fodor &amp; Pylyshyn 1988, Sun 2002</em>).</p><p>And they certainly got very efficient and accurate, better than humans in many fronts, no question. AI, machine learning and neural networks are now behind any major service, predicting our credit score, detecting fraud, rendering face recognition, assisting us through Siri, Google, Cortana or Alexa, and soon driving our cars.</p><p>But, as in the old computing paradigm, <strong>the process of learning still requires an ‘<em>update</em>’, what, in machine learning and neural networks jargon is called ‘<em>retraining the network</em>’</strong> with a new dataset and new features required to incorporate new learning or a new functionality.</p><p>Retraining any AI network takes well experienced engineers, top notch hardware (GPUs) and time, a lot of computing time.</p><p>That’s why we can’t teach Siri, Google, Cortana or Alexa new things on the fly. If they don’t understand what we say they typically default to a simple search on the web, we can’t simply tell them ‘<em>learn this new word</em>’ or ‘<em>remember my favorite team are the Red Sox</em>’. Same applies for the rest of large neural networks behind other services, they need to be retrained with the new data and that takes days, weeks or even months depending on the size of the the network and the dataset.</p><blockquote><strong>Now, imagine for a moment if we could teach machines ourselves and make them learn the same way we humans do, wouldn’t that be awesome?</strong></blockquote><p>Imagine if we could teach Siri, Cortana, Google o Alexa new words or expressions, or even new action commands <em>‘hey Alexa, pull out my car from the garage’:</em></p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FCAP3DbyOtGE%3Ffeature%3Doembed&amp;url=http%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DCAP3DbyOtGE&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FCAP3DbyOtGE%2Fhqdefault.jpg&amp;key=d04bfffea46d4aeda930ec88cc64b87c&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/aff94ac1e332fde688d15dadbde1ed2b/href">https://medium.com/media/aff94ac1e332fde688d15dadbde1ed2b/href</a></iframe><p>The answer to this is in the brain.</p><p>And <a href="https://www.youtube.com/watch?v=w4aoQUxqlZg">some researchers</a>, devoted to <strong>reverse engineering cognition mechanisms of the brain</strong> have unlocked <strong>brain-like algorithms and new machine learning models </strong>solving this problem, turning the traditional machine learning ‘black box’ into a ‘clear box’ neural network where <strong>new learnings can happen on the fly, in real time and at a fraction of today’s computational cost</strong> (no retraining over the whole dataset required).</p><p>In a simplistic way, the underlying problem is that all traditional machine learning models are primarily <em>feedforward</em> based, in other words, the basic calculations in the network happen ultimately in the form of a simple multiplication where the output Y is just the input X <em>weighted</em> (feedforward multiplied by W, the Weight). <strong>Y = W * X</strong></p><p>Determining the set of weights W for a given input dataset X with a known labeled output Y is called <em>‘training the network’</em>. The process is long, can take hours, days or even months for large networks, but, once all those weights W are calculated and refined (a process called <em>optimization of weights</em>) the network is capable of amazing wonders like facial recognition or natural language understanding.</p><p>However, as mentioned, if you want the network to learn something new, you need to go back again through all the retraining process and start from the beginning, recalculating and optimizing the new set of weights W.</p><p>But <em>‘this is not how the brain works’</em>, <a href="https://www.youtube.com/watch?v=w4aoQUxqlZg">Tsvi Achler, MD/PhD in neuroscience and BSEE from Berkeley, told us at a talk in Mountain View</a>.</p><blockquote><em>‘The brain does not turn around and recalculates weights, it computes and learns differently during recognition while the context is still available, and it does not only do feedforward, all sensorial neural recognition mechanisms show some form of feedback loop’</em></blockquote><p>In all traditional machine learning methods (<strong>deep learning, convolutional networks, recurrent networks, support vector machines, perceptrons, </strong>etc) there is a ‘disconnect’ between the learning &amp; training phase and the recognition phase. <strong>What Tsvi Achler proposes is not to recalculate (learn) weights but to determine neural network activation (Y, output) by optimizing during recognition, factoring in feedback as well as feedforward, and more importantly, focusing on the current pattern in context </strong>(vs all of the training dataset).</p><p>With this approach and this new machine learning algorithm we can ‘see’ the weights and change them in real time, while recognizing, add new nodes to the network (patterns) and features on the fly without the need to go over the re-training process.</p><p>At his startup <a href="http://optimizingmind.com/">Optimizing Mind</a>, Tsvi ran his machine learning model on a Celeron Quad core laptop, 2GHz, 4GB memory, which is equivalent to a high end smartphone. He tested it against traditional methods such as SVM or KNN and the scalability results were astonishing, showing off up to <strong>two orders of magnitude of computational cost reduction</strong>.</p><p>The ability to embed this new machine learning technology in a smartphone will enable true real time learning from end users’ interaction while preserving data locally (no need to go back and forth to servers).</p><p>The time when we will be able finally to teach machines by ourselves, as well as learn from the environment, all in real time, is getting closer.</p><p>This may be even a first very early step to enable machine to machine learning, and with that, who knows, exponential intelligence maybe?.</p><p>Exciting times ahead, what a moment to be alive.</p><p><a href="https://medium.com/u/809593f1f2ef">ed fernandez</a> <a href="http://twitter.com/efernandez">@efernandez</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7c07adec3f2e" width="1" height="1" alt=""><hr><p><a href="https://medium.com/no-i-wont-fix-your-computer/how-to-make-machines-learn-like-humans-brain-like-ai-machine-learning-7c07adec3f2e">How to make machines learn like humans: Brain-like AI &amp; Machine Learning</a> was originally published in <a href="https://medium.com/no-i-wont-fix-your-computer">No, I won’t fix your computer</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The emerging 4th sector of the economy and a (tangential) AI arms race]]></title>
            <link>https://medium.com/@efernandez/the-emerging-4th-sector-of-the-economy-and-a-tangential-ai-arms-race-20bfa5211cb?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/20bfa5211cb</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[sharing-economy]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Wed, 30 Mar 2016 02:38:14 GMT</pubDate>
            <atom:updated>2016-03-30T02:38:14.604Z</atom:updated>
            <cc:license>http://creativecommons.org/publicdomain/zero/1.0/</cc:license>
            <content:encoded><![CDATA[<p>Social &amp; mission driven enterprises in the age of AI: mixing up two otherwise unrelated albeit mesmerizing topics.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/856/1*A-X4Ex8conECeork36BoCw.jpeg" /><figcaption>From the movie The <em>Matrix Revolutions</em>: Neo waiting to be interrogated by Agent Smith</figcaption></figure><blockquote>Agent Smith Clone 1: It is purpose that created us.</blockquote><blockquote>Agent Smith Clone 2: Purpose that connects us.</blockquote><blockquote>Agent Smith Clone 3: Purpose that pulls us.</blockquote><blockquote>Agent Smith Clone 4: That guides us.</blockquote><blockquote>Agent Smith Clone 5: That drives us.</blockquote><blockquote>Agent Smith Clone 6: It is purpose that defines us.</blockquote><blockquote>Agent Smith Clone 7: Purpose that binds us.</blockquote><p>— <em>From the movie: The Matrix Revolutions</em></p><p>Any entity &amp; system exists for a reason, an objective to meet (even if only for <a href="https://medium.com/no-i-wont-fix-your-computer/two-duck-rabbit-paradigm-shift-anomalies-in-physics-and-one-maybe-in-machine-learning-86e6e1fbdcd7#.6r14r96y8"><em>dissipation purposes</em></a><em> </em>in the case of living organisms).</p><p>AI is no different in this respect, at least for the time being:</p><blockquote>(a self-improving AI may eventually change its own scope of objectives developing some degree of self-control).</blockquote><p>In the short term, narrow AI will be/has been pervasively spreading into all businesses and activities. This is an ongoing process, no news in here.</p><p>In the current economy, incentives are all in to support narrow AI development fostering efficiencies in every aspect of business, industry and vertical.</p><p>The inevitable search for profit or cost reduction will drive corporations to push narrow (and in the future general) AI based technologies to drive these efficiencies (i. e even <a href="https://www.ted.com/talks/andrew_mcafee_are_droids_taking_our_jobs?language=en"><em>replacement of workers</em></a>, <em>Brynjolfsson and McAfee 2011</em>).</p><blockquote>We’re coming to a point where Machine Learning and artificial intelligence is even being <a href="http://www.eecs.mit.edu/news-events/media/energy-friendly-chip-can-perform-powerful-artificial-intelligence-tasks">embedded at chipset level.</a></blockquote><p>This remains a valid statement whilst we remain in a competitive and evolutionary based economic system, the question is open if the economy can shift to a different model, altruism (will talk about this later on).</p><p>Short term (in the following 10–20 years), <strong>a clash of AI clans</strong>, an arms race for profit, will define how this technology will evolve, somehow a deterministic future.</p><p>A simple example are the all mighty everyday virtual voice assistants: Siri, Cortana, Google Now or Amazon Echo, fiercely compete for end users attention.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/700/0*WANtOzVEGJyhV3qI." /></figure><p>While all of this happens and AI goes on making the headlines, <strong>the economic system</strong> is inconspicuously living a profound change as a consequence of <strong>shifting from a for-profit/self-interest into an altruistic model</strong>.</p><blockquote>Non-for profit, social enterprises, public and government entities are all <strong>converging in what’s called the 4th sector of the economy.</strong></blockquote><p>In the context of this tectonic shift, the sharing (or collaborative) economy, a booming greenfield for unicorn startups &amp; disruptive companies today, is some sort of precursor, initial enabler or tadpole of an <a href="https://en.wikipedia.org/wiki/Post-scarcity_economy"><strong>economy of abundance.</strong></a></p><p>These are the sharing &amp; collaborative economies, driving low end disruption, extending on-line access to virtually any resource and democratizing prices for on-demand services (<a href="http://thenextweb.com/entrepreneur/2014/05/18/rise-uberized-economy-means-business/"><em>Uber-ization</em></a>).</p><blockquote><strong>Sharing economy is an early precursor to the economy of abundance</strong>.</blockquote><p>In a sufficiently abundant economy, altruism based models will thrive by its own nature.</p><h3><strong>The 4th sector</strong></h3><p>Over the past few decades, the boundaries between the public (<em>government</em>), private (<em>business</em>), and social (<em>non-profit</em>) sectors have been blurring with many organizations pioneering and blending social and environmental aims with traditional business approaches.</p><p>There are many manifestations of this trend, such us corporate social responsibility, microfinance, venture philanthropy, sustainable businesses, social enterprise, privatization, community development and others. As these activities mature, this is becoming formalized as a <a href="http://www.fourthsector.net/learn/fourth-sector"><strong>‘Fourth Sector’</strong></a> of the economy.</p><blockquote><strong>The mission driven economy is based on a fundamental change in purpose.</strong></blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/0*1NWrMBU1tzsIUCM1." /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/955/0*gi2cMfJW-e8ONgi3." /></figure><h4><strong>“Hybrid” entrepreneurs</strong></h4><p>They do operate today within the constraints of the other three existing sectors. Often requires them to compromise their objectives, complicate their organizational structures, and invent new processes that distract their focus and deplete resources.</p><p>The plain fact is that the infrastructures that have come into being to support the three main sectors are not ideal for these hybrid enterprises.</p><p>Despite this, many have grown and developed major corporations and global brands.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/457/0*c-T0MAIm6lVrShbV." /></figure><p>Toms’ Shoes: the ‘<em>one for one</em>’ business model concept is an example of that. Toms’ offer blends footwear sales with donation, making the buyer a benefactor for impoverished children each time a pair of shoes is sold.</p><p>The company has given away more than 20M shoes, became a global brand with global presence and extended its businesses into other categories such as eyewear, bags and coffee.</p><p>Given its growth and notoriety, Toms is probably the poster child of ‘<em>conscious capitalism</em>’ of this decade.</p><h3><strong>The sharing &amp; collaborative economy</strong></h3><p>Enabling an altruistic economic model</p><p>Uber-ization of the economy, the process of disrupting existing industries towards everything-as-a-service, creates enormous efficiencies leveraging technology, enabling availability of (otherwise idle) resources for on-demand services.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*xxEOJ-yOkVgdSrF-." /></figure><p>At the same time, we are moving from a transactional, ownership based economy to a <em>pay-per-use</em> and <em>subscription service</em> economy.</p><p><strong><em>It’s no longer about owning</em></strong>, but having <strong><em>convenient access to assets</em></strong> when needed or wanted. Even access to luxury is democratizing.</p><blockquote>With the uber-ization of industries, prices move to a dynamic model, self-regulating with demand, which, again, leverage efficiencies to deliver better services at lower prices by enabling idle resources.</blockquote><p><a href="http://venturebeat.com/2015/06/04/the-sharing-economy-has-created-17-billion-dollar-companies-and-10-unicorns/">The sharing economy has already created $17 Bn companies, with 60.000 employees and 10 Unicorns</a> (startups with value above $1Bn). This sector of the economy is poised to be a significant engine for resources and assets sharing. This is an early foundation for altruism-based services, core to the 4th sector activity.</p><blockquote>As an example, imagine Uber drivers, who, once a week on some spare time could offer rides for free to moms and their children on their way to public libraries.</blockquote><p>Or e-Bay giving away toys for free for children in Christmas, or Airbnb (real state owners), offering shelter for a week to people seeking jobs in a different city or country. Or Elance (freelancers) offering some spare time from teachers for people in need of education.</p><p>Many similar use cases are already happening at a relative small scale.</p><p><a href="http://www.huffingtonpost.com/melissa-sher/how-a-drycleaning-busines_b_4900366.html">Dry cleaners in Portland offer their services for free to unemployed, and also in New York.</a></p><p>The sharing (or collaborative) economy is just enabling altruistic offerings at a bigger scale. This sector is thriving right at the crossroad between traditional economy and the non-for-profit &amp; public sectors. The latest two now blending into the 4th sector.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/739/0*MiqBnawNwPaokr2F." /><figcaption>Handwritten notes from the plane, couldn’t find any well designed graph to express the idea, sorry</figcaption></figure><h3><strong>An afterthought</strong></h3><p>Considering the acceleration of globalization fostered by ubiquitous Internet access, mobile computing and social networks, now more than ever <strong>the world is starting to think and act globally.</strong></p><p>A global world, with global altruistic interests, represented and manifested in the 4th sector of the economy will certainly <strong>leverage technologies and AI in particular for the greater good.</strong></p><p>Oversimplifying this picture, we could think of for-profit; for-control AI oriented developments, relentlessly looking for better efficiencies vertically (traditional economy), coexisting with altruistic, mission driven AI entities at the other side of the spectrum, focused on solving global human problems and challenges (the 4th sector).</p><p>With the development of narrow AI in this context, a strong AI or AGI will emerge eventually. <em>This is the moment when humanity, as we understand it, may be exposed and challenged.</em></p><p>Heavy development of narrow AI will lead to strong AI, eventually leading also to ‘bad’ AI or ‘misused’ AI in the same way other technologies were and are wrongfully used.</p><blockquote>The 2004 <a href="https://en.wikipedia.org/wiki/Madrid_train_bombings">Madrid train bombers</a>, who killed 191 people and wounding 1,800, downloaded their bomb-making instructions from the Internet.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/450/0*FMnXnPiB7s5ODpkO." /></figure><blockquote>After the 1995 <a href="https://en.wikipedia.org/wiki/Oklahoma_City_bombing">Oklahoma City bombing</a>, anonymous <a href="https://en.wikipedia.org/wiki/Usenet">UseNet</a> posts criticized the construction of the bomb, and offered suggestions on how to overcome the failure of the bomb to do its maximum intended damage.</blockquote><p>Many scientists and personalities of our time like Hawking, Elon Musk or Wozniak have already taken steps towards bringing awareness to this issue.</p><blockquote><strong>To join these efforts, an open letter against autonomous AI powered weapons can be </strong><a href="http://futureoflife.org/AI/OPEN_LETTER"><strong>signed here</strong></a><strong>.</strong></blockquote><h3><strong>Another afterthought</strong></h3><p>The 4th sector will definitively leverage new technologies and AI eventually for the benefit of humanity, and this is where <strong>history gives us hope:</strong></p><p><em>The Apollo program was a public funded enterprise involving more than 400.000 talented individuals led by NASA with a common objective, to land a man on the moon.</em></p><p>It was the product of the Space race in the 20th century and left a technology legacy that <em>has improved our lives in countless ways</em>.</p><blockquote>Reaching the moon or reaching AGI is a human race level task.</blockquote><blockquote>Even if posing risks or threats (nuclear power is another example), history proves the human race has done the right thing to preserve the species,</blockquote><p>…at least so far.</p><p>Richard Dawkins, in his book, <a href="https://en.wikipedia.org/wiki/The_Selfish_Gene"><em>the selfish gene</em></a>, gave us all the clues about the meaning of life and the purpose of our existence. <strong>Preserving our species (by self-replication for now) remains our ultimate goal</strong>.</p><p>That goal leads us inexorably to <strong><em>space exploration and colonization</em></strong> (as the other scenario, facing extinction in a star system with a deadline, may not be ideal) and AGI is a major enabler as well as an earlier event to it.</p><p>We just need to get it right within this century.</p><blockquote>@<em>efernandez</em></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=20bfa5211cb" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Two Duck-Rabbit Paradigm-Shift Anomalies in Physics and One (maybe) in Machine Learning]]></title>
            <link>https://medium.com/no-i-wont-fix-your-computer/two-duck-rabbit-paradigm-shift-anomalies-in-physics-and-one-maybe-in-machine-learning-86e6e1fbdcd7?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/86e6e1fbdcd7</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[science]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Tue, 09 Feb 2016 05:55:51 GMT</pubDate>
            <atom:updated>2016-03-21T04:32:30.586Z</atom:updated>
            <cc:license>https://creativecommons.org/licenses/by-sa/4.0/</cc:license>
            <content:encoded><![CDATA[<p>You never know what a meeting for a quick coffee in Palo Alto can turn into.</p><p>What was supposed to be an ‘informal’ chat (if there is such thing when talking with PhD’s) about <a href="https://www.youtube.com/watch?v=S8zGWaRFVPQ&amp;feature=em-upload_owner"><em>feedforward-feedback machine learning models</em></a>, turned into a philosophical discussion on <em>duck-rabbit</em> paradigm shifts (disclaimer 1: <em>I’m just a nerd without credentials either topic you choose, with a genuine interest though</em>)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-fU5oZmkoZ5gGFiZDPaCfw.jpeg" /></figure><h3>First, the <em>theory</em>:</h3><blockquote>I see a rabbit, You see a duck.</blockquote><p><em>Thomas Kuhn</em> described the nature of scientific revolutions back in 1962 (his book <a href="https://en.wikipedia.org/wiki/The_Structure_of_Scientific_Revolutions"><em>The Structure of Scientific Revolutions</em></a>).</p><p>A contrarian back in the time, as he re-defined progress by moving from <em>development-by-accumulation</em> (on pre-established assumptions) into <em>paradigm shifts</em>, or revolutions in scientific progress by looking into anomalies inferring a drastic change of assumptions.</p><p>In other words, Kuhn advocated for a change of rules over the pre-existing framework as the ultimate scientific progression method.</p><blockquote>The Copernican revolution, Newton’s reformulation of gravity, Einstein’s relativity or Darwin’s evolution all were ‘anomalies’ as theories</blockquote><blockquote>Sun vs Earth at the center, relativity vs linear spacetime, Apes evolving into Humans vs creation.</blockquote><p>The <em>ethos</em> of the scientific progress theory rests on identifying the right anomalies which support new paradigms. <strong>Anomalies come up as revolutions in disguise</strong> and, utterly (<em>and I love this</em>), expand on the previous paradigm which ends up nested within remaining perfectly valid.</p><p>Anomalies create rejection by opposition (<em>it’s a Duck!, no is not, it’s a Rabbit</em>), but after the new paradigm takes over (…<em>I can see the Duck now ?!?</em>) both paradigms co-exist (<em>it’s a duck AND a rabbit!</em>, illustration above)</p><blockquote>For a true paradigm shift to happen, the anomaly needs to grow from exception into alternative:</blockquote><blockquote>‘<em>it’s a Rabbit AND a Duck!’</em></blockquote><p>Ok, fair enough on the lecture, <em>where is this Machine Learning anomaly this click-bait headline was all about?</em></p><p>It’s coming, bear with me, will be worth the reading while we get there, first, a couple of <em>jaw-dropping-no-longer-anomalies-but-paradigm-shifts</em>: the first one explains the origin (and meaning) of life, the second one may redefine physics forever.</p><h3>1. <a href="https://www.quantamagazine.org/20140122-a-new-physics-theory-of-life/">Dissipation-driven adaptation</a></h3><p><strong><em>Jeremy England. MIT, biophysics</em></strong></p><p>This incredibly simple idea is intuitively so powerful and makes so much sense that is difficult to resist. It explains Darwinian evolution and survival of the fittest, ultimately dwelling on the inherent reasons why life comes to exist.</p><p>At an intuition level, in Jeremy’s words:</p><blockquote>“You start with a random clump of atoms, and if you shine light on it for long enough, it should not be so surprising that you get a plant”</blockquote><p>Jeremy, a MIT’s researcher, has developed a mathematical model based on current physics, exerting that a given set of atoms, exposed to a continuous source of energy (i.e the Sun), surrounded by a hot bath (i.e the Ocean) will self-organize to dissipate energy in the most efficient way (i.e life).</p><p>We, <em>carbon-based lifeforms</em>, in Spock’s vulcan language, are much better at dissipating heat than inanimate objects. Both living and non-living organisms show this efficiency driven, self-organizing dissipation behavior.</p><p>Photosynthesis and self-replication (of RNA molecules, precursor to DNA-based life) are consequences of dissipation driven adaptation.</p><p>Photosynthesis is about capturing sunlight energy transforming and storing it chemically (sugar) so it can be transported and reprocessed for plant growth and replication (hence forests).</p><p>Don’t believe it yet? see for yourself, here is <em>Dr. Hubler’s</em> Stanford professor experiment on self-wiring ball bearings, an example of dissipation driven matter structure reorganisation.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FPeHWqr9dz3c%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DPeHWqr9dz3c&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FPeHWqr9dz3c%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/8819ac94bee8bedcf4f35b9ec1f43401/href">https://medium.com/media/8819ac94bee8bedcf4f35b9ec1f43401/href</a></iframe><h3>2. <a href="http://www.popsci.com/science/article/2012-09/book-excerpt-there-no-such-thing-time">Timeless physics</a></h3><p><a href="http://www.platonia.com/"><strong><em>Julian Barbour</em></strong></a><strong><em>. British Physicist. Quantum gravity.</em></strong></p><p>Remember the school/college days?:</p><p>Speed = space / time, power = energy / time, theorem of calculus df/dt, Maxwell’s equations, Einstein’s relativity, Thermodynamics, etc, etc.</p><blockquote>In physics, anything dealing with change, requires t (time) as a variable, isn’t it? …may be not any more.</blockquote><p>How is it possible anyone dares to defy physics by removing time from centuries old proven equations?</p><p>If you think about it, time is just an abstraction we use to facilitate our understanding of how things (matter in particular) transitions from one state to another (change). Because we live in an universe governed by the <strong><em>2</em></strong><em>nd law of thermodynamics</em> (fighting an increasing<em> </em><a href="https://en.wikipedia.org/wiki/Entropy"><em>entropy</em></a>) we perceive linear time as our most reliable and dependable reference.</p><p>At an intuition level, if we look at the Universe as a simple but immense ‘cloud’ of matter in permanent change (motion) since the big bang occurred, then, if we reduce our view to atoms transitioning for one state to another, you could remove time entirely.</p><p>Our Universe could be viewed as a continuum of matter in ‘motion’ (actually, according to Barbour, not motion, but matter in permanent change, removing in full the spacetime continuum).</p><blockquote>Our senses and limited computing capacities can’t deal with such enormous entity so we take partial ‘pictures’ with a reference point (time) to deal with reality and make sense out of it (a constrained and partial view).</blockquote><p>Another intuitive line of thought, if Newton’s physics were based on linear time (absolute fixed time), and then Einstein’s relativity made time relative, hence flexible (unlocking a bigger scope for physics), what if we make time super-mighty-flexible to the point of making it irrelevant? wouldn’t this even offer an even wider and extended view as we remove the constraints of a time dimension itself?</p><blockquote>“If, at first, the idea is not absurd enough, then there is no hope for it”. A. Einstein</blockquote><p>…. <em>and now, </em><a href="https://en.wikipedia.org/wiki/And_Now_for_Something_Completely_Different"><em>for something completely different</em></a> (Monty Python)</p><h3>3. Flexible recognition in machine learning</h3><p><strong>Tsvi Achler. Neuroscience (PhD), Medicine (MD), Electrical Engineering (BS-EECS Berkeley) — </strong><a href="http://optimizingmind.com/"><strong>Optimizing Mind</strong></a></p><p>Our brains are ‘<em>computationally flexible</em>’, this means we can immediately learn and use new patterns as we encounter them in the environment.</p><p>We actually ‘<em>like</em>’ to develop those patterns, as we unleash our curiosity, see and try new things for the sake of enjoyment.</p><p>Learning, tasting and traveling feed our brains with new patterns. Riding a hover-wheel, flying a drone, speaking to Amazon echo or playing a new game are examples of behaviors where our brains confront and develop new patterns for different uses and purposes.</p><p>Now, let’s look into it from a machine learning perspective:</p><p><em>(disclaimer 2: as said at the beginning of this post I’m just a nerd without credentials trying to convey the message. Standing in the shoulder of giants when I wrote what you’re about to read on)</em></p><p>Tsvi Achler has been studying the brain from multidisciplinary perspectives looking for a single, compact network with new machine learning algorithms and models who can display brain phenomena as seen in electrode recordings, performing <strong>flexible recognition</strong>.</p><blockquote>The majority of popular models of the brain and algorithms for machine learning remain <em>feedforward</em> and the problem is that even when they are able to recognitze they are not optimal for <em>recall, symbolic reasoning or analysis.</em></blockquote><p>For example you can ask a 4 year old why they recognized something the way they did or what do they expect a bicycle to look like. However it is difficult to do the same with current machine learning algorithms.</p><p>Let’s take the example of recognizing a bicycle over a dataset of pictures. A bicycle, from a pattern perspective, would consist of two same or similar size wheels, a handle, and some sort of supporting triangular structure.</p><p>In feedforward models the weights are optimised for successful recognition over the dataset (of a bicycle in our example).</p><p>Feedforward methods will learn what is unique within a bicycle compared to all other items in the training set and learn to ignore what is not unique. The problem is that subsequently it is not easy to recall what are the original components (two wheels of same of similar size, a handle, a supporting triangular structure) that may or may not be shared with other items.</p><p>Moreover when something new must be learned, feedforward models have to figure out from what is unique to the new item but not to the bicycle and other items it already knows how to recognize. This requires re-doing learning and rehearsing all over the whole dataset.</p><p>What Tsvi suggests is to use a <strong><em>feedforward-feedback machine learning model</em></strong> to estimate uniqueness during recognition by <em>performing optimization on the current pattern</em> that is being recognised, and determining <em>neuron activation</em>. (this is NOT optimization to learn weights by the way).</p><p>With this distinct model, weights are no longer feedforward, learning is more flexible and can be much faster, as there is no need to do rehearsal over the whole dataset.</p><blockquote>In other words, <strong>this model is closer to how our brain actually works</strong>, as we don’t need to rehearse a whole dataset of samples to recognize new things.</blockquote><p>Think about it, how many samples of the much hyped hoverwheels do you need to see first before recognizing the next one on the street?. Same for a bicycle.</p><p>And, the most important thing, with feedforward-feedback models learning happens with significantly fewer data.</p><p><strong>Much less data required to learn, an much faster learning.</strong></p><p><strong>Optimization during recognition</strong> displays also properties observed in brain behaviour and cognitive experiments, like predicting, oscillations, initial bursting with unrecognized patterns (followed by a more gradual return to the original activation) and more importantly even, <strong>speed-accuracy trade off </strong>(so here is your catch if you were looking for it).</p><blockquote>All in all, feedforward-feedback models will make machines learn faster using less data.</blockquote><blockquote>They also mimic better how our brain works.</blockquote><p>I met Tsvi for the first time at a t<a href="https://www.youtube.com/watch?v=S8zGWaRFVPQ">alk in Mountain View: available here</a>.</p><p>I will be helping him and his startup along his journey which (as all new ventures) starts with funding, so if anyone has an interest or wants to know more please <strong>do not hesitate to reach out and leave a message for Tsvi or me in the comments</strong>, or even better, tweet me at <a href="http://www.twitter.com/efernandez"><strong><em>@efernandez</em></strong></a>.</p><p><em>Thanks also to Bart Peintner, Co-founder &amp; CTO at Loop.ai, for his advice, insights and shared interest for the ideas mentioned in this article (note-to-ourselves: keep always bandwidth in your mind to entertain challenging ‘anomalies’)</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=86e6e1fbdcd7" width="1" height="1" alt=""><hr><p><a href="https://medium.com/no-i-wont-fix-your-computer/two-duck-rabbit-paradigm-shift-anomalies-in-physics-and-one-maybe-in-machine-learning-86e6e1fbdcd7">Two Duck-Rabbit Paradigm-Shift Anomalies in Physics and One (maybe) in Machine Learning</a> was originally published in <a href="https://medium.com/no-i-wont-fix-your-computer">No, I won’t fix your computer</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The best answer to “Will computers ever be as smart as humans?” is probably “Yes, but only briefly”]]></title>
            <link>https://medium.com/no-i-wont-fix-your-computer/the-best-answer-to-will-computers-ever-be-as-smart-as-humans-is-probably-yes-but-only-briefly-a1a6f27e4bc6?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/a1a6f27e4bc6</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[future]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Sun, 17 Jan 2016 20:44:08 GMT</pubDate>
            <atom:updated>2016-01-18T04:18:23.590Z</atom:updated>
            <cc:license>http://creativecommons.org/publicdomain/zero/1.0/</cc:license>
            <content:encoded><![CDATA[<p><em>Vernor Vinge</em> coined this descriptive question-answer posted in an IEEE spectrum essay almost a decade ago.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OYEG-XIwR6u1mM1-ws-Hgg.jpeg" /></figure><p>Artificial Intelligence, as opposed to what we have seen so far, radio, TV, PC, smartphones, or the internet, can’t be tracked as a single technology in adoption process, it is too pervasive and has way too many different manifestations. Internet can be measured in terms of penetration, by the number of connections or traffic, AI can’t.</p><p>However, there is a conspicuous set of drivers or <strong>Intelligence accelerators behind AI evolutionary path </strong><em>(previous work by Luke Muehlhauser, Anna Salamon; Machine Intelligence Research Institute)</em></p><p>Here are six intelligence accelerators. Beyond those technology based enablers and drivers, the economic incentives will eventually create an arms race of AI systems, a <em>clash of AI Clans.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/497/1*2ss07CcYlIEfMWxFZX_f9w.jpeg" /></figure><p>Early signs of the AI arms race are already here: Google and Facebook, among others have been protagonists of the machine learning open-sourcing frenzy lately, <a href="http://www.wired.com/2015/12/facebook-open-source-ai-big-sur/"><strong>starting a race to scale and achieve network effects by releasing deep learning tools</strong></a><strong> to the public</strong>.</p><p>1. <strong>More than Moore (MtM) Hardware</strong>: quantum computing, spintronics and related technologies. A new computing paradigm.</p><p>2. <strong>Better &amp; more efficient algorithms</strong>.</p><p>IBM’s Deep Blue played chess at the level of world champion Garry Kasparov in 1997 using about 1.5 trillion instructions per second (TIPS), but a program called Deep Junior did it in 2003 using only 0.015 TIPS.</p><p>Thus, the computational efficiency of the chess algorithms increased by a factor of 100 in only six years (Richards and Shaw 2004).</p><p>3. <strong>Big Data &amp; Analytics</strong> (Massive datasets).</p><p>The greatest leaps forward in speech recognition and translation software have come not from faster hardware or smarter hand-coded algorithms, but from access to massive data sets of human-transcribed and human-translated words (<em>Halevy, Norvig, and Pereira 2009</em>).</p><p>Datasets are expected to increase greatly in size in the coming decades, and several technologies promise to actually outpace “Kryder’s law” (<em>Kryder and Kim 2009</em>), which states that magnetic disk storage density doubles approximately every 18 months (<em>Walter 2005</em>).</p><p>4. <strong>Progress in psychology and neuroscience.</strong></p><p>Cognitive scientists have uncovered many of the brain’s algorithms that contribute to human intelligence (<em>Trappenberg 2009; Ashby and Helie 2011</em>).</p><p>Methods like <strong>neural networks </strong>(imported from neuroscience) and <strong>reinforcement learning</strong> (inspired by behaviorist psychology) have already resulted in significant AI progress, and experts expect this insight-transfer from neuroscience to AI to continue and perhaps accelerate (<em>Van der Velde 2010; Schierwagen 2011; Floreano and Mattiussi 2008; de Garis et al. 2010; Krichmar and Wagatsuma 2011</em>).</p><p>5. <strong>Accelerated crowd sourced science efforts.</strong></p><p>Finally, new collaborative tools, open source projects and other corporate driven initiatives as Google Scholar are already yielding results such as the Polymath Project, which is rapidly and collaboratively solving problems in mathematics (<em>Nielsen 2011</em>).</p><p>6. <strong>Economic incentives</strong></p><p>As the capacities of “narrow AI” programs approach the capacities of humans in more domains (<em>Koza 2010</em>), <strong>there will be increasing demand to replace human workers with cheaper, more reliable machine workers</strong> (<em>Hanson 2008, 1998; Kaas et al. 2010; Brynjolfsson and McAfee 2011</em>).</p><p><strong>First-mover incentives</strong>. Once AI looks to be within reach, political and private actors will see substantial advantages in building AI first.</p><p>AI could make a small group more powerful than the traditional superpowers — a case of “bringing a gun to a knife fight”.</p><p>The race to AI may even be a “winner take all” scenario. Thus, political and private actors who realize that AI is within reach may devote substantial resources to developing AI as quickly as possible, provoking <strong>an AI arms race</strong> (<em>Gubrud 1997</em>).</p><p><strong>An AI arms race will eventually happen amid social and economic changes</strong>. Changes that have already started.</p><p>More importantly, there are signs of structural shifts in the economy related to the core principles of the system itself. As poverty reduces and wealth redistributes gradually, <strong>the economy starts shifting from a for-profit (self-interest) model into an altruistic based one.</strong></p><p>Non-for profit, social enterprises, public and government are all converging in what is called <strong>the 4th sector of the economy.</strong></p><p>(continuation to: <a href="https://medium.com/no-i-wont-fix-your-computer/the-second-arms-race-artificial-intelligence-b33a7b147a00#.3wkj6hqgx">The Second Arms Race: AI</a> and <a href="https://medium.com/no-i-wont-fix-your-computer/an-end-to-moore-s-law-the-dawn-of-a-new-computing-era-80c1f466ad37#.iir8dfa0r">An End to Moore’s Law</a> to be continued: <em>The emerging 4th sector of the economy: Social &amp; mission driven enterprises and the AI arms race</em>)</p><blockquote><a href="http://www.twitter.com/efernandez">@e<em>fernandez</em></a></blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a1a6f27e4bc6" width="1" height="1" alt=""><hr><p><a href="https://medium.com/no-i-wont-fix-your-computer/the-best-answer-to-will-computers-ever-be-as-smart-as-humans-is-probably-yes-but-only-briefly-a1a6f27e4bc6">The best answer to “Will computers ever be as smart as humans?” is probably “Yes, but only briefly”</a> was originally published in <a href="https://medium.com/no-i-wont-fix-your-computer">No, I won’t fix your computer</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An end to Moore’s Law: The Dawn of a New Computing Era]]></title>
            <link>https://medium.com/no-i-wont-fix-your-computer/an-end-to-moore-s-law-the-dawn-of-a-new-computing-era-80c1f466ad37?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/80c1f466ad37</guid>
            <category><![CDATA[tech]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[science]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Fri, 04 Dec 2015 06:25:23 GMT</pubDate>
            <atom:updated>2015-12-04T06:28:20.525Z</atom:updated>
            <content:encoded><![CDATA[<p>(continuation to: <a href="https://www.linkedin.com/pulse/second-arms-race-artificial-intelligence-ed-fernandez"><em>The Second Arms Race: Artificial Intelligence</em></a>)</p><h3>An end to Moore’s Law (Mack 2011)</h3><p>Many information technologies have evolved at exponential rate (<em>Nagy et al, 2011</em>), Moore’s law, stating the transistor count doubles every 2 years, has been at the core of causality for 50 years.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/0*0_A7oBQtMPgCJaYL.png" /></figure><p>But this trend may not hold for much longer (<em>Mack 2011, Lundstrom 2003</em>) as per physical limitations of silicon, or, maybe we <a href="http://spectrum.ieee.org/semiconductors/processors/the-multiple-lives-of-moores-law">don’t see the forest for the trees</a>.</p><ul><li><em>There are limits to the exponential growth inherent in each paradigm. Moore’s law was not the first paradigm to bring exponential growth to computing, but rather the fifth.</em></li><li><em>In the 1950s they were shrinking vacuum tubes to keep the exponential growth going and then that paradigm hit a wall. But the exponential growth of computing didn’t stop.</em></li><li><em>It kept going, with the new paradigm of transistors taking over. Each time we can see the end of the road for a paradigm, it creates research pre, quest for the pressure to create the next one.</em></li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/485/0*wmISdUebHJeWW7ze.png" /></figure><ul><li><em>That’s happening now with Moore’s law, even though we are still about fifteen years away from the end of our ability to shrink transistors on a flat integrated circuit.</em></li><li><em>We’re making dramatic progress in creating the sixth paradigm, which is three-dimensional (quantum) molecular computing.</em></li></ul><p><em>Ray Kurzweil — The Singularity is near</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/349/0*7925ApNDUcfyaaIs.png" /></figure><h3>The dawn of a new computing era: More than Moore MtM</h3><p>Moore’s law will come to an end as a consequence of physical limitations of silicon; three dimensional quantum computing is poised to take over as the new paradigm.</p><p><strong>Quantum computing timeline</strong>:</p><p>2013</p><blockquote>Coherent superposition of an ensemble of approximately 3 billion qubits for 39 minutes at room temperature. The previous record was 2 seconds.</blockquote><p>2014</p><blockquote>Documents leaked by Edward Snowden confirm the Penetrating Hard Targets Project, by which the National Security Agency seeks to develop a quantum computing capability for cryptographic purposes.</blockquote><blockquote>Scientists transfer data by quantum teleportation over a distance of 10 feet (3.048 meters) with zero percent error rate, a vital step towards a quantum Internet.</blockquote><p>2015</p><blockquote>Optically addressable nuclear spins in a solid with a six-hour coherence time.</blockquote><blockquote>Quantum information encoded by simple electrical pulses.</blockquote><blockquote>Quantum error detection code using a square lattice of four superconducting qubits</blockquote><blockquote>Quantum computing promises to augment computing power a billion fold, however, we may not need to get there to develop a strong Artificial Intelligence, one that has the capacity to improve and evolve by itself.</blockquote><blockquote>The expectation is that soon after we reach a strong AI matching a human brain, the ability to replicate it rapidly and limitlessly will generate a self-improving general AI, which in turn would accelerate intelligence exponentially.</blockquote><p><a href="http://twitter.com/efernandez"><em>@efernandez</em></a></p><p>Next: <a href="https://www.academia.edu/s/484e7034e9?source=news"><strong><em>An Explosion of Intelligence: The A.I. Arms Race</em></strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=80c1f466ad37" width="1" height="1" alt=""><hr><p><a href="https://medium.com/no-i-wont-fix-your-computer/an-end-to-moore-s-law-the-dawn-of-a-new-computing-era-80c1f466ad37">An end to Moore’s Law: The Dawn of a New Computing Era</a> was originally published in <a href="https://medium.com/no-i-wont-fix-your-computer">No, I won’t fix your computer</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Second Arms Race: Artificial Intelligence]]></title>
            <link>https://medium.com/no-i-wont-fix-your-computer/the-second-arms-race-artificial-intelligence-b33a7b147a00?source=rss-809593f1f2ef------2</link>
            <guid isPermaLink="false">https://medium.com/p/b33a7b147a00</guid>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[singularity]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[ed fernandez]]></dc:creator>
            <pubDate>Fri, 30 Oct 2015 20:20:01 GMT</pubDate>
            <atom:updated>2015-10-30T23:57:31.229Z</atom:updated>
            <cc:license>http://creativecommons.org/licenses/by/4.0/</cc:license>
            <content:encoded><![CDATA[<p><em>posted originally at </em><a href="http://www.naiss.io/blog"><em>naiss.io/blog</em></a></p><blockquote>The second arms race is actually the third. The first one was the naval race during World War I, followed by the Cold War between United States and the Soviet Union, scaling up nuclear weaponry right after the end of World War II.</blockquote><blockquote>The human race has been able to manage the prisoner’s dilemma inherent to these competitions so far, and now faces a new test with the advent of a new technology breakthrough: Artificial Intelligence.</blockquote><blockquote>This is a series of articles on the topic providing a vision toward an artificial intelligence explosion in the context of current economic changes supporting a shift in our economy toward an altruistic model.</blockquote><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0lZraFMiL-EX8a61wpPFwg.jpeg" /></figure><p><strong>Introduction</strong>:</p><p>· <strong>Technological singularity seems plausible</strong> and recent advancements in machine learning and AI suggest the ‘intelligent explosion’ event is within reach <strong>in this century</strong>.</p><p>· <strong>An arms race of narrow AI entities </strong>will happen in the framework of today’s traditional economy. <strong>Strong intelligence or AGI will eventually emerge</strong> followed by an explosion of intelligence.</p><p>· New globalization processes driven by technology are fueling <strong>the sharing economy</strong>, as well as<strong> the 4th sector where public, non-profit, social and mission oriented enterprises are converging.</strong></p><p>· <strong>The 4th sector is poised to grow and thrive enabled by the sharing and collaborative economy; </strong>mission driven enterprises will have more resources enabling them to play a key role<strong> shaping the right path for AI evolution.</strong></p><p>· The <strong>AI arms race</strong> will provide <strong>‘good’ and ‘bad’ entities</strong> in the context of existing and new economy environments (traditional and altruistic economies)</p><p>· <strong>We, humans, as a species, can succeed managing the risks of a superintelligence event as we did in the past overcoming other technology threats </strong>(i.e nuclear).</p><p>__________________________________________________________________</p><p>We have the capacity to anticipate the future with a certain degree of precision. Our prediction accuracy is lower as we increase the time horizon we aim at.</p><p>It’s pretty straight forward for us to predict short term events, those more likely to impact our survival chances; mechanical or physical, like anticipating when a car is going to cross at the juncture we are on, or, more long term and qualitative, anything related to replicating our gene pool, for instance the chances to date a specific person of the opposed sex.</p><p>However, when we look further ahead in time, and, because of our brainpower limitations and the effort required we struggle to foresee all potential possibilities and combinations.</p><p>Our brains, during evolution, developed a pattern-based approach to efficiently solve this problem. Identifying patterns allow us to see the big picture of a possible future, although we remain unable to predict the smaller details within (stacking up to conform to the pattern).</p><blockquote><em>The </em><strong>Singularity</strong><em>, as defined by </em>Ray Kurzweil<em>, arguably the biggest ambassador of this concept in our times, is a period of time in future where technological advances will evolve so rapidly (</em>exponentially<em>) that humanity will not be able to keep up with them.</em></blockquote><p>This definition needs to be broad because is a concept coined after careful analysis of the evolution of many technologies. It looks into historical data and speed of change rather than specific events themselves (although Singularity is mostly associated to the dawn of a super-intelligence entity capable of self-improvement).</p><p>We say <em>can’t see the forest for the trees</em> referring to short term events clouding our ability to see the big picture. The opposite is true for forward-looking statements.</p><p>With sufficient historical data we can develop patterns (and see the forest) but we will remain clueless about details (trees).</p><p>To state the analogy, let’s have a look at a practical example, a piece of technology we are all very familiar with, our phones.</p><p>Wireless phones (<em>smartphones</em>) have undoubtedly been the protagonists of the technology revolution in recent times.</p><p>The way smartphone technology has been adopted is well described by the diffusion of innovations theory (<em>Everett Rogers — 1962</em>), expressed graphically by an S curve (logistic function) or the widely popular bell curve (derivative of the S curve).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/500/1*UlHLMIIh2-OJxvy3qACsIg.png" /></figure><p>The process is well documented using available data from smartphone manufacturers (sales of devices over time) to the point we can track and predict with a certain degree of accuracy what the future will be for this particular technology.</p><p>The graph for US smartphone adoption, now above 70% penetration (<a href="http://www.asymco.com/2013/11/06/the-diffusion-of-iphones-as-a-learning-process/"><em>Horace Dediu — Asymco</em></a>), follows accurately the bell curve pattern to the point we can predict overall sales volumes in the years to come (this would be the forest in our analogy), but we are unable to predict which manufacturer will get the greater share in the same way we couldn’t predict Apple’s iPhone explosive growth since 2007 (those are the trees).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/490/1*7tQw7_aBNIXrGoaa4zsmGA.png" /></figure><p>Thus, in a competitive and evolutionary environment as the current economy creates, with sufficient historical data, these well known patterns allow us to anticipate how technology breakthroughs will penetrate the markets and impact the population as a social group.</p><p>Details (trees) remain hidden though. We can’t predict which species (corporations) will be winners or losers; however, the scope and length of the ‘race’, market size and time span can be forecasted with fair accuracy.</p><p>The social aspects of technology adoption, with increasing mobile computing and ubiquitous Internet, are <a href="http://www.asymco.com/2013/11/19/a-way-to-measure-ones-life/">shrinking adoption cycles</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OB9H6331YPNCbl7CWAj90w.png" /></figure><p>The number of new technology breakthroughs is also increasing over time. The intuitive idea of a singular future with unlimited wonders driven by technology makes more sense than ever.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/340/1*BKdtV--2vrkmqvFEahtCzw.png" /></figure><p>This vision has fuelled Sci Fi literature and movies since the 50ies. The concept of Singularity, a future time where technology outwits human capabilities, may be now perceived as stating the obvious, a self-fulfilling prophecy.</p><p>The question is <strong>when</strong>.</p><p>But, not so fast…. First, let’s ‘<em>take a selfie’</em> of the present and look at today’s status quo.</p><p>@<a href="http://www.twitter.com/efernandez">efernandez</a></p><p><strong>Next: An end to Moore’s Law </strong>(<em>Mack 2011</em>)</p><p><a href="https://medium.com/u/809593f1f2ef">ed fernandez</a></p><p>�� ���</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b33a7b147a00" width="1" height="1" alt=""><hr><p><a href="https://medium.com/no-i-wont-fix-your-computer/the-second-arms-race-artificial-intelligence-b33a7b147a00">The Second Arms Race: Artificial Intelligence</a> was originally published in <a href="https://medium.com/no-i-wont-fix-your-computer">No, I won’t fix your computer</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>