<?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 Austin Tanney on Medium]]></title>
        <description><![CDATA[Stories by Austin Tanney on Medium]]></description>
        <link>https://medium.com/@AustinTanney?source=rss-98b2d5eec69d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*vZfCyDuRwwWOYgVzCNIbWw.jpeg</url>
            <title>Stories by Austin Tanney on Medium</title>
            <link>https://medium.com/@AustinTanney?source=rss-98b2d5eec69d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 02:08:55 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@AustinTanney/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[The Sublime to the Ridiculous]]></title>
            <link>https://medium.com/@AustinTanney/the-sublime-to-the-ridiculous-c620de49471a?source=rss-98b2d5eec69d------2</link>
            <guid isPermaLink="false">https://medium.com/p/c620de49471a</guid>
            <category><![CDATA[gans]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Austin Tanney]]></dc:creator>
            <pubDate>Wed, 14 Mar 2018 10:32:06 GMT</pubDate>
            <atom:updated>2018-03-14T10:35:11.485Z</atom:updated>
            <content:encoded><![CDATA[<p>A year ago, something happened that I wasnt expecting.</p><p>A little company called <a href="https://arterys.com/">Arterys</a> over in San Francsisco did something that I truly did not see coming. They managed to get FDA approval for a diagnostic test that was based on deep learning and run in the cloud. At the time (according to Linkedin) they had 20 employees and they managed to do something that I did not expect to see for at least another 5 years. This initial approval was for a <a href="https://medcitynews.com/2017/01/arterys-fda-ai/">Cardiac imaging analysis system</a> and the test was developed on a training set of just 1000 patients and produces a result in 15 seconds.</p><p>The test recieved <a href="https://www.accessdata.fda.gov/cdrh_docs/pdf16/K163253.pdf">510k approval</a> form the FDA. For anyone familiar with the US regulatory system, you will appreciate that this is a pretty big deal. I dont want this post to turn into a 40 page document outlining the US regulatory process, but I will try and give a succinct overview.</p><p>If you want to bring a <strong>new</strong> diagnostic test to the market in the United States, there are two main routes you can follow, Premarket notification (510k) or pre-market approval (PMA). Premarket approval (PMA) is required when you are bringing a device or test to market that has a risk associated, eg risk resulting in misdiagnosis. Obviously in a cardiac test, misdiagnosis can lead to dire consequences so the Arterys test fits this bill. The pre-market notification route (510k) is a lower burden pathway that can be followed when you can prove that your test is “substantially equivalent” to other tests on the market. This is an easier and much cheaper route, but its not an easy get out clause. When the first generations of molecular diagnostic tests hit the market they had to go the PMA route. In many cases it could easily be claimed that these tests were substantially equivalent to current diagnostics tests, however becuase the technology was new, the PMA route had to be followed.</p><p>When we look at the Arterys test, it can for sure be claimed to be equivalent to current approaches.. but wait.. Deep learning is used? A black box algorithm? And the validation dataset is pretty small? If I worked in the FDA Id have assunmed this had to go the PMA route for sure. thankfully, both for me and for Arterys, I dont and they managed to take this totally new technologically approach down the 510k pathway which massively reduced their costs and the time to market. Thats enough of my primer on regulations, if you want to read more you can go to <a href="https://www.fda.gov/MedicalDevices/DeviceRegulationandGuidance/Overview/default.htm">the source</a>.</p><p>So Arterys astonished me by successfully going down this route, developing a diagnostic that seems to be significantly better than the current norms and we have a cloud based diagnostic test based on deep learning available in the clinic. Arterys have grown dramatically. Thanks to Linkedin once more we can see that they have seen 140% growth in the last two years and now have over 60 employees.</p><p>This is the “sublime” part of my title. A company taking an AI based technology and using it to improve healthcare and potentially save lives. Healthcare and diagnostics is the world that I come from and for me this is a fantastic success story for healthcare, for the diagnostics industry and for AI.</p><p>So whats the ridiculous part of this story?</p><p>Well a few months ago, Nvidia <a href="http://research.nvidia.com/sites/default/files/pubs/2017-10_Progressive-Growing-of/karras2018iclr-paper.pdf">published a paper</a> (i<a href="https://arxiv.org/abs/1710.10196">n arXiv</a>) in which they used a generative adversarial network (GAN) to generate images based on photos of celebrities. GANs were pretty new to me at the time so I read the paper and found it interesting, albeit a rather silly use case. Of course, due to the celebrity angle the story got picked up in a <a href="https://www.theverge.com/2017/10/30/16569402/ai-generate-fake-faces-celebs-nvidia-gan">plenty</a> <a href="https://mashable.com/2017/10/30/ai-machine-uses-celebrity-photos-to-create-fake-faces/">of</a> <a href="https://www.geek.com/tech/nvidia-ai-generates-fake-faces-based-on-real-celebs-1721216/">places</a> (<a href="http://www.dailymail.co.uk/sciencetech/article-5035045/Creepy-video-shows-AI-faces-fake-celebrities.html">these</a> <a href="http://www.dailymail.co.uk/news/article-5116605/Program-creates-faces-celebrities-never-existed.html">being</a> <a href="http://uk.pcmag.com/nvidia-shield-tv/91791/news/nvidias-ai-creates-artificial-human-faces-from-celebrity-pho">a</a> <a href="https://petapixel.com/2017/11/07/ai-creates-photo-realistic-faces-people-dont-exist/">very</a> <a href="https://www.engadget.com/2017/10/30/neural-network-nvidia-images-celebs/">select</a> <a href="http://www.alphr.com/artificial-intelligence/1007519/watching-an-ai-create-fake-celebrity-faces-is-nightmare-fuel">few</a> <a href="http://www.alphr.com/artificial-intelligence/1007519/watching-an-ai-create-fake-celebrity-faces-is-nightmare-fuel">found</a> <a href="https://flowingdata.com/2017/10/31/ai-generated-celebrity-faces-look-real/">via</a> <a href="https://www.nytimes.com/interactive/2018/01/02/technology/ai-generated-photos.html">google</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*fx-G1RSfhG09wLCpVkYcTA.png" /><figcaption>None of these people exist</figcaption></figure><p>Now this is, you have to admit, as a use of technology somewhat ridiculous. I guess the authors either just did it for fun OR the marketing team in Nvidia are smart as shit and realised this would get them a tonne of coverage. Cuz, y’know, people love celebrities.</p><p>I personally dont have much interest in the bizarre cult of celebrity worship, but I liked the technology so I persevered in my reading, and I have to say that the results are really genuinely impressive. In the photos above we see a series of people who do not exist. Imaginary humans generated by machines that looks like… well.. actual humans. Anyone interested in the concept of the <a href="https://en.wikipedia.org/wiki/Uncanny_valley">Uncanny Valley</a> has got to take at least some interest in this.</p><p>Outside the results though, the technology behind this is actually really fascinating. I cant speak to the detail much, but I can give some idea of the concepts. If you want the detail, <a href="http://research.nvidia.com/sites/default/files/pubs/2017-10_Progressive-Growing-of/karras2018iclr-paper.pdf">read the paper</a>. The idea of the GAN is that its essentially a pair of neural nets. A generator and a discriminator. So one NN generates an image and the second scores it.</p><p>My first thought around this was that it was like two people working together on a creative project</p><p>“How about we do X”</p><p>“Nah, I dont like it, though there are certain features of that idea, like Y and Z”</p><p>“OK then, how about we take Y and Z and put them in a different context”</p><p>etc…</p><p>But then when I thought more about it, I realised that its actually not far off my own creative process. When I write this paragraph I will initially write something off the top of my head. Then, I’ll go back and reread it and make changes. Or indeed more commonly I will make changes as I see the words appear on the page and realise that the word “realise” works better than “see” in this sentence. (I just tried to use an example of self editing while self editing, it might not work, shhh).</p><p>Anyway, back to the point.</p><p>While this paper initially seemed ridiculous when i read the mainstream coverage, the more I read into it, the more it astonished me. So while the output is ridiculous (but impressive) the potential is huge.</p><p>How about is we turned a good GAN towards generating molecular structures that could be medicines. What impact can they have in the gaming industry? In entertainment? In creating personalised UX?</p><p>Since I started writing this and (and got distracted for several weeks and forgot to finish it) there has been a lot more coverage of GANs. Im not going to go into one example you may have heard of (the pornelephant in the room), though I did read a <a href="https://www.biorxiv.org/content/biorxiv/early/2017/12/30/240317.full.pdf">fascinating paper </a>on the use of a deep nerual network for generation of images basd on functional MRI data. That may be the subject of a future post… Or maybe not, either way, read <a href="https://www.biorxiv.org/content/biorxiv/early/2017/12/30/240317.full.pdf">the paper</a> and imagine the clearly rudimentary results shown here and what they could be if a GAN was applied.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c620de49471a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Our Machine Future]]></title>
            <link>https://medium.com/@AustinTanney/our-machine-future-33fe941f951b?source=rss-98b2d5eec69d------2</link>
            <guid isPermaLink="false">https://medium.com/p/33fe941f951b</guid>
            <category><![CDATA[elon-musk]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[self-driving-cars]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[automation]]></category>
            <dc:creator><![CDATA[Austin Tanney]]></dc:creator>
            <pubDate>Tue, 15 Aug 2017 11:24:49 GMT</pubDate>
            <atom:updated>2017-08-15T11:24:49.487Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PXDtv8PXutDpuqCtH-KHPg.png" /><figcaption>Image by Theo Radomski: Instagram — @flannelogue</figcaption></figure><p>There is no shortage of articles about Artificial Intelligence (AI) and machine Learning (ML) in the mainstream media these days. What has heralded the onslaught of interest in this area? Well, as with so much that hits the media, there is a huge element of sensationalism. In 2015, the Guardian published an article on AI which quoted Elon Musk as stating that humanity risked “S<a href="https://www.theguardian.com/technology/2014/oct/27/elon-musk-artificial-intelligence-ai-biggest-existential-threat">ummoning a Demon</a>”.This article also refers back to a previous article from 2014 in which Musk stated that he invested in Google’s DeepMind over “<a href="https://www.theguardian.com/technology/2014/jun/18/elon-musk-deepmind-ai-tesla-motors">Terminator Fears</a>”.</p><p>Let me dwell for one moment on that Terminator idea.</p><p>I gave a talk on machine learning a while back in which I, jokingly, had a picture of Arnold Schwarzenegger as Terminator. If I remember rightly, I had captioned this slide “What AI is <em>not</em>”. Later that day I saw someone else also use a terminator photo in a talk that mentioned AI. Several Ted talks on the subject later made me realise that what I thought an original and amusing concept was already a tired hacky ploy. Note to self… no more terminator references in talks.</p><p>The truth of the matter is that for most of us, our introduction to Artificial Intelligence <em>has</em> in fact been through science fiction. Do you remember the first time you heard of the concept of AI? I actually do. It was probably around 1990.</p><p>I was 15 or so at the time and reading <a href="http://www.williamgibsonbooks.com/">William Gibson</a>’s seminal masterpiece “Neuromancer”. This book influenced me enormously. There is no doubt that Gibson was a visionary, Neuromancer was published in 1984. I was a little too young to read it upon its release but even in 1990 when I did read it, I don’t think I had heard of the Internet. In fact at that time it existed in a nascent form pretty much exclusively in universities. Our real perception of the internet (the world wide web) is attributed to Tim Berners Lee in 1989 so Gibson’s “Cyberspace” predates this by many years. This is, of course, old ground so let me get back to my point.</p><p>I remember reading Neuromancer and discovering the character “Wintermute,” one of the novels two AI’s. I remember being fascinated by Wintermute and, weirdly, actually liking the character a lot. But Wintermute wasn’t a person; it was a machine, machine intelligence. For me, almost 30 years ago, this was an astonishing concept</p><p>As a sideline, the AI concept wasn’t the only one that fascinated me about Neuromancer, in addition to the previously mentioned concept of cyberspace, I’m pretty sure that this was where I first heard about Biotechnology, which is what I studied for my undergrad degree.</p><p>Here we are though, nearly three decades later and while the Internet and biotech are just part of our everyday world, AI has been a slower development.</p><p>Until relatively recently.</p><p>In the last few years, the pace of success in AI development has been precipitously fast. At this years CES, the worlds biggest tech show, Amazon’s Alexa appears to have <a href="https://www.wired.com/2017/01/ces-alexa-in-everything/">dominated the proceedings</a>. It now no longer seems strange that we have machines in our homes that we can talk to. Machines that can answer questions and play music and movies for us. These machines can even order our shopping while adjusting our heating and turning down the lights so we can watch the movie in perfect environmental conditions.</p><p>Of course, Alexa is far from being Wintermute and Amazon is not the only player in town. Microsoft, Apple and Google have had personal assistants like Alexa living in our phones for a while now, but Amazon seems, with Alexa, to have fundamentally changed something. The personal assistant AI is no longer a feature of a phone or an operating system; it’s a standalone product that is being bought (in record numbers) specifically for this purpose. AI has become part of a great number of people’s everyday lives.</p><p>Many will be well aware that this explosion of interest in AI is based on decades of work and research in machine learning. While machine learning has been in common use in many industries for years, its impact on everyday life has taken a little longer. I don’t have time to go into a discussion here on the relationship between machine learning and artificial intelligence, so lets skip past that and if there is sufficient interest I would be happy to discuss or debate that in a future article.</p><p>So with artificial intelligence becoming so common, should we be worried? Is Elon Musk right to be concerned that AI is an existential risk and that we risk “summoning a demon?” Well for a moment lets step past the scaremongering and focus on the reality. Despite voicing his concerns in interviews and having signed the <a href="https://futureoflife.org/ai-open-letter">open letter of concern</a> about AI, along with Stephen Hawking and hundred of others, Musk is leading the way in the pioneering use of machine learning and AI. He recently stated that by the end of 2017, there will be a fully self-driving Tesla capable of bringing a passenger from <a href="http://uk.businessinsider.com/elon-musk-autonomous-tesla-drive-across-country-by-end-of-2017-2016-10">LA to New York</a>.</p><p>Perhaps unexpectedly, this is actually where I want to focus for the remainder of this article. Not on terminator, nor on the existential threat but on the very real near future impact of automated automotion. I think most who have voiced concerns about AI, Elon Musk, Stephen Hawking, Bill Gates, Sam Harris, et al, are particularly concerned about the next wave of AI, artificial <em>general </em>intelligence. This is a topic covered admirably by <a href="http://www.nickbostrom.com/">Nick Bostrom</a> in his fantastic book “Superintelligence”. Anyone interested in this topic should read this book and form their own opinions on the existential threat. I’m pretty sure it was Bostrom’s work that fuelled the fire of concern in the recent Ted talk on the subject by <a href="https://www.ted.com/talks/sam_harris_can_we_build_ai_without_losing_control_over_it">Sam Harris</a>. Bostrom himself also gave an excellent <a href="https://www.ted.com/talks/nick_bostrom_what_happens_when_our_computers_get_smarter_than_we_are">Ted talk</a> on the subject.</p><p>The truth is that even in the absence of artificial general intelligence, there are very real and growing concerns with the (narrow) AI of today. Lets get back to self-driving cars.</p><p>Last year Lyft made a claim that within 5 years its entire fleet will be <a href="https://www.bloomberg.com/news/articles/2016-09-18/exec-most-lyft-rides-will-be-in-autonomous-cars-in-5-years">self-driving cars</a>. In this same article Lyft’s president and co-founder John Zimmer also stated that he believed that in time personal car ownership will end completely. Many have claimed this to be aspirational and even delusional, however in December Lyft’s competitor Uber launched a pilot scheme in <a href="http://www.theverge.com/2016/12/14/13921514/uber-self-driving-car-san-francisco-launch-volvo-xc90">San Francisco </a>of self-driving cars available for routine use. This was a short-lived endeavour as <a href="https://www.theguardian.com/technology/2016/dec/21/uber-cancels-self-driving-car-trial-san-francisco-california">the California DMV shut it down <em>fast</em></a>. But for a company like Uber, whose margins will clearly be dramatically improved if they no longer have to pay drivers, this setback will not halt their progress. In fact the setback seems to have been very short lived and the self driving fleet has now been move to <a href="https://www.bloomberg.com/news/articles/2016-12-22/uber-pulls-self-driving-cars-from-california-for-arizona">Arizona</a> and there is also a pilot scheme on-going in <a href="http://www.digitaltrends.com/cars/uber-pittsburgh-robo-taxi-experiment/">Pittsburgh</a>.</p><p>This is cool, right? How many science fiction books have we read or movies have we seen where there were automated taxis? Total Recall jumps instantly to mind (no pun intended) but also Minority Report, Batman, Knight Rider, Logan’s Run and I, Robot are great examples. Progress towards science fiction reality… Where is my hoverboard and time travelling Delorean?</p><p>Let me stifle my inner tech obsessed nerd and try to be more analytical. What will this mean for society? Well lets look to Uber who seem to be the pioneers that will push this tech into our every day lives at breakneck speed. When Uber launched in London in 2012, they had 50 drivers. In 2015, Uber’s CEO stated that he expected to have 42,000 drivers in London in 2016. Globally they now have, apparently, over<a href="https://www.washingtonpost.com/news/dr-gridlock/wp/2015/06/04/uber-turns-5-reaches-1-million-drivers-and-300-cities-worldwide-now-what/?utm_term=.8598f10219b2"> 1 million drivers</a>. Lyft, Uber’s closest comparator has over 300,000 drivers. Now with both of these companies being highly focussed on automating their service, there is a potential for 1.3 million people to be out of a job. With Lyft’s aspiration to be fully driverless in 5 years, this is not a long-term distant future problem that we face. It’s happening now. Uber already have self-driving cars operating in their fleet. I have a few friends in the US who drive for Uber part time to make extra money. One of them recently told me that he isn’t ruling out moving full time, as it seems to be working well for him. In the UK, the regulations on Taxis are somewhat different so they have less part time drivers who do it for extra money. But realistically, if the car is self driven and there is no driver to pay, it will be easy for Uber to reduce their fares to a point where traditional taxi companies will no longer be able to compete as they are now. Here in Belfast, Uber has a footprint, but it doesn’t really offer anything that you can’t get from a local taxi company. Our larger firms here have apps that let you call a cab just like Uber, you can pay via the app and the prices are very equivalent. I’ve asked quite a few taxi drivers here if they feel that Uber is impacting their business and few have expressed any concern. But a driverless car may be able to make a sufficient difference in fare that could change that situation dramatically. At that point, we are no longer concerned about the Uber or Lyft drivers losing their jobs, it now expands to everyone in the UK driving a taxi.</p><p>Of course, it doesn’t stop at Taxis. If we have self-driving cars, why stop there? Last year the UK approved a scheme to test <a href="http://www.bbc.co.uk/news/uk-politics-35737104">self-driving fleet lorries</a>. According to the BBC, <a href="http://www.bbc.co.uk/news/uk-england-tyne-30496851">there were 285,000 HGV drivers in the UK in 2014</a>. Now while there appears to be a driver shortage for HGVs, self-driving vehicles could be the solution to that driver shortage, while in the process putting hundreds of thousands of people out of a job.</p><p>What other jobs are threatened? Well through the combination of AI and robotics, 3 that seem threatened in the short term are Personal Assistants, Technical Support and Factory workers. Richard and Daniel Susskind have stated that it won’t be long before white collar jobs are threated too with even <a href="https://qz.com/521848/robots-are-taking-your-white-collar-jobs-too/">Doctors and Lawyers being in the crosshairs</a>.</p><p>Of course I don’t want to get too bogged down in the potential future of AI, the potential for artificial general intelligence and a complete paradigm shift that may occur. There are plenty of debates and discussions around the long-term problems of AI and besides the existential risk there are lots of jobs <em>can potentially </em>be automated. What I am focussing on here is the genuine and real short-term impact. If we limit this problem to self-driving cars alone, there is a real and genuine problem that can impact people’s jobs and future career prospects.</p><p>So what does this all mean? Well… ultimately this means we need to step sideways from focussing on technology and technology development and focus on the social, political and economic situation. If we approach this analytically we can make some simple statements that I feel are self-evident. We have a real example (professional drivers) where artificial intelligence will threaten people’s jobs and livelihoods. We can project that this threat will expand as technology improves and develops. The impact of this will be an increase in unemployment and an increasingly challenging job market as our population grows, but job opportunities decrease. Obviously the growth of this field also <em>creates </em>jobs, but it seems likely that automation will remove human jobs faster than it will create them. There is also a skills gap as todays Uber driver cannot be tomorrows software engineer overnight.</p><p>Is this a bleak picture? Is there a solution? Well it would not be difficult to write another full article on this. For now though, let me very briefly touch on the concept of a universal basic income. This is the idea that everyone is provided with a basic living wage that is not means tested. This provides a potential solution to unemployment by automation as well as having the potential to stimulate innovation. <a href="https://www.vice.com/en_us/article/something-for-everyone-0000546-v22n1">Nathan Schneider in his excellent Vice article</a> describes this as a “peculiar class of political notions that can warm Leninist and libertarian hearts alike”.</p><p>Due to my excessive verbosity, rather than expand on this in any great detail I highly recommend you read Schneider’s article (or one of the thousands of others on the topic). In reality, can it work? Well there are currently trials of this idea occurring in Finland, The Netherlands and Canada. Just a few months back it was announced that this is being <a href="https://www.theguardian.com/politics/2017/jan/01/universal-basic-income-trials-being-considered-in-scotland">considered in Scotland</a>.</p><p>To conclude, I am fascinated by AI and have been since I first encountered the notion as a teenager, decades ago. The progress in this area is incredible and has the potential to significantly positively impact human society. However as with any new technology there are risks and some of these are clear and present risks.</p><p>It is important that we focus on AI and its impacts. We must look at the challenges and the risks and try to engineer a society that can utilise this technology for the benefit of all. Our politicians need to be educated in the potential of this technology and of the potential challenges so we may put structures and solutions in place before these risks become problems.</p><p>Written by <a href="https://twitter.com/austintanney">Austin Tanney</a></p><p>Head of Life Science and Healthcare at <a href="http://www.analyticsengines.com/">Analytics Engines</a></p><p>Illustration by <a href="https://www.instagram.com/flannelogue/">Theo Radomski</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=33fe941f951b" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>