<?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 Casey Stegman on Medium]]></title>
        <description><![CDATA[Stories by Casey Stegman on Medium]]></description>
        <link>https://medium.com/@cstegman?source=rss-6b16e6e3f4c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*0lEaFbEuE4zING8xvD4bnw.jpeg</url>
            <title>Stories by Casey Stegman on Medium</title>
            <link>https://medium.com/@cstegman?source=rss-6b16e6e3f4c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 05:08:58 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@cstegman/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[What an old German horse has to do with AI security]]></title>
            <link>https://medium.com/@cstegman/the-people-behind-openai-1fc42afc9109?source=rss-6b16e6e3f4c------2</link>
            <guid isPermaLink="false">https://medium.com/p/1fc42afc9109</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[openai]]></category>
            <dc:creator><![CDATA[Casey Stegman]]></dc:creator>
            <pubDate>Thu, 06 Apr 2017 20:13:39 GMT</pubDate>
            <atom:updated>2017-04-06T21:57:42.469Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/700/1*HOn-i0UC_Ds9UT-AABdvEw.jpeg" /><figcaption>Clever Hans</figcaption></figure><p><em>(Here’s snippet from a new longform article that </em><a href="https://medium.com/@bsimoneaux"><em>Brent Simoneaux </em></a><em>and I wrote for Red Hat’s </em><a href="https://www.redhat.com/en/open-source-stories"><em>Open Source Stories</em></a><em>. It’s a behind-the-scenes look at the researchers and engineers at OpenAI and how they’re revolutionizing machine learning.)</em></p><p>Ian Goodfellow focuses on something people care a great deal about, adversarial training, or — to put it another way — AI security.</p><p>“In the past, security has revolved around application-level security, where you try to trick an application into running the wrong instructions,” he explains. “Or network security, where you send messages to a server that can get misinterpreted. Like you send a message to a bank saying, ‘Hey, I’m totally the account owner, let me in,’ and the bank gets fooled into doing it, even though you’re not actually the account owner.”</p><p>But with AI, and specifically machine learning, security is a different animal.</p><p>“With machine learning security, the computer is running all the right code and knows who all the messages are coming from,” he says. “But the machine learning system can still be fooled into doing the wrong thing.”</p><p>Goodfellow equates this with phishing. With standard phishing, the computer isn’t tricked, but the person operating the computer is.</p><p>It’s the same for AI. Its code remains uncorrupted. But it is tricked into doing different tasks than it was trained for.</p><p>We’ve all heard stories about someone’s grandfather getting a Nigerian-prince-scam-style phishing email, promising untold riches in exchange for sending $1,000 or $2,000. The grandfather, of course, ends up losing the money and gets nothing in return.</p><p>Well, it turns out AI is even more vulnerable than someone’s grandfather.</p><p>“Machine learning algorithms are really, really gullible, compared to people,” Goodfellow says.</p><p>To make things worse, AI has the potential to be more powerful than anyone’s grandfather. This is no knock against your or anyone else’s elder patriarch. It’s just that Gramps falling for the Nigerian Prince scam is not as problematic as, say, a machine learning algorithm used for the financial services sector being tricked into helping hackers defraud a major bank or credit card company.</p><p>“If you’re not trying to fool a machine learning algorithm, it does the right thing most of the time,” Goodfellow says. “But if someone who understands how a machine learning algorithm works wanted to try and fool it, that’d be very easy to do.”</p><p>Furthermore, it’s very hard for the person building the algorithm to account for the myriad ways it might be fooled. The builder could maybe account for one specific type of attack, like an AI-focused Nigerian Prince scam. But that doesn’t mean that the AI can’t be tricked into another, even more simplistic con.</p><p>Goodfellow’s research focuses on using adversarial training on AI agents. This approach is a “brute force solution” in which a ton of examples meant to fool an AI are generated. The agent is given these examples and trained not to fall for them.</p><p>For example, you might train the AI used in a self-driving car not to fall for a fake sign telling the AI to halt in the middle of the highway.</p><p>Goodfellow has developed (along with Nicholas Papernot) cleverhans, a library for adversarial training.</p><p>The name comes from a German horse who became famous in the the early 20th century for his ability to do arithmetic.</p><p>A German math teacher (also a self-described mystic and part-time phrenologist) bought the horse and claimed that he had taught it to add, subtract, multiply, divide, and even do fractions. People would come from all over and ask Clever Hans to, for example, divide 15 by 3. The horse would then tap his hoof 5 times. Or people would ask it what number comes after 7. The horse would tap his hoof 8 times.</p><p>The problem was, Clever Hans wasn’t that clever — at least not in the way his teacher thought.</p><p>A psychologist named Oskar Pfungst discovered that the horse wasn’t actually doing math. Rather, he was taking his cues from the people around him. He’d respond to these people’s body language, tapping his hoof until he got a smile or a nod. Pfungst illustrated this by having the horse wear a set of blinders. When asked a question, the horse began tapping his hoof. But, unable to see the person who’d asked the question, he just kept tapping indefinitely.</p><p>“Machine learning is a little like Clever Hans,” Goodfellow says, “in the sense that we’ve given the AI these rewards for, say, correctly labeling images. It knows how to get the rewards, but it may not always be using the correct cues to get to those rewards. And that’s where security researchers come in.”</p><p>Goodfellow’s cleverhans library has been open sourced.</p><p>“With traditional security, open source is important because, when everybody can see the code, they can inspect it and make sure it’s safe,” he says. “And if there’s a problem, they can report it relatively easily or even send the fix themselves.”</p><p>A similar dynamic holds for machine learning security. Generally speaking, that is.</p><p>“For machine learning, there isn’t really a fix yet,” Goodfellow says. “But we can at least study the same systems that everybody is using and see what their vulnerabilities are.”</p><p>When asked if there’s anything that has surprised him about his experiences doing machine learning research, Goodfellow talks about the time he ran an experiment for a machine learning algorithm to correctly classify adversarial examples.</p><p>He had just read a research paper that made some claims he thought were questionable. So he decided to test them. While his experiment was running, Goodfellow decided to step out to grab some lunch with his manager.</p><p>“I told him,” Goodfellow recalls, “‘when we get back from lunch, I’m not sure the algorithm’s going to correctly classify these examples. I bet it will be too hard. And, even after this training, it will still misclassify them.’”</p><p>But when he came back, Goodfellow found that the algorithm not only recognized the adversarial examples, it had also set a record for accuracy in classifying the normal ones.</p><p>“The process of training on the adversarial examples had forced it to get so good at its original task that it was a better model than what we had started with,” Goodfellow says.</p><p>At that moment, Goodfellow realized that, for AI, adversarial training wasn’t just important for finding vulnerabilities.</p><p>“By thinking about security,” he says, “we could actually make everything better across the board.”</p><p><em>(To read the full article, head over </em><a href="https://www.redhat.com/en/open-source-stories/ai-revolutionaries/people-behind-openai"><em>here</em></a><em>.)</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1fc42afc9109" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What Plato has taught me about artificial intelligence]]></title>
            <link>https://medium.com/@cstegman/what-plato-has-taught-me-about-artificial-intelligence-3a17f31a5805?source=rss-6b16e6e3f4c------2</link>
            <guid isPermaLink="false">https://medium.com/p/3a17f31a5805</guid>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[philosophy]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[machine-learning]]></category>
            <dc:creator><![CDATA[Casey Stegman]]></dc:creator>
            <pubDate>Tue, 31 Jan 2017 22:18:10 GMT</pubDate>
            <atom:updated>2017-01-31T22:23:22.513Z</atom:updated>
            <cc:license>https://creativecommons.org/licenses/by-nc-nd/4.0/</cc:license>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/639/1*0bUEP8Esj7-UXseAaE2MKQ.jpeg" /><figcaption>“Philosoputer” ©2011 <a href="http://braydon.com/">Braydon Fuller</a> <a href="https://creativecommons.org/licenses/by-sa/2.0/">(CC BY-SA 2.0)</a></figcaption></figure><p><em>This is part of a series on my quest to learn as much as possible about AI. To know why I’m doing this, check out </em><a href="https://medium.com/@cstegman/a-writers-quest-begins-ai-open-source-and-a-ton-of-unanswered-questions-6eec600bb263#.kuj5id89s"><em>my first post</em></a><em>.</em></p><p>In grad school, I primarily studied Ancient Greek philosophy. Mostly Plato. My only published piece of academic research is on <a href="http://ptx.sagepub.com/content/early/2015/07/02/0090591715594661.abstract">his Atlantis myth</a>.</p><p>But now I work at a <a href="http://www.redhat.com/">tech company</a> and I’m writing about artificial intelligence — a subject I’m very interested in but know little about.</p><p>Still, knowing what I do about Plato — and what Plato wrote about his famous teacher, Socrates — turns out to be the one advantage I have going for me.</p><p>Let me explain.</p><h3>AI’s knowledge problem</h3><p>There’s plenty of hype surrounding the benefits of AI. A recent <a href="https://ai100.stanford.edu/sites/default/files/ai_100_report_0831fnl.pdf">Stanford study</a> aimed to quell fears about potential Skynets and <em>Matrix</em>-like dystopias. And, instead, highlighted the many advancements AI will make in areas like transportation, healthcare, education, public safety and security.</p><p>But even though evil machines are unlikely, that doesn’t mean there isn’t cause for concern as AI progresses.</p><p>It turns out, with all of AI’s promised intelligence, a machine’s actual knowledge is and could continue to be a very real problem.</p><p>What’s the difference between <em>intelligence</em> and <em>knowledge</em>, you ask?</p><p>This is where Plato comes in.</p><p>According to Plato it’s possible to be intelligent but have little-to-no knowledge about the world or, more importantly, yourself.</p><p>And there’s nothing wrong with this.</p><p>In fact, as I’ll talk about in a minute, to <em>know</em> that you are not knowledgeable was seen by Plato as being a marker of an extremely advanced intellect. This is why he devoted much of his early writings to his famous teacher, Socrates.</p><p>(Side note: Socrates did not produce any writings. So much of what we understand about him comes to us from his student, Plato, who often cast Socrates as the main character in his dialogues. Hence, this is why I will be talking about both Plato and Socrates.)</p><p>For most people, though, intelligence is assumed to be proof that they are extremely knowledgeable. The problem is that lots of other assumptions about the world follow this major one.</p><p>And, what’s worse, when you start questioning these assumptions, you see that a lot of them have almost no basis in reality.</p><p>That’s why most of the people in Athens hated Socrates.</p><p>He liked to stop his fellow citizens in the street or at parties, and ask them a lot of annoying questions about the nature and meaning of certain ideas, like justice or goodness. In a relatively short period of time, Socrates’s questions revealed that these people — who thought they <em>knew</em> what these concepts meant — had never actually thought about them and, as such, didn’t really know anything about them. Their knowledge of these important concepts were filled with all kinds of unquestioned assumptions.</p><p>AI has a similar sort of problem, primarily because machines are taught by human beings, many of whom — like the ancient Athenians and most of us today (including yours truly)— have unquestioned assumptions about the world.</p><p>And here’s how these sorts of unquestioned assumptions can lead to unintended problems.</p><p>Motherboard <a href="http://motherboard.vice.com/read/why-an-ai-judged-beauty-contest-picked-nearly-all-white-winners">reported</a> that, a few months back, an AI called “Beauty.ai” — developed by researchers in Russia and Hong Kong (and backed by Microsoft and Nvidia) — ran an online beauty pageant where 600,000 men and women from around the world sent in selfies. The AI judged the entrants based on facial symmetry, wrinkles, and age. It then picked 44 winners that it deemed “most attractive.” Almost all of them were white.</p><p>ProPublica ran <a href="https://www.propublica.org/article/machine-bias-risk-assessments-in-criminal-sentencing">a similar story</a> in May of last year. This time it was about bias in a risk assessment software that judges use to make sentencing decisions for criminal cases. African American defendants were 77% more likely to be identified as being a “higher risk” for future offenses than white defendants. This score was based on 137 questions given to the defendants. And, while none of the questions mentioned race, the end result was a racial bias in favor of white defendants.</p><p>These recent news stories point to a painstakingly obvious fact about AI. As<a href="https://psmag.com/artificial-intelligence-will-be-as-biased-and-prejudiced-as-its-human-creators-38fe415f86dd#.4togk6up9"> Nathan Collins</a> writes, “Just as we learn our biases from the world around us, AI will learn its biases from us.”</p><p><a href="https://www.bloomberg.com/news/articles/2016-06-23/artificial-intelligence-has-a-sea-of-dudes-problem">Bloomberg</a> reported this past June that the majority of AI researchers are men and highlighted two sets of statistics. The first was that only 17% of computer science graduates today are women. The second was that, at 2015’s premier AI conference, only 13.7% of the attendees were women. Margaret Mitchell, a Microsoft researcher, is quoted as saying that AI has a “sea of dudes” and it’s a problem. As the article notes: “If everyone teaching computers to act like humans are men, then the machines will have a view of the world that’s narrow by default and, through the curation of data sets, possibly biased.”</p><p>How can we avoid this problem going forward?</p><p>Well, again, this is where Plato comes back in.</p><h3>Why Plato wrote that Socrates was the wisest man in Greece</h3><p>Plato tells us that Socrates’s annoying questions reached their tipping point in 399 BCE when he was tried and then <em>(spoiler alert</em>) subsequently sentenced to death. Although the official charge against him was “corrupting the youth of Athens,” most everyone, including Plato, knew that it was because Socrates had made almost everyone he encountered feel like an idiot.</p><p>In the trial, Socrates decided against appeasing his critics, and instead took a different approach.</p><p>He declared himself the wisest man in all of Greece.</p><p>But this label, he told his fellow Athenians, was not self-applied. It was given to him by the Oracle at Delphi — who was a <em>big</em> deal in Ancient Greece.</p><p>For a long time, Socrates said he didn’t want to believe that he was the wisest man in Greece. He couldn’t. Although he loved to question others about complex topics, Socrates constantly admitted that he, himself, knew nothing.</p><p>That’s when the light bulb (or the torch if you want to be historically accurate) went off in his head.</p><p>He was the wisest man in Greece because he <em>knew</em> that he knew nothing.</p><p>For Plato, Socrates’s acknowledgment of his own limited knowledge was game-changing. It was proof that his intellect was far beyond his contemporaries — primarily because he had what they didn’t have: Perspective.</p><p>But how does AI get to this same place?</p><p>It starts with the researchers themselves and how they access and distribute knowledge.</p><h3>Openness and transparency as a way forward</h3><p>In December 2015, Elon Musk made headlines with his decision to join other high-profile investors in backing <a href="https://openai.com/blog/">OpenAI</a> — a non-profit research initiative focused on conducting AI within the philosophy of open source. Musk’s huge investment was characterized by much of the news media as <a href="http://nerdist.com/elon-musk-announces-openai-a-non-profit-to-help-prevent-skynet/">his effort to prevent Skynet</a>.</p><p>But as Greg Brockman, OpenAI’s co-founder and CTO points out, OpenAI is about much more than preventing villainous machines, <a href="https://blog.gregbrockman.com/my-path-to-openai">it’s about creating an AI that benefits humanity.</a></p><p>And the way to do that is by doing it in the open for others to see and comment on.</p><p>In a way — and not to stretch the analogy too far — it’s AI researchers opening themselves up to a potential Socrates. Some other researcher or group who can look at what they did and find the unquestioned biases that slipped through the cracks.</p><p>Judging from the <a href="http://www.techradar.com/news/world-of-tech/how-will-open-source-ai-change-the-tech-industry--1322242">recent wave</a> of open source AI projects — both from big proprietary corporations and small start-ups — it seems that the majority of research projects out there are warming to this approach.</p><p>On my end, I still have a long way to go toward understanding what’s going on in these research projects. Still, Plato’s discussion of Socrates has helped me get there a little bit easier.</p><p>While I don’t know what AI researchers know, I know some of what they don’t know.</p><p>And that helps. At least a little.</p><p><em>Note: Please drop your knowledge bombs on me in the comments below. Tell me what I’m getting right. And, more importantly, tell me what I’m getting wrong — but please also tell me why. Thank you.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3a17f31a5805" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A writer’s quest begins: AI, open source, and a ton of unanswered questions]]></title>
            <link>https://medium.com/@cstegman/a-writers-quest-begins-ai-open-source-and-a-ton-of-unanswered-questions-6eec600bb263?source=rss-6b16e6e3f4c------2</link>
            <guid isPermaLink="false">https://medium.com/p/6eec600bb263</guid>
            <category><![CDATA[b2b-content-marketing]]></category>
            <category><![CDATA[content-marketing]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[open-source]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Casey Stegman]]></dc:creator>
            <pubDate>Thu, 15 Dec 2016 20:50:15 GMT</pubDate>
            <atom:updated>2016-12-15T20:50:15.945Z</atom:updated>
            <cc:license>https://creativecommons.org/licenses/by-nc-sa/4.0/</cc:license>
            <content:encoded><![CDATA[<p><em>What’s it like to devote your life to teaching machines how to think?</em></p><p>Over the last several months, I’ve been obsessed with this question.</p><p>Before I brew my morning coffee, it’s there. When I eat my lunch, it’s there. Snack time at my desk, it’s there. Dinner with my wife and many pets (3 dogs and 4 cats), it’s there. In bed, trying to fall asleep, it’s there.</p><p>It doesn’t go away.</p><p>So, I guess it makes sense that I’m devoting the next several months of my life to researching and writing about it with my colleague, <a href="https://medium.com/@bsimoneaux">Brent Simoneaux</a>.</p><p>And that’s why I am starting this series of posts. It’s going to document my research, my findings, my questions, my attempts at answers, and my occasional ineptitude in deciphering all the technical aspects that go into artificial intelligence research.</p><p>A little bit about me: I’m a writer and editor at <a href="http://redhat.com/">Red Hat</a>.</p><p>Red Hat is the premier provider of open source software solutions for enterprises.</p><p>Some of you may wonder why I’m researching and writing about AI when I work at an enterprise tech company.</p><p>Well, it’s because Red Hat cares about open source, and it wants to tell the world about all the cool stuff that’s happening in the world because of open source.</p><p>And, right now, one of the coolest things happening is <a href="https://medium.com/software-is-eating-the-world/what-s-next-in-computing-e54b870b80cc#.ogzgvfpzr">open source AI research</a>.</p><p>A few years back, Red Hat launched a film series on a variety of topics happening in the world of open source, titled <a href="https://www.redhat.com/en/open-source-stories">Open Source Stories</a>. These films have received a huge amount of attention, not only in viewership and <a href="https://www.redhat.com/en/open-source-stories">news coverage</a>, but they’ve also won some <a href="https://www.redhat.com/en/about/blog/open-source-stories-film-wins-big-west-chester-film-festival">awards</a>.</p><p>Open Source Stories is now moving beyond film and into the realm of longform narrative journalism (for more on this, check out Brent’s latest <a href="https://medium.com/@bsimoneaux/make-brand-journalism-slow-4cb53167f264#.3uqwz06oe">post</a>).</p><p>Brent and I are developing a limited series of articles under the Open Source Stories umbrella on open source AI research that will be released in April of 2017.</p><p>This means that we have only a few months to do as much research as we can, interview as many folks as possible, and write a ton of compelling words.</p><p>It’s going to be fun. It’s going to be stressful. And it’s going to be educational.</p><p>Which is why I am also doing these posts. Every other week, I’ll discuss this research journey. And, hopefully, I’ll get comments from you all out there on what I’m getting right and what I’m way off base on.</p><p>Because I need your help.</p><p>Also, for those neophytes like myself who are interested in AI, this will also serve as a resource repository.</p><p>I’m looking forward to posting my findings and hearing what you all think.</p><p>Now, let the epic quest begin…</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=6eec600bb263" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>