AI radiology has arrived

This has been predicted for a long time, but AI radiology is here:

A commercial artificial intelligence (AI) system matched the accuracy of over 28,000 interpretations of breast cancer screening mammograms by 101 radiologists. Although the most accurate mammographers outperformed the AI system, it achieved a higher performance than the majority of radiologists.

Artificial intelligence versus 101 radiologists

Almost anything to do with recognizing objects or features in images are going to be the first tasks mastered by convolutional neural networks. Radiology, surveillance, counting stuff, etc.

AI beats esports world champion team for first time

Some humans have gotten very good at playing the video game Dota 2. It’s a complex game with over 100 different character types, an in-game economy, and an audience of spectators on Twitch.tv. Oh, and tournaments in which professional Dota 2 players have earned over $100M. And now the championship team has been crushed by an AI:

Within the simplified bounds of the game, OpenAI Five was an astounding triumph. One thing to look for in evaluating the performance of an AI system on a strategy game is whether it’s merely winning with what gamers call “micro” — the second-to-second positioning and attack skills where a computer’s reflexes are a huge advantage. 

OpenAI Five did have good micro, but it also did well in ways that human players, now that they’ve seen it, may well choose to emulate — suggesting that it didn’t just succeed through superior reflexes. The commentators watching the game criticized OpenAI Five’s eagerness to buy back into the game when its heroes died, for example, but the tactic was borne out — maybe suggesting that human pros should be a bit more willing to pay to rejoin the field. 

And OpenAI had a deeper strategic understanding of the board than the human commentators. When the commentators were asserting that the game looked evenly matched, OpenAI would declare that it perceived a 90 percent chance of victory. (It turns out that soberly announced probability estimates make for great trash talk, and these declarations frequently rattled their opponents OG). To us, the game may have seemed open, but to the computer, it was obviously nearly over.

AI triumphs against the world’s top pro team in strategy game Dota 2

Three points to note here:

  1. Rate of improvement. AI’s are improving at an astonishing rate. Chess fell, then Go, now very complex multi-player strategic games like Dota 2. It used to be that game-playing algorithms were customized for specific games and had little applicability to other domains. This is truly a revolution.
  2. Scale of computation. The scale of computation available to the AI’s matters a lot. OpenAI, the researchers behind this AI victory, improved on their previous performance by utilizing eight times more training compute. They trained this model on 45,000 years of Dota self-play over 10 realtime months. Good luck humans.
  3. Real-world applications. Dota 2 is a very complex game with many characters making independent real-time judgments as part of teams trying to take over each other’s bases while protecting their own. It’s a complex simulation of war. The real world is of course still more complex, but this is a domain in which the AI’s appear to do well. Defense departments around the world are paying attention.

Update: The OpenAI team let their AI play against regular Dota 2 players. Out of 7,257 matches, the AI’s won 7,215 (99.4%) and lost just 42.

U.S. facial recognition also rolling out

Jon Porter, writing for The Verge:

The Department of Homeland Security says it expects to use facial recognition technology on 97 percent of departing passengers within the next four years. The system, which involves photographing passengers before they board their flight, first started rolling out in 2017, and was operational in 15 US airports as of the end of 2018. 

The facial recognition system works by photographing passengers at their departure gate. It then cross-references this photograph against a library populated with facesimages from visa and passport applications, as well as those taken by border agents when foreigners enter the country.

US facial recognition will cover 97 percent of departing airline passengers within four years

It’s not automated racism, but it’s similar in scope to China’s rollout. Routine facial recognition for tracking is here, like it or not.

Automated Racism in China

Paul Mozur, writing for the New York Times:

Now, documents and interviews show that the authorities are also using a vast, secret system of advanced facial recognition technology to track and control the Uighurs, a largely Muslim minority. It is the first known example of a government intentionally using artificial intelligence for racial profiling, experts said.

The facial recognition technology, which is integrated into China’s rapidly expanding networks of surveillance cameras, looks exclusively for Uighurs based on their appearance and keeps records of their comings and goings for search and review. The practice makes China a pioneer in applying next-generation technology to watch its people, potentially ushering in a new era of automated racism.

One Month, 500,000 Face Scans: How China Is Using A.I. to Profile a Minority

Bill Gates recently said that AI is the new nuclear technology: both promising and dangerous.

Our long term survival probably requires being good at managing the dangers of increasingly powerful technologies. Not a great start.

AI Transparency Tension: NYPD Sex Chat Bots

The NYPD is using AI chat bots to surface and warn individuals looking to buy sex:

A man texts an online ad for sex.

He gets a text back: “Hi Papi. Would u like to go on a date?” There’s a conversation: what he’d like the woman to do, when and where to meet, how much he will pay.

After a few minutes, the texts stop. It’s not unexpected — women offering commercial sex usually text with several potential buyers at once. So the man, too, usually texts several women at once.

What he doesn’t expect is this: He is texting not with a woman but with a computer program, a piece of artificial intelligence that has taught itself to talk like a sex worker.

A.I. Joins the Campaign Against Sex Trafficking

The article posts an example of an actual chat conversation and it is worth reading to get a sense of the AI capabilities.

Ethics tension. It’s worth noting that many AI ethics frameworks emphasize the importance of informing humans when they are interacting with bots. See also the Google Duplex controversy. Instead, this is indeed deception-by-design. How does this fall within an ethical framework? Are we immediately making trade-offs between effectiveness and transparency?

European Commission publishes “framework for achieving Trustworthy AI”

Like many recent frameworks, this “High-Level Expert Group” assessment provides a list of fairly vague but nevertheless laudatory principles that AI developers should respect:

Trustworthy AI has three components, which should be met throughout the system’s entire life cycle:

1. it should be lawful, complying with all applicable laws and regulations;

2. it should be ethical, ensuring adherence to ethical principles and values; and

3. it should be robust, both from a technical and social perspective, since, even with good intentions, AI systems can cause unintentional harm.

Ethics Guidelines for Trustworthy AI (via Commission reports website)

Great: lawful, ethical, and robust. Ok, how do we do that? Well, the report also lays out four ethical principles to help achieve Trustworthy AI:

  • Respect for human autonomy
  • Prevention of harm
  • Fairness
  • Explicability

Ok, great: lawful, ethical, and robust. And ethical means respect human autonomy, prevent harm, be fair, explain what the AI is doing. Got it. No wait, there’s seven more (non-exhaustive) requirements for Trustworthy AI:

  • Human agency and oversight
  • Technical robustness and safety (robustness duplicate!)
  • Privacy and data governance (lawfulness duplicate!)
  • Transparency (explicability duplicate!)
  • Diversity, non-discrimination and fairness (fairness duplicate!)
  • Societal and environmental wellbeing (prevention of harm duplicate?)
  • Accountability

Ok, nail all these and we’re good? No, no, the report also recognizes that, “Tensions may arise between the above principles, for which there is no fixed solution.” For example, “trade-offs might have to be made between enhancing a system’s explainability (which may reduce its accuracy) or increasing its accuracy (at the cost of explainability).” And what should we do if tensions arise? “[S]ociety should endeavour to align them.”

Clear as mud. Of course, to be fair, no one else is doing any better.

AI researchers call on Amazon to stop selling facial recognition technology to law enforcement

A group of 27 AI researchers affiliated with various academic institutions as well as Microsoft, Google, and Facebook have written an open letter calling on Amazon to stop selling its face recognition technology (Rekognition) to law enforcement. The letter gets into the weeds very quickly but the main complaint is that Rekognition is biased against darker skinned individuals:

A recent study conducted by Inioluwa Deborah Raji and Joy Buolamwini, published at the AAAI/ACM conference on Artificial Intelligence, Ethics, and Society, found that the version of Amazon’s Rekognition tool which was available on August 2018, has much higher error rates while classifying the gender of darker skinned women than lighter skinned men (31% vs. 0%).

On Recent Research Auditing Commercial Facial Analysis Technology

Amazon’s response has essentially been “no that’s not quite right and we’re also concerned and continually improving but none of this is any reason to stop selling the product.”

What all of this highlights is:

  1. No consensus on amount of tolerable bias. Perfectly zero bias may be unreachable. Do we insist upon it, or near it? Or is there a level of tolerable bias? Less bias than an average human would be an improvement in most cases.
  2. No framework for assessing bias. We don’t have any standards on how to judge whether an AI system is “tolerably biased” or not. Much of the debate here is over how the biased was measured.
  3. No framework for assessing impact of bias. Objections to Amazon’s Rekognition technology are premised on its commercial sale, especially to law enforcement. If Amazon had simply released the technology as a research project, it would have joined many other examples of bias in AI research that cause concern but not outrage. Should we insist on zero bias for law enforcement applications? Can retail applications be more tolerably biased?
  4. No laws or regulations at all. And of course there are no laws or regulations governing the sale or use of these systems anywhere in the United States. But… perhaps coming soon.

Facebook and Housing Discrimination

The Department of Housing and Urban Development sued Facebook for housing discrimination. The allegations are fascinating and, although we mostly knew all of this before (based on reporting by Pro Publica), I think most people do not realize how impressively targeted advertisements can be on Facebook. For example:

Respondent [Facebook] has provided a toggle button that enables advertisers to exclude men or women from seeing an ad, a search-box to exclude people who do not speak a specific language from seeing an ad, and a map tool to exclude people who live in a specified area from seeing an ad by drawing a red line around that area. Respondent also provides drop-down menus and search boxes to exclude or include (i.e., limit the audience of an ad exclusively to) people who share specified attributes. Respondent has offered advertisers hundreds of thousands of attributes from which to choose, for example to exclude “women in the workforce,” “moms of grade school kids,” “foreigners,” “Puerto Rico Islanders,” or people interested in “parenting,” “accessibility,” “service animal,” “Hijab Fashion,” or “Hispanic Culture.” Respondent also has offered advertisers the ability to limit the audience of an ad by selecting to include only those classified as, for example, “Christian” or “Childfree.”

Complaint at paragraph 14.

But Facebook’s system doesn’t just enable this kind of micro-targeting. It also refuses to show ads to users that its system judges as unlikely to interact with the ads, even if the advertisers want to target those users:

Even if an advertiser tries to target an audience that broadly spans protected class groups, Respondent’s ad delivery system will not show the ad to a diverse audience if the system considers users with particular characteristics most likely to engage with the ad. If the advertiser tries to avoid this problem by specifically targeting an unrepresented group, the ad delivery system will still not deliver the ad to those users, and it may not deliver the ad at all.

Complaint at paragraph 19.

Thus, the allegation is that the system functions “just like an advertiser who intentionally targets or excludes users based on their protected class.”

There is an AI angle to this as well. The complaint specifically references Facebook’s “machine learning and other prediction techniques” as enabling this kind of targeting. And while folks may disagree on whether this is “AI” or just sophisticated statistical analysis, it is a concrete allegation of real-world harm caused by big data and computation. And I think it is an interesting case study in whether we need extra laws to prevent AI harm.

Here is a hypothesis: our existing laws prohibiting various types of harm will work just fine or better in the AI context. Housing discrimination is already illegal, whether you do it subjectively and intentionally or objectively by sophisticated computation. And in fact, it’s easier to prove the latter. The AI takes input and outputs a result. That result is objective and (with the help of legal process) transparent. The AI doesn’t rationalize its decisions or try to explain away its hidden bias because it fears social judgment. If it operates in a biased manner, we will see it and we can fix it.

There is a lot of anxiety around whether our laws are sufficient for the AI future we envision. Will product liability laws be sufficient to determine who is at fault when a self-driving vehicle crashes? Will anti-discrimination laws be sufficient to disincentivize AI-facilitated bias? Yes, yes I think they will. Perhaps the law is more robust than we fear.

US Government Tries to Address AI

Recently there’s been a push by the U.S. government to figure out this AI thing. After all, China has a big long-term plan. We should, right?

So President Trump issued an executive order in February, and the White House put together this glossy website to talk about AI initiatives.

It’s all just noise. Here’s what the executive order says:

  • We should continue to lead in AI by (a) leading; (b) developing standards; (c) training; (d) fostering public trust; and (e) promoting international cooperation.
  • All departments should pursue these objectives: (a) invest in AI; (b) invest in data; (c) reduce barriers to using AI (but not so much that it impacts safety etc.); (d) develop secure standards; (e) train people; and (f) develop an action plan!
  • The National Science and Technology Council Select Committee on Artificial Intelligence should coordinate all this.
  • AI R&D is a funding priority, depending on your mission of course.
  • Publish a bunch of stuff in the Federal Register asking for public comments, and consider these goals within 120-180 days of the order.

Bottom line: someone should really start thinking about this stuff and maybe we should spend some money on it? There is zero vision in any of this.