Podcast: The story of AI, as told by the people who invented it
Welcome to I Was There When, a new oral history project from the In Machines We Trust podcast. It features stories of how breakthroughs in artificial intelligence and computing happened, as told by the people who witnessed them. In this first episode, we meet Joseph Atick— who helped create the first commercially viable face recognition system.
This episode was produced by Jennifer Strong, Anthony Green and Emma Cillekens with help from Lindsay Muscato. It’s edited by Michael Reilly and Mat Honan. It’s mixed by Garret Lang, with sound design and music by Jacob Gorski.
Jennifer: I’m Jennifer Strong, host of In Machines We Trust.
I want to tell you about something we’ve been working on for a little while behind the scenes here.
It’s called I Was There When.
It’s an oral history project featuring the stories of how breakthroughs in artificial intelligence and computing happened… as told by the people who witnessed them.
Joseph Atick: And as I entered the room, it spotted my face, extracted it from the background and it pronounced: “I see Joseph” and that was the moment where the hair on the back… I felt like something had happened. We were a witness.
Jennifer: We’re kicking things off with a man who helped create the first facial recognition system that was commercially viable… back in the ‘90s…
I am Joseph Atick. Today, I’m the executive chairman of ID for Africa, a humanitarian organization that focuses on giving people in Africa a digital identity so they can access services and exercise their rights. But I have not always been in the humanitarian field. After I received my PhD in mathematics, together with my collaborators made some fundamental breakthroughs, which led to the first commercially viable face recognition. That’s why people refer to me as a founding father of face recognition and the biometric industry. The algorithm for how a human brain would recognize familiar faces became clear while we were doing research, mathematical research, while I was at the Institute for Advanced Study in Princeton. But it was far from having an idea of how you would implement such a thing.
It was a long period of months of programming and failure and programming and failure. And one night, early morning, actually, we had just finalized a version of the algorithm. We submitted the source code for compilation in order to get a run code. And we stepped out, I stepped out to go to the washroom. And then when I stepped back into the room and the source code had been compiled by the machine and had returned. And usually after you compile it runs it automatically, and as I entered the room, it spotted a human moving into the room and it spotted my face, extracted it from the background and it pronounced: “I see Joseph.” and that was the moment where the hair on the back—I felt like something had happened. We were a witness. And I started to call on the other people who were still in the lab and each one of them they would come into the room.
And it would say, “I see Norman. I would see Paul, I would see Joseph.” And we would sort of take turns running around the room just to see how many it can spot in the room. It was, it was a moment of truth where I would say several years of work finally led to a breakthrough, even though theoretically, there wasn’t any additional breakthrough required. Just the fact that we figured out how to implement it and finally saw that capability in action was very, very rewarding and satisfying. We had developed a team which is more of a development team, not a research team, which was focused on putting all of those capabilities into a PC platform. And that was the birth, really the birth of commercial face recognition, I would put it, on 1994.
My concern started very quickly. I saw a future where there was no place to hide with the proliferation of cameras everywhere and the commoditization of computers and the processing abilities of computers becoming better and better. And so in 1998, I lobbied the industry and I said, we need to put together principles for responsible use. And I felt good for a while, because I felt we have gotten it right. I felt we’ve put in place a responsible use code to be followed by whatever is the implementation. However, that code did not live the test of time. And the reason behind it is we did not anticipate the emergence of social media. Basically, at the time when we established the code in 1998, we said the most important element in a face recognition system was the tagged database of known people. We said, if I’m not in the database, the system will be blind.
And it was difficult to build the database. At most we could build thousand 10,000, 15,000, 20,000 because each image had to be scanned and had to be entered by hand—the world that we live in today, we are now in a regime where we have allowed the beast out of the bag by feeding it billions of faces and helping it by tagging ourselves. Um, we are now in a world where any hope of controlling and requiring everybody to be responsible in their use of face recognition is difficult. And at the same time, there is no shortage of known faces on the internet because you can just scrape, as has happened recently by some companies. And so I began to panic in 2011, and I wrote an op-ed article saying it is time to press the panic button because the world is heading in a direction where face recognition is going to be omnipresent and faces are going to be everywhere available in databases.
And at the time people said I was an alarmist, but today they’re realizing that it’s exactly what’s happening today. And so where do we go from here? I’ve been lobbying for legislation. I’ve been lobbying for legal frameworks that make it a liability for you to use somebody’s face without their consent. And so it’s no longer a technological issue. We cannot contain this powerful technology through technological means. There has to be some sort of legal frameworks. We cannot allow the technology to go too much ahead of us. Ahead of our values, ahead of what we think is acceptable.
The issue of consent continues to be one of the most difficult and challenging matters when it deals with technology, just giving somebody notice does not mean that it’s enough. To me consent has to be informed. They have to understand the consequences of what it means. And not just to say, well, we put a sign up and this was enough. We told people, and if they did not want to, they could have gone anywhere.
And I also find that there is, it is so easy to get seduced by flashy technological features that might give us a short-term advantage in our lives. And then down the line, we recognize that we’ve given up something that was too precious. And by that point in time, we have desensitized the population and we get to a point where we cannot pull back. That’s what I’m worried about. I’m worried about the fact that face recognition through the work of Facebook and Apple and others. I’m not saying all of it is illegitimate. A lot of it is legitimate.
We’ve arrived at a point where the general public may have become blasé and may become desensitized because they see it everywhere. And maybe in 20 years, you step out of your house. You will no longer have the expectation that you wouldn’t be not. It will not be recognized by dozens of people you cross along the way. I think at that point in time that the public will be very alarmed because the media will start reporting on cases where people were stalked. People were targeted, people were even selected based on their net worth in the street and kidnapped. I think that’s a lot of responsibility on our hands.
And so I think the question of consent will continue to haunt the industry. And until that question is going to be a result, maybe it won’t be resolved. I think we need to establish limitations on what can be done with this technology.
My career also has taught me that being ahead too much is not a good thing because face recognition, as we know it today, was actually invented in 1994. But most people think that it was invented by Facebook and the machine learning algorithms, which are now proliferating all over the world. I basically, at some point in time, I had to step down as being a public CEO because I was curtailing the use of technology that my company was going to be promoting because the fear of negative consequences to humanity. So I feel scientists need to have the courage to project into the future and see the consequences of their work. I’m not saying they should stop making breakthroughs. No, you should go full force, make more breakthroughs, but we should also be honest with ourselves and basically alert the world and the policymakers that this breakthrough has pluses and has minuses. And therefore, in using this technology, we need some sort of guidance and frameworks to make sure it’s channeled for a positive application and not negative.
Jennifer: I Was There When… is an oral history project featuring the stories of people who have witnessed or created breakthroughs in artificial intelligence and computing.
Do you have a story to tell? Know someone who does? Drop us an email at email@example.com.
Jennifer: This episode was taped in New York City in December of 2020 and produced by me with help from Anthony Green and Emma Cillekens. We’re edited by Michael Reilly and Mat Honan. Our mix engineer is Garret Lang… with sound design and music by Jacob Gorski.
Thanks for listening, I’m Jennifer Strong.
Google just launched Bard, its answer to ChatGPT—and it wants you to make it better
Google has a lot riding on this launch. Microsoft partnered with OpenAI to make an aggressive play for Google’s top spot in search. Meanwhile, Google blundered straight out of the gate when it first tried to respond. In a teaser clip for Bard that the company put out in February, the chatbot was shown making a factual error. Google’s value fell by $100 billion overnight.
Google won’t share many details about how Bard works: large language models, the technology behind this wave of chatbots, have become valuable IP. But it will say that Bard is built on top of a new version of LaMDA, Google’s flagship large language model. Google says it will update Bard as the underlying tech improves. Like ChatGPT and GPT-4, Bard is fine-tuned using reinforcement learning from human feedback, a technique that trains a large language model to give more useful and less toxic responses.
Google has been working on Bard for a few months behind closed doors but says that it’s still an experiment. The company is now making the chatbot available for free to people in the US and the UK who sign up to a waitlist. These early users will help test and improve the technology. “We’ll get user feedback, and we will ramp it up over time based on that feedback,” says Google’s vice president of research, Zoubin Ghahramani. “We are mindful of all the things that can go wrong with large language models.”
But Margaret Mitchell, chief ethics scientist at AI startup Hugging Face and former co-lead of Google’s AI ethics team, is skeptical of this framing. Google has been working on LaMDA for years, she says, and she thinks pitching Bard as an experiment “is a PR trick that larger companies use to reach millions of customers while also removing themselves from accountability if anything goes wrong.”
Google wants users to think of Bard as a sidekick to Google Search, not a replacement. A button that sits below Bard’s chat widget says “Google It.” The idea is to nudge users to head to Google Search to check Bard’s answers or find out more. “It’s one of the things that help us offset limitations of the technology,” says Krawczyk.
“We really want to encourage people to actually explore other places, sort of confirm things if they’re not sure,” says Ghahramani.
This acknowledgement of Bard’s flaws has shaped the chatbot’s design in other ways, too. Users can interact with Bard only a handful of times in any given session. This is because the longer large language models engage in a single conversation, the more likely they are to go off the rails. Many of the weirder responses from Bing Chat that people have shared online emerged at the end of drawn-out exchanges, for example.
Google won’t confirm what the conversation limit will be for launch, but it will be set quite low for the initial release and adjusted depending on user feedback.
Google is also playing it safe in terms of content. Users will not be able to ask for sexually explicit, illegal, or harmful material (as judged by Google) or personal information. In my demo, Bard would not give me tips on how to make a Molotov cocktail. That’s standard for this generation of chatbot. But it would also not provide any medical information, such as how to spot signs of cancer. “Bard is not a doctor. It’s not going to give medical advice,” says Krawczyk.
Perhaps the biggest difference between Bard and ChatGPT is that Bard produces three versions of every response, which Google calls “drafts.” Users can click between them and pick the response they prefer, or mix and match between them. The aim is to remind people that Bard cannot generate perfect answers. “There’s the sense of authoritativeness when you only see one example,” says Krawczyk. “And we know there are limitations around factuality.”
How AI experts are using GPT-4
Hoffman got access to the system last summer and has since been writing up his thoughts on the different ways the AI model could be used in education, the arts, the justice system, journalism, and more. In the book, which includes copy-pasted extracts from his interactions with the system, he outlines his vision for the future of AI, uses GPT-4 as a writing assistant to get new ideas, and analyzes its answers.
A quick final word … GPT-4 is the cool new shiny toy of the moment for the AI community. There’s no denying it is a powerful assistive technology that can help us come up with ideas, condense text, explain concepts, and automate mundane tasks. That’s a welcome development, especially for white-collar knowledge workers.
However, it’s notable that OpenAI itself urges caution around use of the model and warns that it poses several safety risks, including infringing on privacy, fooling people into thinking it’s human, and generating harmful content. It also has the potential to be used for other risky behaviors we haven’t encountered yet. So by all means, get excited, but let’s not be blinded by the hype. At the moment, there is nothing stopping people from using these powerful new models to do harmful things, and nothing to hold them accountable if they do.
Chinese tech giant Baidu just released its answer to ChatGPT
So. Many. Chatbots. The latest player to enter the AI chatbot game is Chinese tech giant Baidu. Late last week, Baidu unveiled a new large language model called Ernie Bot, which can solve math questions, write marketing copy, answer questions about Chinese literature, and generate multimedia responses.
A Chinese alternative: Ernie Bot (the name stands for “Enhanced Representation from kNowledge IntEgration;” its Chinese name is 文心一言, or Wenxin Yiyan) performs particularly well on tasks specific to Chinese culture, like explaining a historical fact or writing a traditional poem. Read more from my colleague Zeyi Yang.
Even Deeper Learning
Language models may be able to “self-correct” biases—if you ask them to
Large language models are infamous for spewing toxic biases, thanks to the reams of awful human-produced content they get trained on. But if the models are large enough, they may be able to self-correct for some of these biases. Remarkably, all we might have to do is ask.
Texas is trying out new tactics to restrict access to abortion pills online
Texas is trying to limit access to abortion pills by cracking down on internet service providers and credit card processing companies. These tactics reflect the reality that, post-Roe, the internet is a critical channel for people seeking information about abortion or trying to buy pills to terminate a pregnancy—especially in states where they can no longer access these things in physical pharmacies or medical centers.
Texas has long been a laboratory for anti-abortion political tactics, and on March 15, a US District Judge heard arguments in a case that’s seeking to reverse the FDA approval of mifepristone, a drug that can be used to terminate an early pregnancy. The case would limit online-facilitated abortions and would have far-reaching consequences even in states that are not trying to restrict abortion.
Earlier this month, Republicans in the Texas state legislature introduced two bills to restrict access to abortion pills. The first bill, HB 2690, would require internet service providers (ISPs) to ban sites that provide access to the pills or information about obtaining them. Companies like AT&T and Spectrum would have to “make every reasonable and technologically feasible effort to block Internet access to information or material intended to assist or facilitate efforts to obtain an elective abortion or an abortion-inducing drug.” The bill would also forbid both publishers and ordinary people from providing information about access to abortion-inducing drugs.
The second bill, SB 1440, would make it a felony for credit card companies to process transactions for abortion pills, and would also make them liable to lawsuits from the public.
Blair Wallace, a policy and advocacy strategist at the ACLU of Texas, a nonprofit that advocates for civil liberties and reproductive choice, said the recent developments mark “a new frontier for the ways in which they’re coming for [abortion access],” adding: “It is really terrifying.”
Wallace sees it as a continuation of a strategy that seeks to criminalize whole abortion care networks with the aim of isolating people seeking abortions. More broadly, this strategy of censoring information and language has become a popular tactic in US culture wars in the last several years, and the proposed bill could incentivize platforms to aggressively remove information about abortion access out of concern for legal risk. Some sites, like Meta’s Instagram and Facebook, have reportedly removed information about abortion pills in the past.
So what might the outcome of all the Texas action be? Both the bill that targets ISPs and the mifepristone case this week are unprecedented, which means neither is likely to be successful. That said, the tactics are likely to stay. “Will we see it again next session? Will we see parts of this bill stripped down and put into amendments? There’s like a million ways that this can play out,” says Wallace. Anti-abortion political strategy is coordinated nationally even though the fights are playing out at a state level, and it’s likely that other states will target online spaces going forward.
Online abortion resources can pose risks to privacy. But there are lots of ways to access them more safely. Here are some resources I recommend.