EP 2: The Signal

0:01This episode is brought to you by Google Chrome. You think you know a browser, but Gemini and Chrome, that's new. It can help you with practically anything on the web, like restoring a vintage motorcycle from a 50-page restoration block, or finally break down that long article you've had open for weeks. Gemini and Chrome is here for it. Ready to make anything online make sense? There's no place like Chrome. Check responses set up required, compatibility and availability varies 18+. When you need to build up your team to handle the growing chaos at work, use Indeed Sponsored Jobs. It gives your job post the boost it needs to be seen and helps reach people with the right skills, certifications and more.

0:38Spend less time searching and more time actually interviewing candidates who check all your boxes. Listeners of this show will get a$75 sponsored job credit at Indeed.com slash podcast. That's Indeed.com slash podcast. Terms and conditions apply. Need a hiring hero? This is a job for Indeed Sponsored Jobs.

1:00I'm Gregory Warner, and this is The Last Invention. On November 30th, 2022, the world as we know it changed forever with the introduction of ChatGPT. The robots are taking over. The internet's going crazy over new artificial intelligence called ChatGPT. So much of this moment that we're in, in the AI revolution, so much of this debate that we're having about what we should do next, was triggered by the arrival of ChatGPT. The next generation of artificial intelligence is here. It took Netflix more than three years to reach one million users, but it took ChatGPT just five days. The program can write really complex essays, books, news articles, and even computer code.

1:45And whether you're a ChatGPT fan or not, That single AI chatbot and the models that followed, they supercharged an industry, shifted our relationship to AI, and fueled this debate about our future with intelligent machines. And that moment, that meteoric impact on our collective conversation, that wasn't just predicted. That moment was forged over 70 years ago in the middle of a war. All right, Greg, so in a lot of ways, what we think of today as artificial intelligence, it comes out of the Second World War. Again, reporter Andy Mills. And actually not just the war, but one battle in that war.

2:32The Battle of the Atlantic continues. As our convoys pass to and fro, you boats lurk in the vast waters preparing to cut our lifelines at every opportunity. All right, so for context, it's 1940, and the Germans have all but cut off the supply lines between the U.S. and Great Britain in the Atlantic Ocean, thanks in part to this technological edge that their navy has in the form of U-boats. Hitler has loosed the whole weight of his U-boat force against our lifeline in the Atlantic.

3:04And it became clear that if they couldn't stop these U-boats, they might lose the war. The Battle of the Atlantic holds first place in the thoughts of those upon whom rests the responsibility for procuring the victory.

3:19One great hope that they had to reverse Germany's dominance in the Atlantic was to crack their Enigma code that they used to communicate with these U-boats. But the trouble was, month after month, as hundreds and thousands of soldiers died and ships sank, nobody could crack it. Right. And just to clarify, so code breaking at that point in history was still a very human endeavor. Yes. At this point, it is very common for pretty much every military across the world to have a team of code breakers that they work with where they try and intercept and decode the communications of their enemies. Right.

3:55But none of those teams in the Allied forces were able to crack this code. And so back in England, a team was assembled of a somewhat unlikely group of people. You had mathematicians, academics, even some chess masters. And these were folks who were not in the military. The military was trying to recruit beyond their ranks. Yes, they were recruited out of the classrooms, recruited out of their research labs, and essentially given Enigma. And the government said, look, we need help. Is there anything you could do? And eventually, after a lot of trial and error, they end up constructing this electromechanical, essentially calculator, that could sift through millions and millions of possible configurations of this code.

4:42Something that would have taken a human group of code breakers weeks and weeks in just hours. And lo and behold. German submarines have been surrendering all over the place in fairly satisfactory numbers. They crack Enigma. It opens up the Atlantic. And this is how you get, you know, events like D-Day. The Americans could now travel the Atlantic to Europe. And many people think that this is one of the key factors that leads to the Allies' victory over the Nazis. Time and time again, the very issue of the war depended on the breaking of the U-boat blockade. And now as some of the last U-boats came in from sea, no one forgot the lives given and the battles fought against the very grave menace they implied.

5:25And the main guy behind that code-breaking team, his name was Alan Turing. And according to the people that I interviewed, right there, in the middle of the war, looking at this big electromechanical contraption that he had helped make, he was already envisioning the day when that machine would be able to think for itself. If we were trying to tell the story of where AI is at right now and where it might be headed next, where does that begin? Alan Turing. Alan Turing. Alan Turing, obviously one of the godfathers of computer science and was the father of AI. So from almost day one, Turing saw computers not just as tools that could break codes, but as machines that could think at the highest level.

6:12Yes, and from the very beginning of this field of computer science, he inspired this goal to make what today we call AGI. It is the genesis, the Ur-like philosopher's stone of the field of computer science. Basically, this is the holy grail of the last 75 years of computer science. And even more dramatically than that, he believed that ultimately the machines would think even better than humans. And that when that happened, they would be able to take control. He talks about how surely one day, you know, there will be machines that can converse with each other, improve upon their things, do anything a human can do, and they will surely leave humanity behind.

7:00The idea that this is possible has been around for a very long time. Alan Turing in 1951 said that the default outcome is that the machines then are going to take control. This is something that I talked to Max Tegmark about. He teaches machine learning at MIT and is a very influential voice in the AI debate we're having today. Because he didn't think of AI as just another technology like the steam engine. He thought about it as a new species. But Turing was pretty chill about it and said, you know, don't worry about it. It's far away. He was right. It was far away from 1951. And he said, but I'll give you a test so you know when you're close.

7:36He'll give you a canary in the coal mine. It's called a Turing test. And Max says that this is one of the reasons that he created what's called the Turing test. You know the Turing test, right? I think the Turing test is where you're chatting with possibly a machine, possibly a human, something on the other side of the screen. And the goal of the test is, can the machine fool you into thinking that you're actually chatting with a human? Yes. That, in short, is the Turing test. Can you be in conversation with a machine and not know the difference between it and a human being? And Tegmark, he was saying that this was not a benign test.

8:12This wasn't even necessarily just a test of the machine, but it was a way to send a signal to people in the future to say that once you've crossed this threshold. When machines can master language and knowledge at the level of humans, then you're close. It was like a warning shot to the future to tell them there's no going back, because soon it will be outside of our control. And did Turing see this future of machines taking control as something good? or something he was warning us about? Well, that question is very much up for debate right now. A lot of the people who are worried about this moment we're in with artificial intelligence now, they look at this one line where he said, once the machine thinking method has started, it would not take long to outstrip our feeble powers.

9:15At some stage, therefore, we should have to expect the machines to take control. And they say, obviously, that sounds very foreboding. He was warning us that this was going to be dangerous. But in most of his lectures, he's far more matter of fact. And a lot of people point out that Turing himself, he wasn't a moralist. He was a contrarian. And he often went as far as to say that when this happened, the machines would, quote unquote, deserve our respect. And so a lot of the people that I spoke to said that trying to take Turing or the Turing test and to turn it into something Doomer or something accelerationist is missing the point.

9:56I actually think the difference between the optimists and the pessimists can be overstated. I think the fundamental difference has always more fundamentally been between people who thought this was a real possibility to take seriously and the people who didn't. And I think the main thrust of the argument was to say, look, people, take this seriously. This is a thing. This is something I was talking about with Robin Hanson. He is an economist, but he also for many years was an AI researcher. It's a rhetorical device to try to get people to take a space of possibility seriously. There's certainly in academia, there's just a long tradition, quite reasonable, of saying, until you can even tell me what your words mean, I'm not really interested in taking your conversation seriously, right?

10:45So Turing was primarily trying to just make sure we could talk about an AI and have that be somewhat precisely defined so that you wouldn't dismiss the idea by just saying that's too vague. It sounds like so Turing didn't just give us a test of how we'd know the AI could think, but actually gave us some kind of practical, concrete way of saying, okay, no, this is artificial intelligence when it can talk back to us and convince us it's human. See, it's more complicated than that. That's what I thought the Turing test was for a long time, that once it can do this, it is now officially an AI. But he actually didn't think that it was gonna suddenly be super useful in this moment.

11:27It wasn't that now you are in the presence of a true thinking machine and it's tomorrow going to start taking over. What he was doing was more nuanced. He was giving the field of computer science a goal that they could shoot for. He was saying this is something you could technically go out and build. But in this deeper sense, he was producing this story that the broader public could understand and making this prediction that once this moment happened, once a machine could so thoroughly mimic how an intelligent person communicates, that we would treat it differently. that we would imbue it with something profound and utterly transformative.

12:20So it's not just a test of the machines and how far it's come, but it would change our relationship to the machine. Right. In some ways, it's just as much a test about us as it is a test about the machine. So what happens to that dream of Turing's? Well, sadly, Turing dies in 1954. It's a terrible story. He was prosecuted by his own government for being gay. Because homosexuality was illegal in England at that time. Yes. He was sentenced to what they called chemical castration. And allegedly, he committed suicide. But Turing's dream doesn't die with him. It gets picked up by a group of scientists in the U.S., many of whom actually knew Turing personally.

13:09A lot of them had corresponded with him. And they raised money for a 10-week summer program at Dartmouth where the idea was to come together, create a prototype of a true thinking machine, and to turn the entire pursuit into a proper field of study. So the AI discipline is founded in the summer of 1956. This is a story that I talked to Karen Howe about. She is a tech reporter and the author of Empire of AI. The people that had gathered together were already very accomplished scientists, giants in their own fields. So this summer program included people like Claude Shannon, who was the inventor of information theory.

13:51Also Claude, the namesake of the anthropic AI model, Claude. Indeed, he was. Nathan Rochester was there. He was the maker of the first commercial computer at IBM. There were also people there like Marvin Minsky and John McCarthy who would found the first AI labs at MIT. And the reason why this is considered the origin story of the field is because a Dartmouth professor, John McCarthy, coined the term artificial intelligence and started using it for the first time to form this new field. And this is where we first get the name artificial intelligence. What did they call these thinking machines before that?

14:29Well, Turing called them thinking machines. Some of them were calling it automata. John McCarthy tried originally to call it automata studies. Which, personally, I like. And it just didn't sound exciting. They were trying to attract funding from government and from nonprofits, and it just wasn't working. So he specifically went to cast a bout for a more evocative phrase and hit upon artificial intelligence. And this name, artificial intelligence. Karen says that this would forever shape the field, partly because it would tie it to the thorny question of what actually is human intelligence. And because there is no consensus around where human intelligence comes from, there was plenty of debate and discourse at the time about what would it actually take to get machines to think.

15:22So everybody there believed in what Turing was saying, that machines could think that AI could be built, but the debate was more about, well, what exactly are we mimicking? What is the intelligence we're copying? Yes. How does our intelligence work, and therefore, how would we recreate it? And right away, that question, it splits the AI researchers into these two different groups, and it gives birth to two different paths towards AI that continue to this day. And the dominant two camps that emerged were called the connectionists and the symbolists. So the symbolists believe that human intelligence comes from the fact that we know things.

16:05And so if you want to recreate intelligent computer systems, you need to encode them with databases of knowledge. So this created a branch of AI focused on building so-called expert systems. All right, so the first group, the symbolists, they get their name because they believed you could build intelligence into a machine by writing rules and using symbols like numbers and words. And that you could essentially encode human-like intelligence and create something like a human expert. But the connectionists, they believe human intelligence comes from the fact that humans can learn. And so to recreate intelligent computer systems, then we need to develop machine learning systems, software that can learn from data.

16:56But the second group, the connectionists, they say human intelligence, it doesn't come from shoving a bunch of expertise and logic into our brains. It comes from our ability to find our own patterns in the world and to find our own connections between those patterns. Right. So instead of trying to build something that's like a human expert, we should try and build something closer to a human toddler or a baby. When you watch babies grow up, they're constantly exploring the world. They're gathering all of this experience and they're quickly updating their model of their environment around them.

17:32And that's what the connectionists believe was the primary driver of how we become intelligent. That's kind of cool that you could create an AI, not that knows the things you taught it, but that it could go out and learn new things from the patterns it finds from the data. But thinking about a baby that will grow up, right, and come to its own conclusions, derive its own values, wouldn't that model be a lot less predictable than the symbolist human expert? You're teasing at the current drama, Greg. Yes, the connectionist's model would be far less controllable, far more unwieldy. And eventually, that's going to strike some fear into many a heart.

18:15In the long run, the framing that they picked, the ideas that they discussed back then, the debates that emerged that summer, have continued to have lasting impact to present day. Okay, so the summer program ends and they have a new name. They have a debate. What else do they got? Well, they've got a lot of dreams. They've got a lot of theories, but they do not have a lot of money. Remember, like we said, computer science is a brand new field. Artificial intelligence is like a branch of computer science. It's a new experimental field inside of a new and experimental field. And so they don't have the resources that they need to really build the models and turn their math and their dreams into something functional in the world.

18:58However, all that was about to change because of the arrival of, in many ways, a new battle.

19:10CBS Television presents a special report on Sputnik 1, the Soviet space satellite. On October 4, 1957, the USSR becomes the first nation to ever get a satellite into orbit. Really quite an advancement for not only the Russians, but for international science. It's the first time anybody has ever been able to get anything out that far in space and keep it there for any length of time. It's this absolutely amazing achievement for science. But of course, it comes right in the midst of the Cold War. It gets the American people alarmed that a foreign country, especially an enemy country, can do this.

19:49We fear this. And it triggers all of these fears about the communists winning the space race. Let's not fool ourselves. This may be our last chance to provide the means of saving Western civilization from annihilation. And in response to this, in a move that modern-day accelerationists say that we should embrace in our current AI race with China, the United States decides, all right, let's accelerate. These are extraordinary times, and we face an extraordinary challenge. Our strength as well as our convictions have imposed upon this nation the role of leader in freedom's cause. They flood a bunch of money and a bunch of resources into universities, into science labs.

20:42They make this huge national effort to recruit all kinds of young, talented people to go into space, to go into technology. Now it is time to take longer strides. Time for this nation to take a clearly leading role in space achievement, which in many ways may hold the key to our future on Earth. Ten, nine, ignition sequence start. Six. And it works. Liftoff. We have a liftoff. Liftoff on Apollo 11. The U.S. wins the race. We are the first to get to the moon. That's one small step for man, one giant leap for mankind. And it becomes really one of the most inspiring events in history. This realization of an enormous dream, this example of what can happen when people put everything into this goal of reaching beyond what was previously thought possible.

21:47And it turns out that this also was an absolute boom time for the field of artificial intelligence. The Cold War was really a fountain of youth for research. I found this interview of Marvin Minsky. He was one of the guys at the Dartmouth Summer Program. There was huge amounts of money, more than you needed. And he was saying that suddenly these AI labs had more money than they knew what to do with. They were finally able to build the first AI models. They ended up actually building the first AI chatbot named Eliza. Things proceeded very, very rapidly. A new generation of ideas every two or three years was a wonderful period where you had to change everything you thought.

22:39And in fact, the field of AI was moving at such a rapid pace that a lot of AI researchers began to think that by the time the astronauts got to the moon, that we were going to be living here on Earth alongside thinking robots.

22:58The Thinking Machine. Produced by the CBS Television Network. I found this old CBS archive from the 1960s. Can machines really think? I'm David Wayne, and as all of you are, I'm concerned with the world in which we're going to live tomorrow. A world in which a new machine may be of even greater importance than the atomic bomb. And in it, they're interviewing all these AI researchers at the time. But I think the computers will be doing the things that men do when we say they're thinking. And all of them are confident that this breakthrough was close. I confidently expect that within a matter of 10 or 15 years, something will emerge from the laboratories, which is not too far from the robot of science fiction fame.

23:46I'm convinced that machines can and will think in our lifetime. Okay, so they sound very optimistic about the AI they were going to make. But were there any fears and doubts about what would happen once they made it? Well, this is something that a lot of people, especially those who are worried about our current AI, marvel at about this time period. This is something that came up when I was talking to Nick Bostrom, the author of that book, Superintelligence. The early pioneers were actually quite optimistic about the timelines. They thought maybe, you know, in 10 years or so, we would be able to get machines that can do all that humans can do.

24:24So in some sense, they took that seriously. I guess that's part of what drew them into, you know, trying to program computers to do AI stuff. But in another sense, they weren't serious at all because they didn't then seem to have spent any time thinking about what would happen if they were right, if they actually did succeed in getting machines to do everything that humans can do. So you're saying that in the research labs during this time, there weren't people who were saying, oh my God, what's this going to do to the economy? Oh my God, what if this changes the society forever? Yeah, there was very little, I mean, thinking about, you know, the ethics of this, the political implications, like the safety.

25:04It was as if their imagination muscle had so exhausted itself in conceiving of this radical possibility of human-level AI that they couldn't take the obvious next step of saying, well, probably forget that, we will have superintelligence not too long after. I think that the best explanation for just why this time period had such a different mindset than we do today came from something that I heard from Robin Hanson. Why is it that you think there was not a big, robust discussion about AI safety and that they were moving so fast, asking so few questions? Why wasn't it more safetyist at the time?

25:43Well, first of all, safetyism as a cultural trend is just something that's happened since then, mostly in the world. The world back then wasn't very safetyist, honestly. Right, no seatbelt laws. They didn't have seatbelts, for example. Okay. And secondly, this was a technological framing. This was a, can we do this? Is this technologically feasible? And he was saying you have to put yourself into their mindset, which is that they are living in the aftermath of the Second World War. Many of these people were veterans of the Second World War. They are worried about another war that could break out with the Soviets.

26:21And there was this idea that if your scientists could conceive of a new technology, you have to assume that your opponent's scientists have also conceived of that technology. Because right after World War II, there was clearly this strong perception that the winners are the ones who more effectively pursue possible technological changes. There was this expectation of technological progress and an expectation that in order for your nation to stay competitive with the world, you needed to pursue feasible technologies. The idea being that it would be a better world if we made this technology and not our enemies.

27:06Good maybe for humanity, but plausibly also more good for our people. For our nation. Our nation or whatever, if we are pursuing this before the rest. And do we know if they actually thought the Soviets were trying to compete in AI? Yes. There was a rumor that was going through the Department of Defense, circulating through the government, that the Soviets and their AI researchers were right on America's heels. When you read the communication about it, it sounds so similar to how we think of China and the U.S. with AI today. Right. Like the U.S. has no choice but to barrel forward with creating this AI because, God forbid, the communists get their hands on it first.