Pamela McCorduck: Machines Who Think and the Early Days of AI

00:00:00.000 | The following is a conversation with Pamela McCordick. She's an author who has written on

00:00:04.800 | the history and the philosophical significance of artificial intelligence. Her books include

00:00:10.400 | Machines Who Think in 1979, The Fifth Generation in 1983 with Ed Vangenbaum, who's considered to

00:00:18.160 | be the father of expert systems, The Edge of Chaos that features women, and many more books.

00:00:24.800 | I came across her work in an unusual way by stumbling on a quote from Machines Who Think

00:00:30.320 | that is something like, "Artificial intelligence began with the ancient wish to forge the gods."

00:00:36.880 | That was a beautiful way to draw a connecting line between our societal relationship with AI

00:00:42.960 | from the grounded day-to-day science, math, and engineering to popular stories and science fiction

00:00:50.000 | and myths of automatons that go back for centuries. Through her literary work,

00:00:55.520 | she has spent a lot of time with the seminal figures of artificial intelligence, including

00:01:01.040 | the founding fathers of AI from the 1956 Dartmouth Summer Workshop where the field was launched.

00:01:08.960 | I reached out to Pamela for a conversation in hopes of getting a sense of what those early

00:01:14.400 | days were like and how their dreams continue to reverberate through the work of our community today.

00:01:20.160 | I often don't know where the conversation may take us, but I jump in and see. Having no constraints,

00:01:26.880 | rules, or goals is a wonderful way to discover new ideas. This is the Artificial Intelligence Podcast.

00:01:33.760 | If you enjoy it, subscribe on YouTube, give it five stars on iTunes, support it on Patreon,

00:01:39.680 | or simply connect with me on Twitter @LexFriedman, spelled F-R-I-D-M-A-N.

00:01:45.600 | And now, here's my conversation with Pamela McCordick.

00:01:49.760 | In 1979, your book, Machines Who Think, was published. In it, you interview some of the

00:01:58.240 | early AI pioneers and explore the idea that AI was born not out of maybe math and computer science,

00:02:06.320 | but out of myth and legend. So, tell me if you could the story of how you first

00:02:13.360 | arrived at the book, the journey of beginning to write it.

00:02:18.000 | I had been a novelist. I'd published two novels. And I was sitting under the portal at Stanford

00:02:29.680 | one day, the house we were renting for the summer, and I thought, "I should write a novel about these

00:02:34.480 | weird people in AI I know." And then I thought, "Ah, don't write a novel, write a history. Simple,

00:02:41.920 | just go around, interview them, splice it together, voila, instant book. Ha, ha, ha." It was

00:02:48.880 | much harder than that. But nobody else was doing it. And so, I thought, "Oh, this is a great

00:02:54.960 | opportunity." And there were people who, John McCarthy, for example, thought it was a nutty

00:03:04.400 | idea. The field had not evolved yet, so on. And he had some mathematical thing he thought I should

00:03:11.680 | write instead. And I said, "No, John, I am not a woman in search of a project. This is what I want

00:03:18.480 | to do. I hope you'll cooperate." And he said, "Oh, mutter, mutter. Well, okay, it's your time."

00:03:24.800 | What was the pitch for the, I mean, such a young field at that point? How do you write a personal

00:03:31.920 | history of a field that's so young? I said, "This is wonderful. The founders of the field

00:03:37.920 | are alive and kicking and able to talk about what they're doing." Did they sound or feel like

00:03:43.120 | founders at the time? Did they know that they have founded something? Oh, yeah. They knew what

00:03:49.760 | they were doing was very important. Very. What I now see in retrospect is that they were at the

00:03:58.560 | height of their research careers. And it's humbling to me that they took time out from all the things

00:04:06.240 | that they had to do as a consequence of being there. And to talk to this woman who said,

00:04:12.720 | "I think I'm going to write a book about you." No, it was amazing. Just amazing.

00:04:17.040 | - So who stands out to you? Maybe looking 63 years ago, the Dartmouth conference.

00:04:25.120 | So Marvin Minsky was there. McCarthy was there. Claude Shannon, Alan Newell, Herb Simon,

00:04:32.960 | some of the folks you've mentioned. Then there's other characters, right? One of your co-authors.

00:04:40.160 | - He wasn't at Dartmouth. - He wasn't at Dartmouth.

00:04:43.920 | - No. He was, I think, an undergraduate then. - And of course, Joe Traub. I mean,

00:04:50.960 | all of these are players, not at Dartmouth, but in that era. It's CMU and so on. So who are the

00:04:59.280 | characters, if you could paint a picture, that stand out to you from memory? Those people you've

00:05:04.880 | interviewed and maybe not, people that were just in the... - In the atmosphere.

00:05:09.680 | - In the atmosphere. - Of course, the four founding

00:05:13.200 | fathers were extraordinary guys. They really were. - Who are the founding fathers?

00:05:17.040 | - Alan Newell, Herbert Simon, Marvin Minsky, John McCarthy. They were the four who were not only at

00:05:24.800 | the Dartmouth conference, but Newell and Simon arrived there with a working program called the

00:05:29.920 | Logic Theorist. Everybody else had great ideas about how they might do it, but they weren't

00:05:37.520 | going to do it yet. And you mentioned Joe Traub, my husband. I was immersed in AI before I met Joe,

00:05:48.640 | because I had been Ed Feigenbaum's assistant at Stanford. And before that, I had worked on a book

00:05:56.080 | edited by Feigenbaum and Julian Feldman called Computers and Thought. It was the first textbook

00:06:05.600 | of readings of AI. And they only did it because they were trying to teach AI to people at Berkeley,

00:06:12.080 | and there was nothing, you'd have to send them to this journal and that journal. This was not

00:06:15.920 | the internet where you could go look at an article. So, I was fascinated from the get-go by AI. I was

00:06:23.920 | an English major. What did I know? And yet I was fascinated. And that's why you saw that

00:06:32.320 | historical, that literary background, which I think is very much a part of the continuum of AI.

00:06:39.760 | The AI grew out of that same impulse.

00:06:44.640 | - Yeah, that tradition. What drew you to AI? How did you even think of it back then? What

00:06:53.040 | was the possibilities, the dreams? What was interesting to you?

00:06:56.880 | - The idea of intelligence outside the human cranium, this was a phenomenal idea. And even

00:07:06.480 | when I finished Machines Who Think, I didn't know if they were going to succeed. In fact,

00:07:11.600 | the final chapter is very wishy-washy, frankly. Well, succeed the field did.

00:07:19.040 | - Yeah, so was there the idea that AI began with the wish to forge the gods? So, the

00:07:27.040 | spiritual component that we crave to create this other thing greater than ourselves.

00:07:34.240 | - For those guys, I don't think so. Newell and Simon were cognitive psychologists. What they

00:07:42.720 | wanted was to simulate aspects of human intelligence, and they found they could do it

00:07:50.640 | on the computer. Minsky just thought it was a really cool thing to do. Likewise, McCarthy.

00:08:00.080 | McCarthy had got the idea in 1949 when he was a Caltech student. And he listened to somebody's

00:08:10.400 | lecture. It's in my book. I forget who it was. And he thought, "Oh, that would be fun to do. How

00:08:17.200 | do we do that?" And he took a very mathematical approach. Minsky was hybrid, and Newell and Simon

00:08:24.960 | were very much cognitive psychology. How can we simulate various things about human cognition?

00:08:34.000 | What happened over the many years is, of course, our definition of intelligence expanded tremendously.

00:08:41.680 | I mean, these days, biologists are comfortable talking about the intelligence of a cell,

00:08:47.360 | the intelligence of the brain, not just human brain, but the intelligence of any kind of brain.

00:08:56.720 | Cephalopods, I mean, an octopus is really intelligent by any amount. We wouldn't have

00:09:03.600 | thought of that in the '60s, even the '70s. So, all these things have worked in. And I did hear

00:09:11.120 | one behavioral primatologist, Franz De Waal, say, "AI taught us the questions to ask."

00:09:20.640 | - Yeah, this is what happens, right? When you try to build it is when you start to actually

00:09:27.760 | ask questions. It puts a mirror to ourselves. - Yeah, right.

00:09:31.040 | - So, you were there in the middle of it. It seems like not many people were asking the questions

00:09:38.640 | that you were, or just trying to look at this field the way you were.

00:09:42.640 | - I was solo. When I went to get funding for this, because I needed somebody to transcribe the

00:09:50.320 | interviews, and I needed travel expenses, I went to everything you could think of, the NSF, the

00:10:01.120 | DARPA. There was an Air Force place that doled out money. And each of them said, "Well, that was

00:10:12.720 | very interesting. That's a very interesting idea, but we'll think about it." And the National Science

00:10:22.000 | Foundation actually said to me in plain English, "Hey, you're only a writer. You're not a historian

00:10:27.600 | of science." And I said, "Yeah, that's true, but the historians of science will be crawling all

00:10:32.960 | over this field. I'm writing for the general audience." So, I thought. And they still wouldn't

00:10:40.480 | budge. I finally got a private grant without knowing who it was from, from Ed Fredkin at MIT.

00:10:47.280 | He was a wealthy man, and he liked what he called crackpot ideas. And he considered--

00:10:53.520 | - This a crackpot idea?

00:10:54.720 | - This a crackpot idea, and he was willing to support it. I am ever grateful, let me say that.

00:10:59.920 | - You know, some would say that a history of science approach to AI, or even just a history,

00:11:07.040 | or anything like the book that you've written, hasn't been written since. Maybe I'm not familiar,

00:11:14.400 | but it's certainly not many. If we think about bigger than just these couple of decades,

00:11:20.720 | few decades, what are the roots of AI? - Oh, they go back so far. Yes, of course,

00:11:30.320 | there's all the legendary stuff, the golem and the early robots of the 20th century.

00:11:38.800 | But they go back much further than that. If you read Homer, Homer has robots in the Iliad. And

00:11:46.560 | a classical scholar was pointing out to me just a few months ago, "Well, you said you just read

00:11:52.800 | the Odyssey. The Odyssey is full of robots." "It is?" I said. "Yeah, how do you think Odysseus'

00:11:59.120 | ship gets from one place to another? He doesn't have the crew people to do that, the crewmen."

00:12:03.760 | "Yeah, it's magic, it's robots." "Oh," I thought, "how interesting."

00:12:09.840 | So we've had this notion of AI for a long time. And then toward the end of the 19th century,

00:12:19.200 | the beginning of the 20th century, there were scientists who actually tried to

00:12:24.000 | make this happen some way or another. Not successfully, they didn't have the technology

00:12:29.200 | for it. And of course, Babbage, in the 1850s and '60s, he saw that what he was building was capable

00:12:40.080 | of intelligent behavior. And when he ran out of funding, the British government finally said,

00:12:47.040 | "That's enough." He and Lady Lovelace decided, "Oh, well, why don't we play the ponies with this?"

00:12:54.720 | He had other ideas for raising money too. - But if we actually reach back once again,

00:13:01.520 | I think people don't actually really know that robots do appear, or the ideas of robots. You

00:13:07.440 | talk about the Hellenic and the Hebraic points of view. - Oh, yes. - Can you tell me about each?

00:13:14.240 | - I defined it this way. The Hellenic point of view is robots are great. They are party help.

00:13:22.400 | They help this guy Hephaestus, this god Hephaestus in his forge. I presume he made them to help him,

00:13:30.240 | and so on and so forth. And they welcome the whole idea of robots. The Hebraic view has to do with,

00:13:39.360 | I think it's the second commandment, "Thou shalt not make any graven image." In other words,

00:13:46.640 | you better not start imitating humans because that's just forbidden. It's the second commandment.

00:13:54.480 | And a lot of the reaction to artificial intelligence has been a sense that this is

00:14:07.440 | somehow wicked. This is somehow blasphemous. We shouldn't be going there. Now, you can say,

00:14:16.880 | "Yeah, but there are going to be some downsides." And I say, "Yes, there are,

00:14:20.080 | but blasphemy is not one of them." - You know, there is a kind of fear that feels to be almost

00:14:25.920 | primal. Is there religious roots to that? Because so much of our society has religious roots,

00:14:33.760 | and so there is a feeling of, like you said, blasphemy, of creating the other, of creating

00:14:40.240 | something, you know, it doesn't have to be artificial intelligence, it's creating life

00:14:45.520 | in general. It's the Frankenstein idea. - There's the annotated Frankenstein on my

00:14:51.440 | coffee table. It's a tremendous novel. It really is just beautifully perceptive. Yes,

00:15:01.280 | we do fear this, and we have good reason to fear it, but because it can get out of hand.

00:15:06.880 | - Maybe you can speak to that fear, the psychology, if you've thought about it. You know,

00:15:11.360 | there's a practical set of fears, concerns in the short term. You can think of, if we actually think

00:15:16.320 | about artificial intelligence systems, you can think about bias, of discrimination in algorithms,

00:15:24.400 | you can think about, there's social networks have algorithms that recommend the content you see,

00:15:33.120 | thereby these algorithms control the behavior of the masses, there's these concerns.

00:15:37.760 | But to me, it feels like the fear that people have is deeper than that. So,

00:15:44.320 | have you thought about the psychology of it? - I think in a superficial way I have.

00:15:50.640 | There is this notion that if we produce a machine that can think, it will outthink us,

00:15:59.760 | and therefore replace us. - I guess that's a primal fear of almost

00:16:05.840 | a kind of mortality. So, around the time you said you worked at Stanford with Ed Feigenbaum.

00:16:18.640 | So, let's look at that one person throughout his history, clearly a key person, one of the many

00:16:24.640 | in the history of AI. How has he changed in general around him? How has Stanford changed in the last,

00:16:33.680 | how many years are we talking about here? - Oh, since '65.

00:16:38.400 | - '65. So, maybe it doesn't have to be about him, it could be bigger, but because he was a key person

00:16:45.200 | at Expert Systems, for example, how is that, how are these folks who you've interviewed

00:16:51.440 | in the '70s, '79, changed through the decades? -

00:16:59.480 | In Ed's case, I know him well, we are dear friends, we see each other

00:17:09.760 | every month or so. He told me that when Machines Who Think first came out, he really thought all

00:17:15.760 | the front matter was kind of baloney. And 10 years later, he said, "No, I see what you're getting at.

00:17:25.200 | Yes, this is an impulse that has been, this has been a human impulse for thousands of years,

00:17:31.440 | to create something outside the human cranium that has intelligence."

00:17:36.720 | - I think it's very hard when you're down at the algorithmic level, and you're just trying to make

00:17:47.680 | something work, which is hard enough, to step back and think of the big picture. It reminds me of

00:17:55.920 | when I was in Santa Fe, I knew a lot of archaeologists, which was a hobby of mine. And

00:18:04.400 | I would say, "Yeah, yeah, well, you can look at the shards and say, oh, this came from this

00:18:09.040 | tribe, and this came from this trade route, and so on, but what about the big picture?"

00:18:14.560 | And a very distinguished archaeologist said to me, "They don't think that way. No, they're trying to

00:18:23.120 | match the shard to where it came from. That's, you know, where did this corn, the remainder of this

00:18:30.160 | corn come from? Was it grown here? Was it grown elsewhere?" And I think this is part of the AI,

00:18:36.320 | any scientific field. You're so busy doing the hard work, and it is hard work,

00:18:45.200 | that you don't step back and say, "Oh, well, now let's talk about the, you know,

00:18:49.920 | the general meaning of all this." - So, none of the, even Minsky and McCarthy,

00:18:58.080 | they-- - Oh, those guys did, yeah. The founding fathers did. - Early on, or-- - Pretty early on.

00:19:04.560 | - Yeah. - But in a different way from how I looked at it, the two cognitive psychologists,

00:19:11.200 | Newell and Simon, they wanted to imagine reforming cognitive psychology so that we would really,

00:19:20.960 | really understand the brain. Minsky was more speculative, and John McCarthy saw it as,

00:19:31.520 | I think I'm doing him right by this, he really saw it as a great boon for human beings to have

00:19:40.320 | this technology, and that was reason enough to do it. And he had wonderful, wonderful

00:19:48.880 | fables about how if you do the mathematics, you will see that these things are really good for

00:19:56.800 | human beings. And if you had a technological objection, he had an answer, a technological

00:20:03.440 | answer, but here's how we could get over that, and then blah, blah, blah, blah. And one of his

00:20:09.600 | favorite things was what he called the literary problem, which of course he presented to me

00:20:14.960 | several times, that is, everything in literature, there are conventions in literature. One of the

00:20:22.800 | conventions is that you have a villain and a hero, and the hero in most literature is human,

00:20:36.160 | and the villain in most literature is a machine. And he said, "No, that's just not the way it's

00:20:41.600 | going to be." But that's the way we're used to it, so when we tell stories about AI,

00:20:46.000 | it's always with this paradigm. I thought, "Yeah, he's right." You know, looking back,

00:20:51.840 | the classics, RUR is certainly the machines trying to overthrow the humans.

00:20:59.680 | Frankenstein is different. Frankenstein is a creature. He never has a name. Frankenstein,

00:21:10.160 | of course, is the guy who created him, the human, Dr. Frankenstein. This creature wants to be loved,

00:21:19.280 | wants to be accepted, and it is only when Frankenstein turns his head, in fact,

00:21:26.400 | runs the other way, and the creature is without love, that he becomes the monster that he later

00:21:38.000 | becomes. So who's the villain in Frankenstein? It's unclear, right? Oh, it is unclear, yeah.

00:21:45.520 | It's really the people who drive him. By driving him away, they bring out the worst. That's right.

00:21:54.720 | They give him no human solace, and he is driven away, you're right. He becomes, at one point,

00:22:04.560 | the friend of a blind man. He serves this blind man, and they become very friendly.

00:22:12.000 | But when the sighted people of the blind man's family come in, "Ah, you've got a monster here."

00:22:19.760 | So it's very didactic in its way. What I didn't know is that Mary Shelley and Percy Shelley were

00:22:28.640 | great readers of the literature surrounding abolition in the United States, the abolition

00:22:34.480 | of slavery, and they picked that up wholesale. You are making monsters of these people because

00:22:42.000 | you won't give them the respect and love that they deserve. Do you have, if we get philosophical for

00:22:50.080 | a second, do you worry that once we create machines that are a little bit more intelligent,

00:22:57.280 | let's look at Roomba that vacuums the clean air, that this darker part of human nature where we

00:23:04.720 | abuse the other, the somebody who's different, will come out? I don't worry about it. I could

00:23:15.120 | imagine it happening. But I think that what AI has to offer the human race will be so attractive

00:23:25.680 | that people will be won over. - So you have looked deep into these people,

00:23:33.200 | had deep conversations, and it's interesting to get a sense of stories of the way they were

00:23:40.640 | thinking and the way it was changed, the way your own thinking about AI has changed. So you mentioned

00:23:45.040 | Mr. McCarthy. What about the years at CMU, Carnegie Mellon, with Joe? - Sure. Joe was not in AI, he was

00:23:58.960 | in algorithmic complexity. - Was there always a line between AI and computer science, for example?

00:24:07.280 | Is AI its own place of outcasts? Was that the feeling? - There was a kind of outcast

00:24:14.320 | period for AI. For instance, in 1974, the new field was hardly

00:24:21.360 | 10 years old. The new field of computer science was asked by the National Science Foundation,

00:24:31.680 | I believe, but it may have been the National Academies, I can't remember,

00:24:34.400 | to tell us, tell your fellow scientists where computer science is and what it means.

00:24:44.160 | And they wanted to leave out AI. And they only agreed to put it in because Don Knuth said,

00:24:53.520 | "Hey, this is important. You can't just leave that out." - Really? Don, dude? - Don Knuth, yes.

00:24:59.280 | - I talked to him recently, too. So out of all the people. - Yes. But you see, an AI person couldn't

00:25:06.080 | have made that argument. He wouldn't have been believed. But Knuth was believed, yes. - So

00:25:11.840 | Joe Traub worked on the real stuff. - Joe was working on algorithmic complexity, but he would

00:25:19.680 | say in plain English again and again, "The smartest people I know are in AI." - Really? - Oh, yes. No

00:25:26.080 | question. Anyway, Joe loved these guys. What happened was that, I guess it was as I started

00:25:36.480 | to write Machines Who Think, Herb Simon and I became very close friends. He would walk past our

00:25:42.960 | house on Northumberland Street every day after work. And I would just be putting my cover on my

00:25:48.720 | typewriter. And I would lean out the door and say, "Herb, would you like a sherry?" And Herb almost

00:25:55.840 | always would like a sherry. So he'd stop in. And we'd talk for an hour, two hours. My journal says

00:26:04.000 | we talked this afternoon for three hours. - What was on his mind at the time in terms of

00:26:10.000 | on the AI side of things? - Oh, we didn't talk too much about AI. We talked about other things.

00:26:14.720 | - Just life. - We both love literature. And Herb had read Proust in the original French

00:26:23.280 | twice all the way through. I can't. I read it in English in translation. So we talked about

00:26:29.680 | literature. We talked about languages. We talked about music because he loved music. We talked

00:26:36.000 | about art because he was actually enough of a painter that he had to give it up because he was

00:26:44.720 | afraid it was interfering with his research and so on. So no, it was really just chat, chat,

00:26:50.960 | but it was very warm. So one summer I said to Herb, "You know, my students have all the really

00:26:59.360 | interesting conversations." I was teaching at the University of Pittsburgh then in the English

00:27:03.920 | department. You know, they get to talk about the meaning of life and that kind of thing.

00:27:08.880 | And what do I have? I have university meetings where we talk about the photocopying budget and,

00:27:15.120 | you know, whether the course on romantic poetry should be one semester or two. So Herb laughed.

00:27:22.080 | He said, "Yes, I know what you mean." He said, "But you know, you could do something about that."

00:27:27.680 | Dot, that was his wife, Dot and I used to have a salon at the University of Chicago every Sunday

00:27:34.320 | night and we would have essentially an open house and people knew it wasn't for a small talk. It was

00:27:43.200 | really for some topic of depth. He said, "But my advice would be that you choose the topic ahead

00:27:52.240 | of time." "Fine," I said. So, we exchanged mail over the summer. That was US Post in those days

00:28:01.280 | because you didn't have personal email. And I decided I would organize it and there would be

00:28:10.720 | eight of us--Alan Newell and his wife, Herb Simon and his wife, Dorothea. There was a novelist in

00:28:20.880 | town, a man named Mark Harris. He had just arrived. And his wife, Josephine. Mark was most famous

00:28:29.040 | then for a novel called Bang the Drum Slowly, which was about baseball. And Joe and me, so eight

00:28:35.600 | people. And we met monthly and we just sank our teeth into really hard topics and it was great

00:28:45.440 | fun. How have your own views around artificial intelligence changed through the process of

00:28:54.480 | writing Machines Who Think and afterwards, the ripple effects? I was a little skeptical that

00:29:01.760 | this whole thing would work out. It didn't matter. To me, it was so audacious. The whole thing being

00:29:06.640 | AI. AI generally, yeah. And in some ways, it hasn't worked out the way I expected so far.

00:29:17.680 | That is to say, there's this wonderful lot of apps, thanks to deep learning and so on.

00:29:26.880 | But those are algorithmic. And in the part of symbolic processing, there's very little yet.

00:29:37.920 | And that's a field that lies waiting for industrious graduate students.

00:29:45.600 | Maybe you can tell me some figures that popped up in your life in the 80s with expert systems,

00:29:53.040 | where there was the symbolic AI possibilities of what's... What most people think of as AI,

00:30:00.320 | if you dream of the possibilities of AI, it's really expert systems. And those hit a few walls

00:30:07.520 | and there were challenges there. And I think, yes, they will reemerge again with some new

00:30:12.160 | breakthroughs and so on. But what did that feel like, both the possibility and the winter that

00:30:17.760 | followed, the slowdown in research? - Ah, you know, this whole thing

00:30:22.560 | about AI winter is to me, a crock. - Snow winters.

00:30:26.960 | - Because I look at the basic research that was being done in the 80s, which is supposed to be,

00:30:33.760 | my God, it was really important. It was laying down things that nobody had thought about before,

00:30:40.320 | but it was basic research. You couldn't monetize it. Hence the winter.

00:30:47.600 | You know, research, scientific research goes in fits and starts. It isn't this nice, smooth,

00:30:53.600 | "Oh, this follows this, follows this." No, it just doesn't work that way.

00:30:59.280 | - The interesting thing, the way winters happen, it's never the fault of the researchers.

00:31:03.600 | It's the some source of hype over-promising. Well, no, let me take that back. Sometimes it

00:31:12.000 | is the fault of the researchers. Sometimes certain researchers might over-promise the

00:31:17.280 | possibilities. They themselves believe that we're just a few years away. Sort of just recently

00:31:23.600 | talked to Elon Musk and he believes he'll have an autonomous vehicle, will have autonomous

00:31:27.760 | vehicles in a year. And he believes it. - A year?

00:31:30.640 | - A year, yeah. With mass deployment of autonomous. - For the record, this is 2019 right now.

00:31:37.040 | So he's talking 2020. - To do the impossible,

00:31:40.480 | you really have to believe it. And I think what's going to happen when you believe it,

00:31:45.200 | 'cause there's a lot of really brilliant people around him, is some good stuff will come out of

00:31:50.160 | it. Some unexpected, brilliant breakthroughs will come out of it when you really believe it,

00:31:55.360 | when you work that hard. - I believe that. And I believe

00:31:58.560 | autonomous vehicles will come. I just don't believe it'll be in a year. I wish.

00:32:02.720 | - But nevertheless, there is, autonomous vehicles is a good example. There's a feeling many companies

00:32:09.920 | have promised by 2021, by 2022, Ford, GM, basically every single automotive company has promised

00:32:17.520 | they'll have autonomous vehicles. So that kind of over promise is what leads to the winter.

00:32:22.480 | Because we'll come to those dates, there won't be autonomous vehicles. And there'll be a feeling,

00:32:28.320 | well wait a minute, if we took your word at that time, that means we just spent billions of dollars,

00:32:35.440 | had made no money, and there's a counter response to where everybody gives up on it.

00:32:41.600 | Sort of intellectually, at every level, the hope just dies. And all that's left is a few

00:32:49.200 | basic researchers. So you're uncomfortable with some aspects of this idea? - Well, it's the

00:32:56.320 | difference between science and commerce. - So you think science goes on the way it does?

00:33:04.240 | - Oh, science can really be killed by not getting proper funding, or timely funding.

00:33:11.920 | I think Great Britain was a perfect example of that. The Lighthill Report in,

00:33:19.360 | I don't remember the year, essentially said, there's no use Great Britain putting any money

00:33:26.480 | into this, it's going nowhere. And this was all about social factions in Great Britain.

00:33:35.520 | Ed Murrow hated Cambridge, and Cambridge hated Manchester, and somebody else can write that

00:33:44.320 | story. But it really did have a hard effect on research there. Now, they've come roaring back

00:33:53.760 | with DeepMind, but that's one guy and his visionaries around him.

00:34:01.360 | - But just to push on that, it's kind of interesting, you have this dislike of the

00:34:06.720 | idea of an AI winter. Where's that coming from? - Oh, because I just don't think it's true.

00:34:15.440 | - There was particular periods of time. It's a romantic notion, certainly.

00:34:21.280 | - Yeah, well. No, I admire science perhaps more than I admire commerce. Commerce is fine. Hey,

00:34:32.880 | we all got to live. But science has a much longer view than commerce, and continues

00:34:46.640 | almost regardless. It can't continue totally regardless, but almost regardless of what's

00:34:56.400 | saleable and what's not, what's monetizable and what's not. - So the winter is just something

00:35:01.680 | that happens on the commerce side, and the science marches. That's a beautifully

00:35:08.720 | optimistic and inspiring message. I agree with you. I think if we look at the key people that

00:35:15.440 | work in AI, that work in key scientists in most disciplines, they continue working out of the

00:35:21.200 | love for science. You can always scrape up some funding to stay alive, and they continue working

00:35:28.560 | diligently. But there certainly is a huge amount of funding now, and there's a concern on the AI

00:35:37.520 | side and deep learning. There's a concern that we might, with overpromising, hit another slowdown

00:35:44.000 | in funding, which does affect the number of students, that kind of thing. - Yeah, I know it

00:35:48.480 | does. - So the kind of ideas you had to machine to think, did you continue that curiosity through

00:35:55.120 | the decades that followed? - Yes, I did. - And what was your view, historical view, of how AI

00:36:01.840 | community evolved, the conversations about it, the work? Has it persisted the same way from its

00:36:08.960 | birth? - No, of course not. It's just, we were just talking, the symbolic AI really kind of

00:36:19.200 | dried up, and it all became algorithmic. I remember a young AI student telling me what he was doing,

00:36:27.200 | and I had been away from the field long enough. I'd gotten involved with complexity at the Santa

00:36:33.280 | Fe Institute. I thought, algorithms, yeah, they're in the service of, but they're not the main event.

00:36:40.960 | No, they became the main event. That surprised me. And we all know the downside of this. We all know

00:36:49.840 | that if you're using an algorithm to make decisions based on a gazillion human decisions,

00:36:58.960 | baked into it are all the mistakes that humans make, the bigotries, the short-sightedness,

00:37:05.040 | so on and so on. - So you mentioned Santa Fe Institute. So you've written the novel,

00:37:13.760 | "Edge of Chaos," but it's inspired by the ideas of complexity, a lot of which have been extensively

00:37:22.400 | explored at the Santa Fe Institute. It's another fascinating topic of just sort of emergent

00:37:32.240 | complexity from chaos. Nobody knows how it happens, really, but it seems to where all the interesting

00:37:38.240 | stuff does happen. So how did first, not your novel, but just complexity in general and the

00:37:45.280 | work at Santa Fe fit into the bigger puzzle of the history of AI? Or maybe even your personal

00:37:52.400 | journey through that. - One of the last projects I did concerning AI in particular was looking at

00:38:02.640 | the work of Harold Cohen, the painter. And Harold was deeply involved with AI. He was a painter first.

00:38:15.360 | And what his project, "Aaron," which was a lifelong project, did was reflect

00:38:25.440 | his own cognitive processes. Harold and I, even though I wrote a book about it, we had a lot of

00:38:33.840 | friction between us. And I thought, "This is it." The book died. It was published and fell into a

00:38:44.480 | ditch. "This is it. I'm finished. It's time for me to do something different." By chance, this was a

00:38:53.600 | sabbatical year for my husband. And we spent two months at the Santa Fe Institute and two months

00:38:59.840 | at Caltech, and then the spring semester in Munich, Germany. Okay. Those two months at

00:39:11.760 | the Santa Fe Institute were so restorative for me. And I began to... The Institute was very small

00:39:19.920 | then. It was in some kind of office complex on Old Santa Fe Trail. Everybody kept their door open.

00:39:26.560 | So you could crack your head on a problem, and if you finally didn't get it, you could walk in to

00:39:33.920 | see Stuart Kaufman or any number of people and say, "I don't get this. Can you explain?"

00:39:42.480 | And one of the people that I was talking to about complex adaptive systems was Murray Gell-Mann.

00:39:51.120 | And I told Murray what Harold Cohen had done. And I said, "You know, this sounds to me like

00:39:59.200 | a complex adaptive system." And he said, "Yeah, it is." "Well, what do you know? Harold Aaron

00:40:06.480 | had all these kids and cousins all over the world in science and in economics and so on and so forth."

00:40:13.040 | I was so relieved. I thought, "Okay, your instincts are okay. You're doing the right thing."

00:40:19.680 | I didn't have the vocabulary. And that was one of the things that the Santa Fe Institute gave me.

00:40:25.840 | If I could have rewritten that book -- no, it had just come out, I couldn't rewrite it -- I would

00:40:30.880 | have had a vocabulary to explain what Aaron was doing. Okay, so I got really interested in

00:40:37.680 | what was going on at the Institute. The people were, again, bright and funny and willing to

00:40:46.960 | explain anything to this amateur. George Cowan, who was then the head of the Institute,

00:40:54.080 | said he thought it might be a nice idea if I wrote a book about the Institute.

00:40:57.600 | And I thought about it, and I had my eye on some other project, God knows what. And I said,

00:41:06.640 | "I'm sorry, George. Yeah, I'd really love to do it, but, you know, just not going to work for me at

00:41:11.760 | this moment." And he said, "Oh, too bad. I think it would make an interesting book." Well, he was

00:41:16.560 | right and I was wrong. I wish I'd done it. But that's interesting. I hadn't thought about that,

00:41:21.600 | that that was a road not taken that I wish I'd taken.

00:41:24.960 | Well, you know what, just on that point, it's quite brave for you as a writer, as sort of

00:41:34.560 | coming from a world of literature, the literary thinking, historical thinking, I mean, just from

00:41:40.080 | that world and bravely talking to quite, I assume, large egos in AI or in complexity and so on. How

00:41:54.160 | did you do it? Like, where did you, I mean, I suppose they could be intimidated of you as well,

00:42:00.560 | because it's two different worlds coming together.

00:42:02.880 | I never picked up that anybody was intimidated by me.

00:42:06.160 | But how were you brave enough? Where did you find the guts to sort of--

00:42:08.720 | God, just dumb, dumb luck. I mean, this is an interesting rock to turn over. I'm going to write

00:42:14.320 | a book about it. And you know, people have enough patience with writers if they think they're going

00:42:19.920 | to end up at a book that they let you flail around and so on.

00:42:24.240 | Well, but they also look if the writer has, there's like, if there's a sparkle in their eye,

00:42:30.240 | if they get it.

00:42:31.120 | Yeah, sure.

00:42:31.680 | Right? When were you at the Santa Fe Institute?

00:42:35.920 | The time I'm talking about is 1990, yeah, 1990, '91, '92. But we then, because Joe was an external

00:42:45.840 | faculty member, we're in Santa Fe every summer. We bought a house there. And I didn't have that

00:42:52.160 | much to do with the Institute anymore. I was writing my novels. I was doing whatever I was doing.

00:43:00.560 | But I loved the Institute. And I loved the, again, the audacity of the ideas.

00:43:10.640 | That really appeals to me.

00:43:12.960 | I think that there's this feeling, much like in great, great institutes of neuroscience, for

00:43:21.680 | example, that it's, they're in it for the long game of understanding something fundamental about

00:43:29.840 | reality and nature. And that's really exciting. So if we start now to look a little bit more

00:43:36.320 | recently, how AI is really popular today. How is this world, you mentioned algorithmic, but in

00:43:48.800 | general, is the spirit of the people, the kind of conversations you hear through the grapevine and

00:43:54.400 | so on, is that different than the roots that you remember?

00:43:57.520 | No. The same kind of excitement. The same kind of, "This is really going to make a difference

00:44:03.360 | in the world." And it will. It has.

00:44:05.840 | You know, a lot of folks, especially young, 20 years old or something,

00:44:10.240 | they think, "We've just found something special here. We're going to change the world tomorrow."

00:44:16.640 | On a time scale, do you have a sense of what, of the time scale at which breakthroughs in AI

00:44:26.720 | happen?

00:44:28.080 | I really don't. Because look at deep learning. That was, Jeffrey Hinton came up with the algorithm

00:44:39.040 | in '86. But it took all these years for the technology to be good enough to actually

00:44:51.440 | be applicable. So, no, I can't predict that at all. I can't. I wouldn't even try.

00:44:57.680 | Well, let me ask you to, not to try to predict, but to speak to the,

00:45:02.480 | you know, I'm sure in the '60s, as it continues now, there's people that think,

00:45:08.160 | let's call it, we can call it this fun word, the singularity. When there's a phase shift,

00:45:14.560 | there's some profound feeling where we're all really surprised by what's able to be achieved.

00:45:21.600 | I'm sure those dreams are there. I remember reading quotes in the '60s and those continued.

00:45:26.240 | How have your own views, maybe if you look back, about the timeline of a singularity changed?

00:45:34.960 | Well, I'm not a big fan of the singularity as Ray Kurzweil has presented it.

00:45:45.760 | How would you define the Ray Kurzweil?

00:45:50.480 | How do you think of singularity in those?

00:45:52.560 | If I understand Kurzweil's view, it's sort of, there's going to be this moment when machines

00:45:59.280 | are smarter than humans and, you know, game over. However, the game over is, I mean,

00:46:06.560 | do they put us on a reservation? Do they, et cetera, et cetera. And first of all,

00:46:13.280 | machines are smarter than humans in some ways all over the place. And they have been since

00:46:20.160 | adding machines were invented. So, it's not going to come like some great eatable crossroads,

00:46:28.640 | you know, where they meet each other and our offspring Oedipus says, "You're dead."

00:46:35.440 | It's just not going to happen.

00:46:37.920 | Yeah, so it's already game over with calculators, right? They're already out to do much better at

00:46:45.600 | basic arithmetic than us. But, you know, there's a human-like intelligence. And it's not the ones

00:46:53.680 | that destroy us, but, you know, somebody that you can have as a friend, you can have deep

00:46:59.600 | connections with that kind of passing the Turing test and beyond, those kinds of ideas. Have you

00:47:06.320 | dreamt of those?

00:47:07.520 | - Oh, yes, yes, yes.

00:47:08.880 | - Those possibilities.

00:47:10.240 | In a book I wrote with Ed Feigenbaum, there's a little story called the geriatric robot.

00:47:16.160 | And how I came up with the geriatric robot is a story in itself. But here's what the geriatric

00:47:24.880 | robot does. It doesn't just clean you up and feed you and wheel you out into the sun.

00:47:30.640 | Its great advantage is it listens. It says, "Tell me again about the great coup of '73.

00:47:42.720 | Tell me again about how awful or how wonderful your grandchildren are," and so on and so forth.

00:47:52.080 | And it isn't hanging around to inherit your money. It isn't hanging around because it can't get any

00:47:59.680 | other job. This is its job, and so on and so forth. Well, I would love something like that.

00:48:08.320 | - Yeah, I mean, for me, that deeply excites me. So, I think there's a lot of us--

00:48:15.200 | - Lex, you gotta know, it was a joke. I dreamed it up because I needed to talk to college students,

00:48:20.880 | and I needed to give them some idea of what AI might be. And they were rolling in the aisles

00:48:26.880 | as I elaborated and elaborated and elaborated. When it went into the book,

00:48:32.160 | they took my hide off in the New York Review of Books. This is just what we have thought about

00:48:40.240 | these people in AI. They're inhuman. Oh, come on, get over it.

00:48:44.320 | - Don't you think that's a good thing for the world, that AI could potentially--

00:48:49.040 | - Moi, I do, absolutely. And furthermore, I want, you know, I'm pushing 80 now. By the time I need

00:48:58.480 | help like that, I also want it to roll itself in a corner and shut the fuck up.

00:49:04.080 | - Let me linger on that point. Do you really, though?

00:49:09.760 | - Yeah, I do. Here's why.

00:49:10.960 | - Don't you want it to push back a little bit?

00:49:13.120 | - A little, but I have watched my friends go through the whole issue around having help in

00:49:21.040 | the house. And some of them have been very lucky and had fabulous help. And some of them have had

00:49:29.360 | people in the house who want to keep the television going on all day, who want to talk on their phones

00:49:34.640 | all day. No. Just roll yourself in the corner and shut up.

00:49:39.440 | - Unfortunately, us humans, when we're assistants, we care, we're still, even when we're

00:49:46.320 | assisting others, we care about ourselves more. And so you create more frustration. And a robot,

00:49:53.200 | AI assistant can really optimize the experience for you. I was just speaking to the point,

00:50:01.600 | you actually bring up a very, very good point, but I was speaking to the fact that

00:50:05.440 | us humans are a little complicated, that we don't necessarily want a perfect servant.

00:50:11.200 | I don't, maybe you disagree with that, but there's, I think there's a push and pull with humans.

00:50:21.440 | A little tension, a little mystery that, of course, that's really difficult for AI to get right.

00:50:28.080 | But I do sense, especially in today with social media, that people are getting more and more

00:50:35.120 | lonely, even young folks, and sometimes especially young folks, that loneliness, there's a longing

00:50:42.960 | for connection and AI can help alleviate some of that loneliness. Some, just somebody who listens,

00:50:51.760 | like in person. - So to speak.

00:50:55.600 | - So to speak, yeah, so to speak. Yeah, that to me is really exciting. But so if we look at that

00:51:05.520 | level of intelligence, which is exceptionally difficult to achieve actually,

00:51:10.640 | as the singularity or whatever, that's the human level bar, that people have dreamt of that too.

00:51:18.320 | Turing dreamt of it. He had a date, timeline. Do you have, how have your own timeline

00:51:25.280 | evolved on past-- - I don't even think about it.

00:51:28.960 | - You don't even think. - No. Just, this

00:51:32.880 | field has been so full of surprises for me. - You just taking in and see--

00:51:39.440 | - Yeah, whoa, whoa, that's great. Yeah, I just can't. Maybe that's because I've been around

00:51:46.960 | the field long enough to think, don't go that way. Herb Simon was terrible about making these

00:51:54.000 | predictions of when this and that would happen. And he was a sensible guy.

00:51:59.040 | - Yeah. And his quotes are often used, right, as a--

00:52:03.600 | - As a bludgeon, yeah. - Yeah. Do you have concerns about

00:52:11.280 | AI, the existential threats, as many people, like Elon Musk and Sam Harris and others are thinking

00:52:18.560 | about? - Oh yeah, yeah. That

00:52:21.440 | takes up a half a chapter in my book. I call it the male gaze.

00:52:26.560 | - (laughs) - Well, you hear me out. The male gaze is

00:52:33.120 | actually a term from film criticism. And I'm blocking on the women who dream this up. But she

00:52:41.760 | pointed out how most movies were made from the male point of view, that women were objects, not

00:52:50.720 | subjects. They didn't have any agency, so on and so forth. So when Elon and his pals Hawking and

00:53:00.160 | so on came, "AI's gonna eat our lunch and our dinner and our midnight snack too," I thought,

00:53:07.120 | "What?" And I said to Ed Feigenbaum, "Oh, this is the first guy, first, these guys have always been

00:53:13.120 | the smartest guy on the block, and here comes something that might be smarter. Ooh, let's stamp

00:53:18.800 | it out before it takes over." And Ed laughed. He said, "I didn't think about it that way."

00:53:24.080 | But I did. I did. And it is the male gaze. Okay, suppose these things do have agency.

00:53:34.480 | Well, let's wait and see what happens. Can we imbue them with ethics? Can we imbue them with

00:53:45.520 | a sense of empathy? Or are they just gonna be, "I know we've had centuries of guys like that."

00:53:54.960 | That's interesting that the ego, the male gaze is immediately threatened.

00:54:03.680 | And so you can't think in a patient, calm way of how the tech could evolve.

00:54:13.360 | Speaking of which, your '96 book, "The Future of Women," I think at the time and now,

00:54:20.720 | certainly now, I mean, I'm sorry, maybe at the time, but I'm more cognizant of now,

00:54:25.680 | is extremely relevant. You and Nancy Ramsey talk about four possible futures of women in science

00:54:33.760 | and tech. So if we look at the decades before and after the book was released, can you tell a

00:54:41.920 | history, sorry, of women in science and tech and how it has evolved? How have things changed?

00:54:50.000 | Where do we stand? Not enough. They have not changed enough. The way that women are

00:54:58.080 | ground down in computing is simply unbelievable. But what are the four possible futures for women

00:55:08.160 | in tech from the book? What you're really looking at are various aspects of the present.

00:55:14.240 | So for each of those, you could say, "Oh yeah, we do have backlash. Look at what's happening

00:55:21.200 | with abortion," and so on and so forth. We have one step forward, one step back.

00:55:26.640 | The golden age of equality was the hardest chapter to write. And I used something from

00:55:33.440 | the Santa Fe Institute, which is the sandpile effect, that you drop sand very slowly onto a pile

00:55:41.760 | and it grows and it grows and it grows until suddenly it just breaks apart.

00:55:45.920 | And in a way, #MeToo has done that. That was the last drop of sand that broke everything apart.

00:55:58.240 | That was a perfect example of the sandpile effect. And that made me feel good. It didn't change all

00:56:04.400 | of society, but it really woke a lot of people up. - But are you in general optimistic about

00:56:10.800 | maybe after #MeToo? I mean, #MeToo is about a very specific kind of thing.

00:56:16.560 | - Boy, solve that and you'll solve everything. - But are you in general optimistic about the future?

00:56:22.800 | - Yes, I'm a congenital optimist. I can't help it. - What about AI? What are your thoughts

00:56:30.800 | about the future of AI? - Of course, I get asked, "What do you worry about?"

00:56:36.720 | And the one thing I worry about is the things we can't anticipate.

00:56:41.280 | There's going to be something out of left field that we will just say, "We weren't prepared for

00:56:48.480 | that." I am generally optimistic. When I first took up being interested in AI,

00:56:59.680 | like most people in the field, more intelligence was like more virtue. What could be bad?

00:57:06.880 | And in a way, I still believe that, but I realize that my notion of intelligence

00:57:15.280 | has broadened. There are many kinds of intelligence, and we need to imbue our

00:57:20.880 | machines with those many kinds. - So you've now just finished, or in the process of finishing

00:57:29.520 | the book that you've been working on, the memoir. How have you changed? I know it's just writing,

00:57:38.320 | but how have you changed the process? If you look back, what kind of stuff did it bring up

00:57:44.000 | to you that surprised you, looking at the entirety of it all?

00:57:49.360 | - The biggest thing, and it really wasn't a surprise, is how lucky I was. Oh my! To be,

00:58:01.200 | to have access to the beginning of a scientific field that is going to change the world.

00:58:10.480 | How did I luck out? And yes, of course, my view of things has widened a lot.

00:58:20.880 | If I can get back to one feminist part of our conversation, without knowing it,

00:58:31.680 | it really was subconscious. I wanted AI to succeed because I was so tired of hearing

00:58:39.280 | that intelligence was inside the male cranium. And I thought if there was something out there

00:58:46.240 | that wasn't a male thinking and doing well, then that would put a lie to this whole notion of

00:58:55.920 | intelligence resides in the male cranium. I did not know that until one night, Harold Cohen and I

00:59:04.400 | were having a glass of wine, maybe two, and he said, "What drew you to AI?" And I said, "Oh,

00:59:12.320 | you know, smartest people I knew, great project, blah, blah, blah." And I said, "And I wanted

00:59:17.600 | something besides male smarts." And it just bubbled up out of me, Lex. And I, "What? What's this?"

00:59:27.440 | - It's kind of brilliant, actually. So AI really humbles all of us and humbles the people that

00:59:34.800 | need to be humbled the most. - Let's hope.

00:59:37.600 | - Wow, that is so beautiful. Pamela, thank you so much for talking to me. It was really a huge honor.

00:59:44.400 | - Oh, it's been a great pleasure. - Thank you.

00:59:47.040 | - Thank you.

00:59:48.420 | - Bye.

00:59:49.880 | - Bye.

00:59:50.620 | - Bye.

00:59:52.080 | - Bye.

00:59:53.540 | - Bye.

00:59:55.000 | - Bye.

00:59:56.460 | - Bye.

00:59:57.920 | - Bye.

00:59:59.380 | - Bye.

01:00:00.840 | - Bye.

01:00:01.340 | [BLANK_AUDIO]

Pamela McCorduck: Machines Who Think and the Early Days of AI | Lex Fridman Podcast #34

Chapters