Loading summary
Transcript274 lines
- [00:01]
Sponsor/Ad Voice
This episode is brought to you by indeed. You're ready to move your business forward, but first you need to find the right team. Start your search with Indeed Sponsored Jobs. It can help you reach qualified candidates fast, ensuring your listing is the first one they see. According to Indeed data, sponsored jobs are 90% more likely to report a hire than non sponsored jobs. See the results for yourself. Get a $75 sponsored job credit@ Indeed.com podcast. Terms and conditions apply.
- [00:30]
Thomas Haigh
Limu Emu and Doug Here we have the Limu Imu in its natural habitat, helping people customize their car insurance and save hundreds with Liberty Mutual. Fascinating. It's accompanied by his natural ally, Doug.
- [00:48]
Lee Vincl
Uh, Limu is that guy with the binoculars watching us.
- [00:51]
Thomas Haigh
Cut the camera. They see us.
- [00:52]
Lee Vincl
Only pay for what you need@libertymutual.com Liberty Liberty Liberty Liberty Savings vary underwritten by Liberty Mutual Insurance Co. Affiliates excludes Massachusetts.
- [01:01]
Sponsor/Ad Voice
Starting a business can seem like a daunting task unless you have a partner like Shopify. They have the tools you need to start and grow your business. From designing a website, to marketing, to selling and beyond, Shopify can help with everything you need. There's a reason millions of companies like Mattel, Heinz and Allbirds continue to trust and use them. With Shopify on your side, turn your big business idea into Sign up for your $1 per month trial at shopify.com.
- [01:27]
Lee Vincl
Specialoffer welcome to the New Books Network.
- [01:34]
Welcome to Peoples and Things where we explore human life with technology. I'm Lee Vinc.
- [01:58]
My God, our current AI bubble. What is there to say? Well, it's incredible. We can say that for starters, it's much larger than the dot com bubble which led to the dot com bust. Some recent articles say our current bubble is more than 17 times larger than that earlier one. And it's been so textbook. In this bubble, it's like everyone is playing from a well known standardized, generic basic bitch textbook. I mean the way that people have become irrationally exuberant around generative AI and argue that it's going to change everything. But also the way crit hypey critics, including lots of people in the fields I work in, the critics have come along and made outlandish claims about the potential and real negative effects of this technology, including about its impacts on the environment, on work and wages and employment, on critical thinking, on the degree to which people have a writerly or poetic voice, and so on and so forth. My God, the textbook nature of it all. Both the hype and the criti hype have been so, so disappointing and to be frank, to be real with you, it has broken a part of my soul. A breaking that has at once been very painful, but also ultimately profoundly liberating. When this bubble is over, what's gonna be left? Well, first off, let's just hope the air goes out of it slowly, huh? Not in a pop, or else we could be in a pile of trouble, really. I mean, AI capital expenditure accounted for something like half of US GDP this past year. If that goes all of a sudden, along with a mountain of value in folks retirement accounts and such, well, things could get pretty ugly indeed. But what else will be left over from this bubble? The reality is that none of us know. It will take years for the meso organizational and macroeconomic effects of generative AI adoption to become clear, if they ever do. And it could be, as many are arguing today, that the kinds of irrational infrastructural investments in data centers and such could have unforeseen benefits down the road, as some argue happened with telecom infrastructural investments during the dot com moment. But in the meantime, so many folks will have gone along with the powerful narratives about the promises of this technology. They will have gone along with the marketing, both the marketing of these new technologies and the marketing of critics books about the dangers of these new technologies. Because let me tell you, holy crud. There is a wave of Genii criticism books being published and on the way. Oh my God. I get asked to review and blurb them on a weekly basis. My belly tells me. Well, it tells me that most of them will hold up as well in a decade as podcast bro Sam Altman's ridiculous statements about the existential risk of AGI or whatever the fuck. But what if someone came along and told you that in a deep and non trivial sense, the very concept of AI has always been about marketing? That's what eminent computer historian Thomas Haig does in his forthcoming book, tentatively titled Artificial Intelligence the History of a Brand. Haig, who is a professor of history and affiliated faculty of Computer Science at University of Wisconsin, Milwaukee, has written a lot, including co authored books on ENIAC and the general history of modern computing, which you should look up. In his forthcoming book, Tom examines the history of modern AI going back to its roots in the 1950s, and examines how and why different groups found it useful to slot their technologies, which often had very little do with one another, under the term AI. Let me tell you folks, I love this little concise book. It packs a punch. One of the things I most appreciated about it is that Tom is able to deftly track both the real history of different developments in hardware and software and the history of what people have said about these things over time, which is what we sometimes call historiography. It's very rare for historians to be able to do these things so nicely and compactly together. And the book is fun and ironic with this kind of British understated humor. It had me in stitches. I am very excited to give you a little preview of this book, which you're going to want to buy. You know, there's a famous, in these parts, at least, famous military marching band here at Virginia Tech called the Heidy Tighties, who sometimes play outside my office window here on campus. And I say, thank you, lasses and lads, I know you came here to play just to inspire me, but sometimes I wish, what if they could just accompany me when I was introducing other people's works? My God, they could play a little fanfare.
- [07:38]
Thomas Haigh
Go.
- [07:40]
Lee Vincl
Ladies and gentlemen, I bring you Thomas Haig, talking about his great little book, Artificial Intelligence the History of a Brand, in our stinky bubble of the moment. You won't want to miss it. Hey, get excited.
- [08:18]
Tom, thanks so much for taking the time to talk to me today.
- [08:21]
Thomas Haigh
Pleasure to be here, Lee.
- [08:23]
Lee Vincl
So Artificial Intelligence the History of a Brand is a neat book.
- [08:30]
It's not out yet, but what's the kind of elevator pitch for it when you talk to people about what you're doing with it and what you're trying to do?
- [08:40]
Thomas Haigh
Well, it's a concise history of AI It's a short history in both senses. And as much as it does not start with ancient Greece and run through Frankenstein, right. It starts with the invention of AI and it looks at the things that have actually been called AI, the AI branded research labs and technologies, and particularly at how the category itself has changed over time, the continuities and discontinuities.
- [09:12]
Lee Vincl
And, you know, you.
- [09:16]
Your angle is that it's the history of a brand. And in fact, the first sentence of the current version is the history of artificial intelligence. The history of an intellectual brand, now almost seven decades old. There survived cycles of hype, disillusion, contestation, and redefinition to emerge stronger than ever. So why did you decide to take this brand angle, and what kind of work does it do for you to do that?
- [09:40]
Thomas Haigh
Well, there's a number of things that brands do, and of course, I'm as much of a historian of business as I am of technology, and I do not count myself a historian of science at all. So I think it was natural to reach for brands. So one of the things that I wanted people to understand that I think is obvious to those who have been in the field, but I'm finding is not obvious to people outside it, is that the technologies that are now gestured towards, when you say artificial intelligence, have absolutely nothing in common with the technologies that were being developed in the 20th century under the same category. So one of the things that brands do, for example, Chanel, right? So their.
- [10:28]
Best, you know, classic product really, I believe fragrances, but then there's also the Runway fashion. There are watches, there are bags. So one of the things that brands do is. Is gather together a bunch of things that have no particular inherent connection to each other and make it seem like they have a coherent kind of set of qualities. And, of course, brands evolve over time, so the specific products, or in this case, technologies that the brand is associated with can change completely. But there's this perception of something that's enduring. Another thing that brands do is impute qualities to things. So unless you're a hardcore auto engineer, you probably don't know actually what is supposed to be so great about BMWs, but you know that they're supposed to drive well, or you don't know specifically why a leather bag is worth $3,000. You know, you can't inspect it and see the quality of the stitching and so on and the leather. But you know, if it's got a Chanel or Coach brand on it, then that's kind of imputing a quality to it. So one of the things that the AI brand does is impute a connection to cognition and a connection to some broad project of achieving human, like intelligence. When the work itself, if you dig down on a technical level, may be about developing algorithms to optimize search or optimize the outcome of a function, which on a technical level are very hard to distinguish from work that might be done in operations research, for example. But that's a business school brand, and AI has been a computer science brand at least until recently. And brands can be aspirational. So you probably read the Economist, am I right?
- [12:17]
Lee Vincl
I do, yeah.
- [12:18]
Thomas Haigh
So you've seen a lot of ads for Patek Philippe, right?
- [12:21]
Lee Vincl
Yes.
- [12:22]
Thomas Haigh
That's very aspirational right now, on the face of it, paying maybe $50,000 for a watch.
- [12:30]
It'S kind of a stretch. But assuming that one day you make it big, maybe with a crypto startup, and you've got lots of crazy money lying around, by that point, you'll have seen so many ads that you'll think, well, I really should buy a Patek Philippe and hand it down to my son as an heirloom or whatever. Right. Because that branding kind of like is an aspirational thing and AI is really unusual as a field because it's named after its aspirational objective that has not as yet been reached. So it's kind of like if.
- [13:05]
Economics was named universal prosperity.
- [13:15]
Another thing with brands is we know that over time they can get tarnished. So you probably don't follow this, but one of the things that was in the British news maybe 20 years ago, a Burberry was in trouble because these rich but vulgar people called Chavs had embraced the brand.
- [13:32]
Lee Vincl
Right, right.
- [13:33]
Thomas Haigh
I do remember that actually they managed to get over that. And in a similar kind of way brands can be tainted or.
- [13:44]
Pierre Cardin was a huge one of the original big brands and that got horribly over licensed and you'd only see the products all piling up in Marshalls and so on. So something like that happened to the AI brand in the 80s there was the famous AI winter and it's come back in interesting ways and we can maybe talk about why it came back if we have time. So that's another brand like feature. And I could go on, but I think you kind of get the idea for why there's that fit there. And of course everything is brand like in that sense. Right. Like dei big brand was doing well, ran into some problems. History of technology is a brand, History of computing is a brand. Unfortunately it doesn't sell very well. So I'm not saying AI is unique in this respect, but I think for some of those reasons.
- [14:42]
Highlighting the brand ness of AI is more useful in terms of understanding what it does.
- [14:51]
To the extent to which I think it's even hard to understand what's gone on historically without taking the brand like qualities of AI seriously.
- [14:59]
Lee Vincl
No, I found it very useful in that way and there's a number of ways I'll kind of highlight that as we go along. So how did you know before we kind of jump into the book, how did you come to write this thing?
- [15:11]
You have a long career in the history of computing. You kind of co authored or co revised this big survey history of the history of computing. There we go. A new history of modern computing with Ceruzi. You co wrote this history of eniac. And so how did you go from those things to deciding to write this book?
- [15:33]
Thomas Haigh
Yeah, I'm going to have essentially two versions of this and they start in the same places. So one of them is that way back before I became a historian with my graduate training, I was a computer scientist with my undergrad and first master's degree. This was in the first half of the 1990s at the University of Manchester, which at that point was, it was said to be the largest department in Europe. So it had a broad range of specialists in different areas, including a significant team of AI people. So the thing that I took the most courses in as a computer scientist and that I did my equivalent of a senior thesis in was artificial intelligence. So I didn't do anything with that for a while. But I do draw on that, as you may have seen at some points in the book, to turn myself into an example.
- [16:28]
Not of someone who stayed in AI and did anything, but of what an ordinary rank and file student being trained in the normal science of AI in the late 20th century would have learned. Then we did the new history of Modern computing and one of the things that as I was writing it, I asked some AI people about and so on, was where in this big history should we be talking about AI now? A new history of modern computing is a history of computing practice and technology and applications. It's not a history of computer science. If you write the history of computer science, AI looms very large, and perhaps we'll talk about that if you write the history of computer practice and applications. AI is virtually nowhere until at most the last decade.
- [17:24]
So there are places, for example the famous hacker community and AI lab at MIT where you can draw lines from, that influences the development of time sharing, the development of various technologies around interactive computing and programming languages, et cetera. But obviously to have a big new, first revised, big picture history of computing in a substantial way in a generation, and it barely talk about AI.
- [17:58]
Is kind of an interesting thing to have done, given that the only thing anyone wants to talk about in the present moment is AI. So there's some sense there in following the big book that didn't mention AI, with the little book that is about AI, and having had that experience of doing something that is not very fashionable, which is writing a big picture narrative that doesn't jump around like crazy in time and isn't primarily cultural and tries to join the dots and follow something through in a kind of slightly plodding way to explain incrementally how we got from one eniac in the world that's kind of like a cyclotron or something, very specialized equipment to computers, massively, massively outnumbering humans and costing 10 cents if you buy them in bulk.
- [18:51]
So I had that experience with a big picture kind of synthetic narrative. And there's something intellectually said about trying to write a little book instead of writing another huge one. And another thing was in, I think, 2018, approximately. So you may possibly have come across Bert Grad, who is now well in his 90s, still alive, still active, who organized a whole series of oral history events and roundtable discussions, essentially from a business history viewpoint, around different aspects of the software industry.
- [19:30]
So one of those was organized with the collaboration of David Brock at the Computer History Museum, and they bought in.
- [19:39]
At least a dozen people who had been founding expert system companies in the 80s. That was interesting to me because I, as a student, been only vaguely aware that there had actually, before the present moment, been a substantial moment where AI branded companies were getting startups and hiring people and being hyped in the press. It had actually, in that one area in the 80s, broken through as a business and then collapsed. It was interesting to be there as one of the less than a handful of historians there with the pioneers. And that was also the time I was doing new history, modern computing. So those are the people I was asking, hey, I would like to put AI into this book. Where have your technologies and approaches actually made a difference in computing practice in the 20th century? And they unfortunately did not have many good examples for me, but that kind of put it back on my radar. Then there was an event in Cambridge a bit before the pandemic that I was invited to.
- [20:41]
And that had got a lot of short presentations by many people that was also giving me a sense of how the history of AI was being approached by people with other kinds of backgrounds, very much in the context of long histories. That would be saying, well, the history of AI is really the history of capitalist oppression or the history of Greek myth, or those kinds of things. Okay, so now to the point where two paths diverge. So from what I've said so far, you can see why it would make perfect sense for me to write this book, but how I actually came to write the book.
- [21:14]
So I had for eight years a collaborative visiting appointment in connection with a group that is kind of essentially media studies meets STS at Siegen University in Germany. And from there, some guest editors of a special issue of Social Studies of Science who were putting together something on AI asked me to contribute something. And I said to them, well, I don't actually work on the history of AI, but I do have my experience with writing these short, accessible historiographical things in communications with the acm. So if you think that it can make its way through the review process in social studies of science, and I did say that up front, then I could write one of those for you as a kind of provocative historiographic essay on long histories and short histories in AI. And as I worked on it, and I spent a long time reading the history of AI literature as it currently existed, it got a bit longer. So it finished up even a little bit longer than an article is supposed to be. But pretty much everyone I showed it to loved it.
- [22:28]
And the guest editors loved it and they said, I'll put some more things in, cover this, cover that. Then it was time to send it in for the actual review process. And the real editor desk rejected it and said, this thing isn't even an article. This is weird. Why would you send us this?
- [22:45]
Lee Vincl
Because it was novella in size by that point.
- [22:48]
Thomas Haigh
It was.
- [22:51]
13,000 words, I think. And they say, I think that you can be 10. But I did say, I understand this will need to be cut, but if possible, I'd like to get the reviewers input to help me know what I need to cut, which is not a crazy thing to do. So the length was part of it. But I think also fundamentally they want. I mean, the guy was a sociologist. They want social science type research. And the historiographical argumentative essay is a very specific genre.
- [23:19]
Lee Vincl
That's right.
- [23:19]
Thomas Haigh
Which, if you're not a historian, just looks weird and somewhat unhinged, perhaps.
- [23:26]
And so my concerns about. I mean, I should probably have gone and read a bunch of social citizen science from recent decades and discovered myself that there was no way they were ever going to publish this thing. And so at that point, I've put like a year's work into this. I've read the whole history of AI literature.
- [23:47]
I could just have looked around for some historical venue, right? Maybe history and technology or something that wanted to publish something like that.
- [23:56]
But what I decided to do instead was try and turn into one of those short books with the MIT Essential Knowledge series in mind, still keeping that kind of historiographical thing. So most of what was in the essay is making its way into the book in one form or another, but adding a whole bunch of stuff on the actual history.
- [24:14]
Lee Vincl
Because the problem is, that's interesting.
- [24:15]
Thomas Haigh
The actual history is not widely known, so it's hard to appreciate the historiographical points if nobody is aware of what actually happened with AI in the 20th century.
- [24:31]
MIT was very interested for the Essential Knowledge series. I didn't get a contract which probably I should have done. And then by the time in the summer where I've got the book already and I send it in, they're like, oh, yeah. So we got a new executive director and there was some controversy with some of the things in the Essential Knowledge series. So we're not allowed to have anything in Essential Knowledge anymore that has an opinion or an argument.
- [24:58]
Which, yeah, you know, STS would find that somewhat, indeed.
- [25:04]
Mirthful. But therefore the idea is that the book is still supposed to get some kind of trade press crossover thing. It's not going to be a $60 print on demand paperback with an open source thing like most MIT books are at this point.
- [25:22]
So hopefully still a book people might actually read. And there were pluses and minuses with that. So it had to be extremely concise for the Essential Knowledge. I was shooting for 35,000 words, which seems to be about the length. Finished up 42. I don't know if they'd have wanted cuts for it to stay in the series, but there were.
- [25:45]
For a short standalone book, kind of mid-50s is better. And that's pretty much, I think, where it's going to finish up. So I've been able to add some human detail, go more in depth in a few places and explain some things that might otherwise be too cryptic to people who aren't familiar with the technology.
- [26:07]
So it's going to be nice to have that little bit of space to build out from the book that I had ready at the end of last summer.
- [26:15]
And it's in some ways, I think, going to be a better book. But intellectually, I did like that challenge of trying to fit into that very specific format. And it's also a format where short books make sense because they've got those nice little square pages. And so on the holidays have a.
- [26:32]
Sponsor/Ad Voice 2
Way of sneaking up on you. And I can tell you they snuck up on me. This year I have people coming and I need to buy those people gifts for them. As I say, I just didn't have everything I need. So what I did is I went to Wayfair. From bedding to linens to decor for every room in the house, Wayfair is your one stop shop. Last minute guest prep. Wayfair has you covered. You can refresh bedding and throw pillows and accent chairs for way less. That's what I did. Pretty much all the bedding in my house is threadbare. So I decided to replace it. I went to Wayfair and I ordered some new sheets and pillowcases and I got a comforter which was really cool. I ordered it, the price was great. The shipping was free. It arrived. Now I am ready for the hordes to descend upon me. And it's not just bedding, of course. You can get linens and towels and things for the kids room, kitchen essentials, things for your living room. And of course they have holiday gifts. So get your last minute hosting essentials, gifts for all your loved ones and decor to celebrate the holidays. For way less head to Wayfair.com right now to shop all things home. That's W A Y-F-A-I R.com Wayfair every style, every home.
- [27:42]
Sponsor/Ad Voice
Close your eyes, exhale, feel your body relax and let go of whatever you're carrying today. Well, I'm letting go of the worry that I wouldn't get my new contacts in time for this class. I got them delivered free from 1-800-contacts. Oh my gosh, they're so fast.
- [27:58]
Thomas Haigh
And breathe.
- [27:59]
Sponsor/Ad Voice
Oh sorry. I almost couldn't breathe when I saw the discount they gave me on my first order. Oh sorry. Namaste. Visit 1-800-contacts.com today to save on your first order. 1-800-contacts Experience a membership that backs what you're building with American Express Business Platinum. Get 2 times Membership Rewards points per dollar on eligible purchases and key business categories, as well as on each eligible purchase of $5,000 or more on up to $2 million in eligible purchases per calendar year. American Express Business Platinum there's nothing like it. Terms apply. Learn more@AmericanExpress.com Business Platinum.
- [28:43]
Lee Vincl
That'S a very interesting story and it helps make sense.
- [28:47]
I really noticed and appreciated your kind of historiographical glances on how historians have been telling this story throughout it. And now it makes sense that it kind of like an early version of it was a kind of historiographical examination. And then you've now put in history in there too. So I think it's a nice mix in that way. Do I remember correctly that in a previous position it might even be at the same university you used to teach coding classes and data courses and things like that?
- [29:19]
Thomas Haigh
Yeah, so I've been in University of Wisconsin, Milwaukee since 2004. The first 13 years there were in the School of Information Studies, which.
- [29:31]
Those high schools are in a whole range, and this one is very much at the renamed library school end of things. They had an undergraduate program that was basically teaching vocational IT for students who wanted to do IT jobs and couldn't handle the programming in maths and computer science. I was teaching.
- [29:51]
Systems analysis, project management and sometimes databases, and then.
- [29:59]
Occasionally a.
- [30:00]
Lee Vincl
More.
- [30:02]
Thomas Haigh
Kind of social study of IT and organizations class. So it wasn't really what I would have been doing by choice.
- [30:12]
But I'm happy that I was able to transfer to the history department.
- [30:19]
Lee Vincl
I mean, one thing that stands out about that is that.
- [30:23]
There'S a lot of historians in the kind of STS history of technology worlds who have an undergraduate degree in engineering or some technical field and then go to get a PhD in history. But usually they leave behind those things completely and you are still teaching them in ways that kept you connected to the technical side of things for much longer.
- [30:46]
Thomas Haigh
Yeah, I mean, another thing that kept me connected is the work that I do with the association of Computing Machinery. So you'll see that the Turing Award winners loom fairly large in the book, partly because it's a short book. So talking about the people that the community itself is selected as, you know, being the ones who have achieved the great things or set the intellectual agenda, is a way of keeping the cast of characters more manageable. I edit the Turing Award winners website, where the official biographical profiles and video and oral histories go. And currently I'm also chairing the history committee for the IEEE Computer Society. So in the history of computing community, I'm probably the real historian who has.
- [31:35]
Kept in touch the most with the computer science community.
- [31:40]
Lee Vincl
So as your book says, in a variety of ways, in a variety of places, we're currently in this kind of very hot, bubbly AI in quotation marks moment since the emergence of ChatGPT and generative AI and, and all that. And your book is very much pointed towards this moment and what's been going on. And so I wanted to ask you kind of like what interventions you wanted to make specifically in this moment to remind people or just to really inform people who don't know about this broader history.
- [32:14]
To teach them some things. What are some major interventions you wanted to make today?
- [32:21]
Thomas Haigh
I mean, given the time span of academic publishing, it seems likely, hopefully MIT is going to speed this one on a little bit versus the typical book, but it seems likely that by the time the book comes out, either the robots will have killed us or they won't. So some of what is up in the air, I mean, Ezra Klein just this week said, you know what? I believe it now. AGI going to be here next year and we'll see how that holds up.
- [32:52]
So we're at this moment, I think that is not sustainable either. The hype is going to be delivered on and I have absolutely no expertise to say whether or not that is going to happen.
- [33:07]
Or it won't. And if it won't, I think what's going to happen is something. I mean, the quote at the 1984 conference from which the phrase AI winter comes.
- [33:21]
Is a fascinating paragraph I quoted in the book. But one of the lines is that people are going to start calling whatever they do something else other than AI. And it's not that most of the things that were being branded as AI in the 80s stopped happening, but they found other brands. So it's clear that the current wave of AI branded technologies have got applications and are going to do things and are not going away. The ability to generate plausible customized bullshit for essentially nothing.
- [33:54]
Is not something that is going to be uninvented. Whether or not it's going to translate to technology that can reliably automate high level knowledge workers across the board. Right, different question. But the stuff that is currently branded as AI is going to find applications and have a future. But if the.
- [34:16]
AGI human, like sentience, quickly bootstrapping to godlike intelligence, which renders humanity essentially obsolete, which is.
- [34:26]
To a shocking extent seems to be actually believed by and guiding the decisions of the people running the AI industry, doesn't come true, then I suspect the AI brand is going to be irredeemably tainted and they're going to go back to calling things deep learning or generative tech systems or kind of more prosaic things that don't carry that taint of hubris and failure.
- [34:53]
I don't know which way we're going to go, but what I can say as a historian is the surprising continuity. So in some ways what's defined the AI brand more than anything else for most of its history, although they backed off from this substantially in the 90s and early 2000s, is its connection with this very specific narrative of what we've now called the singularity of the end or transformation of work.
- [35:23]
Of a form of automation that hasn't existed previously.
- [35:29]
And of a technology that's going to be able to do everything mentally that humans can do, but better. And a timeline where.
- [35:40]
People in the field have always argued about the timeline, but where some credible people like Nobel Prize and cheering award winner Herb Simon in 1960 said, within the next decade, as I say in the Treasury, 1970, Marvin Minsky predicted all this great stuff within the next three to eight years. People have consistently been making promises that are essentially identical with what people like Sam Altman are promising today around the AI brand and offering timelines They've been credible, respected, top level people and none of those previous things have come remotely close to coming true. Now that doesn't mean they won't come true this time. In fact, I should revise the introduction to be even kind of clearer on that point. There's nothing in the fact that a prediction has been made every 10 years for the last 70 years and so far it's always been wrong. That means it might not be true this time. But I think being aware of the fact that those exact same promises have been made since 1949 around completely different technologies that were also called artificial intelligence is hopefully going to help people make their own decisions on how compelling they find the current wave of claims.
- [37:01]
Lee Vincl
I hope you're right, man. I think that this is a big part of my work in the last couple of years around hype y stuff and working with people, people like Jeffrey Funk who kind of studies hype waves and things like that. I mean, at least personally, I find studying these kinds of history to be kind of clarifying mentally of just raising the question, like you said, doesn't mean that.
- [37:25]
The kind of classic induction reasoning, black swan stuff, it doesn't mean something's not going to happen just because it's never happened before. But the fact that we've had these waves and it hasn't panned out is good reason to ask critical questions. I think I wanted to kind of ask you. I was at a workshop, I don't know, maybe it was like a year ago and it was on regulating risks around new technologies. And you will not be surprised that that was just coded language for talking about AI the whole time. Basically I found out when I got there and there were some smart folks in the room and one of the guy, John Lindsay, who writes about military technology and it especially in the military context, I think it was him who said something like, you know, every time we use the term AI, we're kind of adding to the mystification and really we're better off talking about specific applications, including for regulatory purposes, you know, and talking about the real risks of specific applications. I wondered if that kind of idea appealed to you given, you know, how you've written around this larger history.
- [38:34]
Thomas Haigh
Yeah, I mean, I've said exactly the same thing. So I get weirdly lots of invitations to Europe.
- [38:43]
Not really. I mean, I've never been asked to speak at STS or history department in the US But I get interesting four invitations a year to go and do things in Europe. And one of the groups there is around the digital humanism project in Vienna around the IWM and the Technical University. And they're getting a bunch of money from a ministry with a very long name that has responsibility, among other things for AI and sustainability. And Austria as part of the EU was involved in drafting this thing. So there were people from the ministry there and there were these discussions and I was saying exactly the same thing that when you talk about AI, it's not a stable thing. So a few years ago when we talked about AI, we were mostly concerned about the dangers of self driving vehicles and systems that were very good at face recognition and so on. That's completely flipped. Now when we talk about AI, it's shorthand essentially for ChatGPT and ChatGPT like generative systems and possibly also images and deepfakes.
- [39:50]
And the issues involved in regulating self driving cars are not the same issues that are involved in regulating deepfake videos. And then there are some other things that maybe are still called AI, like the code that controls the non player characters in video games.
- [40:07]
And we probably aren't as concerned with regulating that.
- [40:10]
Then. For most of the 20th century chess was a central problem in AI. And I mean the International Chess Federation is concerned about that. The whole thing about anal beads as a possible cheating device that you may have heard about. Right. So AI is a big problem for them, but that's not the same AI regulation that the European Union really cares about. So there are very few ways in which trying to approach this at the level of let's come up with regulations for whatever happens to be called AI this week versus let's come up with regulations for specific actual existing deployed technologies that may or may not be termed AI, depending on what it is, that's going to get you attention and career development and venture capital funding at the particular moment.
- [41:02]
Lee Vincl
Well said. So you know, you start us off, you know, the book is chronological, it's a history. And you start us off in the 50s when the term AI is first being coined and taking hold. And you know, you've, we've kind of spelled out your angle, this branding angle and that approach. So when you look back to this kind of origin story moment, what do you see or what do you want to draw out with your kind of branding story? What stands out in that moment?
- [41:34]
Thomas Haigh
So I think that's one of the places where the brand concept is most clearly useful and it's in the first chapter. So hopefully that's going to sell it to people that. John McCarthy, there are three other names on the proposal, but McCarthy is generally credited with having written the main part of it.
- [41:53]
Wants money from a foundation to bring some people together and pay them in the summer to think great thoughts at Dartmouth. So he puts the proposal in 1955, and without explanation, he just drops in. We're going to consider problems in artificial intelligence, and here are some bullet points. Doesn't even say, hey, I just made this up. No one's ever talked about AI before. He just drops it in perfectly naturally.
- [42:21]
To get money. And he succeeds. Doesn't get all the money, but he gets some of the money. It's literally invented to sell a proposal. But in addition, there are other brands around. Brands compete with each other. So that's another part of the brand ness of AI. AI over time has been in competition with different brands in that kind of intellectual, financial, government support kind of marketplaces. And he and Claude Shannon, who were two of the proposers, had done something the previous year and they called it Automata Studies. So this was the second brand they tried. The previous one was Automata Studies. And the problem they had was they wanted like big picture kind of stuff. And there's a specific kind of cellular automata theory piece of the intellectual soup that becomes computer science that's prominent at that point. And he complained that the things that they got were too much on that technical level, not on the bigger level. And the other obvious brand that they could have used that is doing parallel things, although in somewhat different way, is cybernetics. Of course, cybernetics was huge back then, and still in some circles of, I guess, cultural history of science and media studies is the only thing that anyone wants to talk about.
- [43:42]
Lee Vincl
Thank you for saying that.
- [43:43]
Thomas Haigh
Yeah. Which I find sometimes slightly depressing. And.
- [43:50]
Later in life, McCarthy wrote that. Why didn't we call it cybernetics? Well, because I didn't want to have to deal with Norbert Wiener. And if we call it cybernetics, you either have to agree with him or you have to challenge him. And if we call it AI, we can just ignore it. So they wanted. And it's somewhat like that in parts of our thing now, right. I mean, if you want to be famous and successful, you pretty much have to invent a brand like software studies or critical code studies and so on. We've got this sort of big umbrella brand called the History of Computing that people find unsexy and ambitious people be well advised to stay away from. And then the successful ones just invent their own brand and maybe get their own series with mit.
- [44:36]
It's clearly the same thing back then. They don't want the baggage, but particular the fact that the cybernetics brand is owned by somebody else. So they want to invent their own brand.
- [44:48]
Lee Vincl
One of the things you said earlier is there's also these discontinuities between what people mean by AI in these different moments from what we're talking about today or in the 2010s or whatever. So what did intelligence mean in this moment? What is going on during that time that maybe is very different from where we end up?
- [45:08]
Thomas Haigh
Yeah. And by the way, I talk about this part of it if people want to read it in the first of my CACM columns which is called Conjoined Twins and it's about the co evolution of artificial intelligence and computer science. So some of this stuff is already published in that series and they're open access. So if you Google me, they're on my webpage or my CICM author page.
- [45:32]
So intelligence is an interesting term that is never defined. And I follow this through the book. Basically even later when they have AI textbooks, a lot of them say in the first place, well, we're not even going to try and define intelligence. There's a philosophical discourse and we don't want to get into that because we're doing practical things here. But it is an interesting shift. A lot of people assume that the.
- [45:59]
Invention of the AI brand is the first time that people have thought about this question, can computers think? And it really isn't the first book to bring computer technology to a broad audience.
- [46:14]
Giant Brains or Machines that Think by Edmund Berkeley, 1949. And you talk to a lot of old school computer science, they'll be like, this was the book that told me computers existed. It was a best seller, it had multiple editions. You wanted to know what computers were. This was the book in the first half of the 50s, at least that you would find that as a member of the public.
- [46:37]
Again, the title Giant Brains or Machines that Think. So it's coming from a cybernetic place, it's taking seriously.
- [46:45]
You know, I could go on for hours about this, right? But you know, there's McCulloch and Pitts. Everyone knows McCulloch and Pitts. And Neurons. Not as many people know that in the first draft of a report on the EDVAC that lays out the fundamental architecture of the modern computer for the first time, John von Neumann is not talking about digital logic gates, he's talking about neurons. He's not talking about subsystems, he's talking about organs. So he's very much in this kind of cybernetic y kind of discourse, although he doesn't in the end come down on the side of computers being brains. So this isn't just some kind of like, weird popularization that misunderstands it around the modern computer. There's a lot of cybernetic thinking and the interchangeability of humans and non humans and organic and inorganic and brains and control systems is just fundamental to that. So Berkeley believes that any computer that is doing digital program logic type stuff, even an incredibly simple box that he shows you how to build this, flashes two light bulbs in response to a homemade paper tape reader, is thinking.
- [47:55]
So.
- [47:57]
There'S a initially not cranky idea that computers are already brains and that carrying out a computer program is thinking. And what AI says is no, it says computers will be intelligent. Or Turing does this a few years earlier with his famous paper that introduces what we now think of as the Turing Test and brings up the idea that a computer. Well, we can't quite say what intelligence is, but we can answer this other thing, which is not quite equivalent, but could be substituted, which is this whole idea of if a computer can pretend to be human, fool people, then it's intelligent what they come up with. If a computer can do certain creative things which are gestured towards in the report, like proving theorems, composing musical works, or playing chess, then that will be artificial intelligence. And by implication, if it does great things with numerical mathematics and cranks out numbers faster than any human, which computers were already doing in, like 1945.
- [49:03]
So depending on what you count as intelligence, it could be that we're still ahead of computers, or it could be that we fell hopelessly behind in 1945 and have just been losing ground ever since. And they come up with this kind of slightly weird set of things.
- [49:20]
And why do they choose theorem proving, such as a critique even within the AI field. Rodney Brooks, who goes on to head the MIT lab and advocate for favor of building robots that crawl around the floor and experience the world that way, and invents the Roomba, you know.
- [49:41]
Vacuum machines.
- [49:44]
And has explicitly said this so you don't even have to be like Nessie as Right. It's a very cultural and historically specific idea of intelligence. So the people coming up with it are.
- [49:57]
The kind of first generation to come up inside the emerging military industrial complex. They're disproportionately Jewish. They've gone to elite, you know, elite institution after elite institution proved themselves by being very smart.
- [50:14]
So they prized their own intelligence. Intelligence is central to their identity. And they also have that kind of mathy, sciencey, way of looking at the world that are like, ranking other people's intelligence and know, oh, this guy over here is the real genius, and so on, and the things that they set each other apart. Right. So they like this idea of being virtuosos and being able to do cultural stuff, too. So music is on the list. Theorem proving is obviously something that the smartest people are best at. And playing chess. Right. There's a quote from Minsky that I recently put in the book as I was expanding it in that first chapter.
- [50:49]
That he went and did his PhD in the mathematics department at Princeton, which at the. I think was the leading place for that in the world. And at least the way he described it, they pretty much just got money. They faked the courses on the transcript, they sat around, played a lot of chess, shot the breeze with the other geniuses.
- [51:07]
And that was the culture of the department. Is that what the department was really like? I have no idea. I'm not a historian of mathematics, but the fact that that is how Minsky characterized the ideal academic environment and the centrality of games to that. So I think in retrospect, we can see that there's kind of a very specific, post war meritocratic, somewhat Jewish kind of idea of intelligence and its importance that is being baked into the field of AI. And of course, the Jewish aspect is probably the place where I am skating on the thinnest ice and I'm writing very carefully. But, you know, like.
- [51:49]
Especially as none of. While most of the leading figures and pretty much all, I think, of the leading 20th century critics of AI had a Jewish background, I don't think any of them were believed in God. So it's kind of the cultural Jewish thing.
- [52:07]
Lee Vincl
Right.
- [52:08]
Thomas Haigh
We talk about that when we talk about American humor in the period. Right?
- [52:11]
Lee Vincl
Yeah, yeah.
- [52:11]
Thomas Haigh
You know, Mel Brooks, Woody Allen, et cetera.
- [52:15]
Lee Vincl
Well, high culture generally during that period in a lot of ways.
- [52:18]
Thomas Haigh
Yeah, absolutely. And the novelists and so on. I'm not aware of anyone who has explicitly said that about AI.
- [52:28]
So I don't want to go too far out on a limb there, but I think there are elements about the traditions, like answering a question with a question, not deferring to traditional wisdom, not having a centralized authority and the whole kind of insider, outsider element of bourgeois Jews in the immediate post war years that I think can meaningfully be connected to the history of AI And I hope.
- [52:57]
While I can't dig deep into that in the book, I hope at least flagging it might encourage other people to take that element seriously.
- [53:06]
Lee Vincl
So One of the things you also write about, and you touched on this before, is kind of the institutionalization of AI. And, and you really set it up nicely when you were talking about the new history of Modern computing book with Ciruzi, that that was a kind of a business history approach. And so it wasn't as much about the CS kind of as a field, but this like AI, even though you point out in several places in the book, its applications don't actually get applied in the real world in so many examples, and yet it becomes very central to the field of CS as a self identity. So what's going on there?
- [53:47]
Thomas Haigh
Yeah, and I think this is maybe the biggest thing, that there has been something of a rush of people into the field of AI. I myself have essentially been accused of being a carpet bagger who's going to come in and steal from the grad students who've written their PhDs on this. And I hope it's apparent to people that this kind of synthetic short book is not something that a grad student would write, you know, as their dissertation book. That it's. I'm very much trying to highlight the fact that it's now becoming possible to write this kind of synthetic history only because there are dissertations and other things appearing. I'm trying to, you know, foreground the names of other scholars in the text and so on. Not to be accused of like coming and you know, grabbing the glory. But I think the thing that is most obviously missing in most of the work that is being branded as history of AI at the moment is an awareness of the fact that AI in the 20th century existed almost entirely within the emerging discipline of computer science.
- [54:57]
Because the history of computer science has not really been written. And I've spent some time talking to people and kind of asking why this is. Because from my viewpoint, so modern computing, it's not, not even primarily business history, but it's a history of technology. Because the computer becomes so many different things. The common thread through it that lets you link one chapter to another is essentially this particular core affordance of modern computer systems is first developed for military control systems, or first developed for arpanet, or it's first developed in personal computing. Wherever this stack of technologies is, what moves forward from one, one chapter on one thing the computer becomes to another. So it's not even that it's the business history, but it's the history of the technology that get deployed.
- [55:48]
Digging in some ways quite deep into the affordances and the architectures of computing. But.
- [55:56]
I mean, back in the 90s business historians and historians of technology were attempting to produce these big picture comprehensive histories based on a somewhat underdeveloped secondary literature. And, and I'm not saying those fields did an amazing job. But the computer book by Campbell Kelly and Asprey, Paul Sarouz's original history, Martin Campbell Kelly's book on the history of software industry, more recently Jeff Yost's book on the history of computer services and other works, were trying to do this big picture thing. And my view of the world is we've got this kind of. So I'm very much a historian of technology rather than a historian of science. I've never been to the History of Science Society meeting except when it was co located with shot. I've published in isis.
- [56:41]
Lee Vincl
Amen, brother. I'm with you.
- [56:43]
Thomas Haigh
I always had this kind of separate but equal sense that we did the technology, they did the science. And it was a bit perplexing to me that.
- [56:51]
Essentially nobody.
- [56:54]
Has written a book that is squarely the history of computer science. That would tell you the kind of plodding big picture stuff about where computer science came from, how it changed over time, what were the leading subfields, what were the leading departments.
- [57:12]
And even there's essentially no one who would describe themselves as a historian of computer science like in the world, basically. Right, Maybe that's a slight exception. So there's a lot of people who've done work that is framed in a different way and intersects with a piece of the history of computer science. And that certainly sheds light on it. I've got.
- [57:33]
In a recent edited volume by Bill Asprey where I wrote about the history of modern computing and some other things. It's a historiographical essays collection around the history of computing and information. I've got a paragraph with a lot of footnotes where I call out people like Joe November and Stephanie Dick, you know, whose work like fills in pieces of this, but it's all like kind of cutting through a little piece of the history of computer science in the interest of a project that is framed differently. So what I realized was as someone trained in computer science who kind of keeps a foot in the field and has a general sense of what it is. Even though I'm not a historian of science, I am more equipped to ground the story of AI as a part of the history of computer science than any of the actual historians of science because for some reason, and I had some reasons for that. So one of them, I think is this whole cultural turn.
- [58:33]
Another thing I think is that Ever since Kuhn. The history of science, the thing that you do to succeed in the history of science is to take. And this was actually pretty true of history in general when I was in grad school. Right. You take some received like oversimplified over broad master narrative and you break it down and problematize it and show it's more complicated and more contingent and more local. And if you're a field like the history of physics that people had already written the boring plotting histories back when that was a thing that you could do and have a career, then they had something to work with. But I think they've never really known what to do with computer science because they're trying to do that. But computer scientists haven't been doing this either. There's very little internalist history of computer science. So there is, you know, how do you do that? You make up some received narrative.
- [59:26]
It's kind of hard to do. And I've also been told that it's seen as old fashioned in the history of science to define yourself as a historian of discipline X. Right. Because everyone should be doing this work that flits around between time periods and crosses boundaries and so on, which is great. But if nobody is calling themselves a historian of computer science.
- [59:49]
How are people in the history of science supposed to know what computer science was and where it came from and so on. And there has been work.
- [59:58]
Particularly in Europe. I don't offend anyone. Good work around things like Algol and so on. In the 50s, in the period basically immediately before computer science comes into institutional existence. But for the period in the 60s and later where computer science actually exists.
- [60:18]
If you wanted a book for your orals list or something where you could go and read what, what's the history of computer science? There just isn't one.
- [60:29]
Lee Vincl
Part of the institutionalization story that maybe has been handed down or in the literature is about the relationship between the field and.
- [60:39]
Military patronage, basically. And you had some very to me funny stuff in here about the kind of received STS literature on computers and the military and this kind of Langdon winner esque vision that like military values are being built into these things. And you had, you had this wonderful line I pulled out. You're right. Personally though, I see the founders of AI as less as militaristic imperialists and more as inspired boondogglers who diverted a few buckets of money from a tsunami of Cold war spending to advance their quirky personal obsessions. And I just love this as an image. I think that's true in a lot of Sciences during that period, actually, that there was so much money around that you could divert it. So just tell us a bit about that kind of counter picture you're painting there of what getting Cold War money means for a field like computer science.
- [61:38]
Thomas Haigh
Yeah, so one of the interesting things to me is the three centers for AI. So four of the people who are at the famous Dartmouth meeting later become known as the founders of AI. They all win Turing Awards in the first decade of Turing awards in the 60s and early 70s, and four of the first 11 Turing Awards for computer science go to AI people. So it's kind of the intellectual high ground of the discipline as they're trying to establish themselves.
- [62:14]
By having what they hope will become a prestigious award, which it eventually does.
- [62:20]
And they found the programs at Carnegie Mellon, MIT and Stanford.
- [62:26]
Now if you look at the present day rankings of computer science departments, those are still the top three, not just for AI, but for computer science in general in the world. So that's probably not a coincidence. And AI.
- [62:42]
I think was an important part of building up those departments to preeminence. But at least two of the names on that list. Right. So Carnegie Mellon is an up and coming institution this period that gets in the top rank. And it does it in large part driven by computer science and AI and robotics, but MIT and Stanford. Right. So Bill Leslie's book, what is it? Science, Science and Cold War. Cold War, it's something like that.
- [63:11]
That's the title. But the actual book is MIT and Stanford. And his argument is those are like two unique top tier institutions that get more federal money more easily across more things and invent a new kind of university in the early Cold War period.
- [63:29]
Those are two leading centers for AI. And if you look at the later accounts and how it would be good to back this up with archival evidence, but I can't dig in too deep for this kind of book. But just the way the actors themselves describe it is that they were setting the thing up with a handshake that there was a lot of money already at mit. They meet Jerome Wisner one day by chance describes the project. He's yeah, here, have some of my money, I've got so much money, go have a lab. And then ARPA is giving out the money in this kind of rather insider dealy kind of way that people essentially from MIT take it in turns to go to ARPA and shower money on their friends. And then they step back and the next guy goes, and he gives them money. Which is not the kind of way the NSF and other things do things. On the other hand, it was spectacularly successful in fostering the development of computer networking, graphics, interactive stuff and so on. So I don't want to of like. But.
- [64:34]
But certainly in those privileged institutions they have easy access to significant amounts of federal money.
- [64:43]
From ARPA in particular in a way that other institutions don't. And they have computers lying around and they have kind of infrastructure. So I think you absolutely can't understand what's going on here without talking about the Cold War context. And even in Paul Edwards like wonderful, the Cold World, there's kind of an element and I think this goes with this 80s and later kind of maybe Foucaultiany discoursey thing that kind of things seep into things like by osmosis. So there's an original sin kind of element there almost and in David Noble as well, particularly forces of production. Right. That if the money is coming from the military than like what comes out of it is, you know, tainted and somehow embeds those values.
- [65:35]
And that of course is a critique that's easy to make. And you see in things like Yardenkatz's, you know, artificial whiteness and a few other things.
- [65:43]
Somewhat in Johnny Penn's dissertation, which I like, but he's kind of grappling with that and he's. And I think like.
- [65:54]
In the recent past.
- [65:57]
Condemning militarism and the military industrial complex and so on has been kind of an easy and obvious move for an aspiring scholar. You will not find many people in humanities and media studies type things who are going to stand up for the military industrial complex.
- [66:19]
I'm not going to do that either. But I do want to de exceptionalize computing a bit because at this point.
- [66:30]
Pretty much.
- [66:33]
I quote the figures from Stuart Leslie's book. But military funding is so important across.
- [66:42]
Science and technology in lots of fields.
- [66:45]
Lee Vincl
Exactly.
- [66:45]
Thomas Haigh
It's not that AI is in any way unusual or exceptional.
- [66:50]
From this viewpoint and particularly in computing because the NSF doesn't have a computer science specific directorate until the 70s.
- [67:01]
It's not just AI, it's also operating systems and graphics. And of course everyone knows about the ARPANET and pretty much computing in general and most of the other tech. And if you look at the number of zeros attached, the money that goes to AI in this period.
- [67:18]
While setting up a pretty cushy life for a small number of people, is missing many, many, many zeros compared with things like the Sage Aircraft Defense Network or the nuclear submarines or the missiles or the jet fighters. So it really is like a few drops from this massive stream of Cold War funding.
- [67:43]
And if you look at the people involved, I mean, John McCarthy at this point was still a communist, although he turns into a right wing crank later in his career.
- [67:54]
The famous hacker culture that grows out of mit, as Stuart Levy writes about in Hackers, was determinedly apolitical. So they really don't understand why the square outside the building that they're on the sixth floor of is full of protesters who've got something against military funding in Vietnam and they install barriers and security keys and they continue to go up and do their kind of geeky hacking. And if you're a hardcore person, you'd say, well, it doesn't matter, they're still stooges in the project of American imperialism. But given that none of the stuff they produce does anything useful.
- [68:33]
I want to backtrack on that. The stuff that they produce to advance the mission that they say that they're doing of making intelligent systems doesn't produce anything useful, but they do create important things that essentially become the infrastructure of computing as byproducts on that mission. They're being useful to the military and everybody else. If you could somehow.
- [68:57]
Total all that up and do a cost benefit analysis, I suspect it was a pretty great return on the investment. But if you look at in terms of doing the things that they promised they were going to do as their direct goals.
- [69:11]
Not a lot of that happened in the 50s, 60s, 70s.
- [69:16]
Sponsor/Ad Voice
Tito's handmade vodka is America's favorite vodka for a reason. From the first legal distillery in Texas, Tito's is six times distilled till it's just right and naturally gluten free, making it a high quality spirit that mixes with just about anything from the smoothest martinis to the best Bloody Marys. Tito's is known for giving back, teaming up with non product profits to serve its communities and do good for dogs. Make your next cocktail with Tito's, distilled and bottled by 5th Generation Inc. Austin, Texas. 40 alcohol by volume saver, responsibly deck your homewithlines.com.
- [69:52]
Thomas Haigh
Diy or let us install.
- [69:58]
Free design consultation.
- [70:03]
Lee Vincl
Plus free samples and free shipping.
- [70:07]
Thomas Haigh
Head to blinds.com now for up to 50% off site wide plus a free professional measure.
- [70:13]
Rules and restrictions may apply.
- [70:16]
Lee Vincl
Well, I mean, even though it's a short book, there's a lot in there that we're not going to have a chance to get to all of it by any stretch of the imagination. But there's two more things I wanted to briefly chat with you about the first is, you mentioned this earlier is the kind of expert systems moment in the 80s and the coming of the first AI winner. I don't know, I don't know if there's been multiple at this point or at least two, something like that. And this was my friend, the science journalist John Horgan in the 2010s when it was like the machine learning bubble was going. He pointed me to some publications from the 80s of that were expert systems publications where people were, you know, they were promising technological unemployment and the destruction of the professions and all this kind of stuff. And so, you know, coming for our moment, what do you, what do you see in this kind of earlier.
- [71:12]
You know, this, this, this expert systems moment. And also, you know, how do you understand the, the AI winner today?
- [71:22]
Thomas Haigh
Sure.
- [71:24]
So I'll take the first one of those first. Right. The expert systems thing. Expert systems is interesting because it's a different brand. If you look at the Google Ngram, which is very crude, there even seems to be a moment where expert system shoots up from nowhere, is briefly discussed more than AI itself, which is also shot up massively in the 80s and then plummets away. And unlike the AI brand, expert systems does not make a comeback.
- [71:52]
On the other hand, though, expert systems in some ways is a retreat from this dream of what we now call AGI. So in the 60s they focus mostly on the idea of automating, finding great ways to automate reasoning. In the 70s, the big challenge is structuring knowledge representation.
- [72:14]
And some of the people, a lot of the people doing that are still kind of dreaming about that idea of the general purpose AGI type thing, although it doesn't seem to be getting any closer.
- [72:28]
And then expert systems is the idea that we don't need to solve that problem. What we can do is.
- [72:40]
Collaborate with experts. And that's kind of interesting. This comes out particularly from a guy called Edward Feigenbaum who trains at Carnegie Mellon with Herb Simon and goes to Stanford. So there's a lot of that with these three main centers. People train in one go to another. I talk about the way that they really retain the power to shape this brand. But he's also an intellectual entrepreneur and he gets the expert systems brand. He's saying, don't think about these big picture things. The point is this technology is here now. It's economically viable. It turns out.
- [73:17]
That automating what experts do is much easier than automatically automating common sense, daily life stuff. So if all you want to do is replace someone who is diagnosing a blood infection or who is interpreting data from like Schlumberger style, like oil.
- [73:41]
Well prospecting and so on, that you can do that relatively easily by having an expert talk to them, turn their knowledge into rules, put those rules into a fairly simple computer program, have it interpret the data, and initially it's going to get a lot of stuff wrong. And then you go back to the expert and they say, oh yeah, I forgot to mention that you don't do that in this circumstance. So you get back and forth and after some months and a few hundred rules you've got a system that they can claim that you do. Kind of like a blind test. Here's what the system says, here's what a panel of experts say, and it'll outperform the experts. And that's a very powerful pitch. And an industry springs up around this idea that companies are all going to need to set up their internal groups and hire AI people to build expert systems. Companies selling expert systems products and services. Boom.
- [74:36]
That is the moment where an AI brand for the first time is selling not just to funding agencies and graduate students and university administrators, but to venture capitalists incorporated as something they need to set up. Although interestingly, it's distancing itself from the AGI kind of mission in a way that actually AI textbooks in the 70s and 80s are also doing so. Another variant term that Feigenbaum I think also invents or promotes is knowledge based systems. So the idea is, so you finish up with a move away from the AGI hype actually and towards the this stuff is economically valuable right now, regardless of whether we ever get to generalized intelligence or not, because experts are scarce and expensive and they retire. But if we can put their knowledge into a box and deploy it widely, then that's going to be a great thing. So it's particularly interesting both in terms of how the expert system brand and the AI brand interact with each other and in terms of this being the previous moment where there's a.
- [75:52]
The AI thing spreads beyond the laboratory and into venture capitalists and startups and businesses. All right, now the other part of that was the AI winter question. So.
- [76:07]
There'S a lot of interest in AI history at the moment and.
- [76:15]
There'S also kind of a lot of bottom feeding content on the way web. So you Google kind of AI history, you're going to find a lot of these kind of unsourced things, et cetera, the Wikipedia page and so on, and everyone is going to tell you that there's been a Whole series of AI winters that it's going to up and down and up and down, you'll find kind of graphs illustrating that. And because people have not generally been looking at the big picture thing of this, this also finds its way into scholarly articles as kind of the assumed existence because people just use that to frame the specific projects that they actually researched. And one of the ide is that there is a so called first AI winter in the 70s.
- [76:54]
Lee Vincl
Yeah, you really nicely debunked that in the book. I really appreciated your data analysis in that part.
- [77:00]
Thomas Haigh
Right. I'm writing this kind of synthetic thing. I'm not going really deep, but I mean also I think historians have just become terrified of numbers. It's not just that people don't do regression analysis and put the census data onto punch cards anymore, but like even numbers, percentages, is there more or less of this thing?
- [77:22]
Lee Vincl
So.
- [77:25]
Thomas Haigh
Two obvious places to look were membership of the main interest group for AI, which rose rapidly over the 70s and showed absolutely no slowdown. Google Ngram for occurrences of the term AI. And then you can also kind of look a bit at other people who were in the field, in the area who were actually saying firsthand, like, I found I couldn't get money anymore, it was dreadful, I had to give up my lab or something. Also, the discourse has so much been around these elite institutions where everybody who becomes prominent in the field trains. And by the way, there's a complete discontinuity between that and the current wave of AI branded technologies. Those are associated with different institutions because the neural network stuff was kicked out of AI because it didn't fit well with computer science. But.
- [78:18]
If you kind of go to Europe, say, and look at where I didn't even put this in the book yet. But you look at what become the leading sensors for AI, they tend to get founded in the 70s. There is absolutely no evidence that I've seen whatsoever of a broad based slowdown. There was the shift from darpa, from ARPA to darpa. The people at MIT and Stanford had to start justifying what they were doing somewhat more and having actually proposals that promise to do things.
- [78:49]
Et cetera. So for them, maybe it was a bit of a shock, but to go from there to there was some kind of asteroid that hit the field and wiped out most of AI for a few years in the late 70s. Maybe I'm missing something, but it would be nice if anyone out there who does think there was a first AI winter in the 70s can find some actual data beyond just this guy who was from mit, wrote a book where he said that someone who'd been at MIT at the period mentioned to him that this happened. And that's pretty much where the entire idea of a 70s AI winter is coming from. The US in the UK, Edinburgh was the big center and I think they did get their funding cut back genuinely, but it did not completely wipe out AI, even in the uk, where there may have been more of like an actual AI winter. And by the way, the other thing I find compelling here, 1984, where the term AI winter is introduced, whole panel discussion, like a dozen pages, multiple presenters. They're concerned about a lot of things that have a lot in common with the field now, like that.
- [80:00]
The people in the field are being poached by companies giving huge salaries. So the pipeline of new researchers is drying up. People who have very vague and shoddy credentials are able to get jobs leading AI groups because there aren't enough real trained people.
- [80:18]
Companies are promising too much. The hype is getting out of control. It's interesting that a lot of the things they're talking about in 84, where they're worried that their academic conference is turning into a trade show.
- [80:30]
Are both well grounded and have a lot of commonality with the present moment, but we're now doing it on a much huge scale. But one of the things they don't say in that session is, wouldn't it be terrible if we had another AI winter like the one that we just lived through a few years ago? Right. Nobody there says anything to suggest that something like what they're talking about has recently happened. Which is strange, right? If there really was a big AI slowdown in the 70s.
- [80:59]
Across the board, not just in specific things like machine translation, which for the most part wasn't even branded as AI anyway. It was kind of its own identity. Identity.
- [81:09]
It's kind of strange that none of them mentioned that.
- [81:12]
Lee Vincl
So, I mean, one of the things that has driven me nuts about the generative AI bubble is I started writing about bubbles and thinking about the sociology and economic bubbles almost a decade now ago. And it was in that earlier 2010s moment of like self driving cars, delivery drones, the sharing economy.
- [81:37]
And machine learning was very much a part of that. Right? I mean, I remember like books like the Rise of the Robots and the Second Machine Age, and not only in the kind of discourse and also in kind of like startup land, there was a lot of bubbly stuff going on.
- [81:55]
But also, you know, a friend of mine went and gave a historian friend went and gave a Talk about AI story stuff at. Or just technology stuff at a major beverage company in the United States, and they had just spent like $10 million on machine learning applications and got apparently nothing out of it at all. They were telling him that they just wasted a bunch of money. And so, you know, the generative AI moment is just like, I'm just like, I need to get over it and just assume that there's no such thing as collective learning and we don't learn from experience. And like the next bubble comes along and people just go for it, but I just feel like we just live through this kind of bubble. Ten, you know, less than a decade ago that just the air came out in like 20, 23 of that thing or something like that. You know, it wasn't even that long ago. So. Yeah, yeah. I just wonder, like, you know, like, you have a. I want to tell listeners that you have a really beautiful and to my knowledge, the best treatment of the history of machine learning and neural networks and that history that I've seen so far. But what do the 2000 and tens kind of look like to you in this project?
- [83:02]
Thomas Haigh
Yeah, so first, I don't want to claim credit for that. Right. I'm drawing on dissertations and publications, Zhao Qiang Li.
- [83:13]
Some other scholars who are coming up. And because people are so focused on that machine learning, data history, modern neural net stuff, I kind of feel like. Like I need to deal with it, but I don't even need to understand it on a technical level the same way I understand the earlier stuff reasonably on a technical level. I mean, I got the general idea back, Propagation algorithm, ways of doing it better, different classes, something called convolutional. But I mean, like, probably literally 100 people pitching books to MIT right now where they explain that stuff. So I need to have it there enough to say what was different about the old stuff. So I don't want to claim. I don't want to claim to be the history of machine learning guy, but yeah, that's another continuity. So I mentioned the continuity to the beginning that these same claims about the utopian and or possibly dystopian revolutionary singularity moment that's coming for us have been made around computers since 1949 at least.
- [84:16]
And consistently updated thereafter. And yet the other thing is, if you look at the last few things that Silicon Valley got really excited about, so We've got the VR augmented reality, which enormous bust. We've got the Web 3.0 DeFi blockchain thing.
- [84:38]
And certainly Trump is doing his best for that stuff. But I don't think NFTs are coming back. Right? I mean, that was all just insane. Huge flop. You've got the Alexa voice assistant stuff that they wrote off billions for and have cut back on, which is kind of essentially AI. You've got the Amazon stores that got closed down because they never got to the point where the computers could actually do the work. So there were people watching you on a video camera and trying to make out what you'd picked up. And there's a self driving car stuff where.
- [85:17]
If you've got enormous big tech monopoly money to keep you in business like Waymo has, maybe someone's going to break through with that. But GM just gave up on it. Most of the other companies have given up on it.
- [85:28]
Lee Vincl
Uber Lyft gave up on it, on.
- [85:31]
Thomas Haigh
The idea of an economic robo taxi thing because it turns out you need more people.
- [85:37]
In the control center keeping an eye on the car and keeping the thing serviced than you do drivers. And the economic case is very far away from existing. So it's like overhyped flop after overhype flop in this space and in adjacent spaces. And.
- [85:54]
That does not seem to be leading to any kind of skepticism. There's a phrase I heard that I kind of like, that venture capitalists are lemmings crushed with sheep.
- [86:08]
That I think describes some of this kind of group things that happens around the next big thing. And of course we can't lose track of the fact that this is happening because you have these gigantically profitable monopoly businesses that are looking for things to do with all their spare capital and are terrified that one of these things is actually going to pan out and they might face some real competition and they're spending vast amounts of money to suppress it.
- [86:35]
I'm not an apologist for big tech in any way. Here again, disclaimer. Maybe this one will work. Right, Right.
- [86:47]
On a meta level, the venture capitalist idea that you fund 20 things and 15 of them are going to go bust, three of them going to have mediocre returns, and maybe one of them is Google and it pays for all the others. Maybe Silicon Valley.
- [87:03]
Needs to throw itself off a cliff 15 times and the 16th time they're going to lock up the technology that is going to dominate the world for the rest of history. So it was still a pretty good investment. Investment.
- [87:17]
Lee Vincl
Yeah. So I mean, your last chapter is called, you know, thanks to Chat. It's called thanks to Chatbots, AI Finally Conquers the World. And yeah, I mean, maybe you just said what your take on that thing, you know. But it is striking to me, the kind of deeper history of chatbots where, you know, I'm very open to, you talk about productivity and, you know, organizational change around these things. In that chapter, I'm open to there being changes.
- [87:49]
As a result of people adopting these things that I can't see yet. And in fact, I really try to very closely listen to and watch people adopting these things in ways that I didn't understand. You know, there's a lot of people in my Life who use ChatGPT for all kinds of shit, you know, in ways that I wouldn't because, you know, it just doesn't make sense for me. But, but it, but it, you know, my initial reaction to this technology was also like when the, that first initial ChatGPT moment was like, this seems like not really new to me and in fact, like an expression of these older, deeper chatbot technologies that have been around. So, I mean, where are you at with these things right now? And you're very. You say more than once in the book that you're a historian and you're not going to like, do prophecy other than to say, like, probably, you know, some people will probably lose a lot of money off this because that's happened to every major technology going back to the railroad. Yeah. So, I mean, where are you at with these things these days?
- [88:49]
Thomas Haigh
I don't personally use them.
- [88:56]
Most of what I know about them honestly comes from listening to the Hard Fork podcast, because that's kind of once a week, and those guys bring everyone on. You get a sense of how the people in the industry are thinking about things.
- [89:12]
It seems from what I hear secondhand, with absolutely no added value, that the thing that I would have expected to happen is happening, that things just don't scale indefinitely. So the word is that ChatGPT 4.5 uses massively more resources to achieve.
- [89:38]
Very slightly better results.
- [89:41]
Lee Vincl
And folks like Kary Marcus have been predicting that for over, and Steve Jobs too, for over a year and a half or something like that.
- [89:48]
Thomas Haigh
On the other hand, the excitement is around something, and this is another sub brand.
- [89:57]
Large language models working by predicting text statistically seem to me an unlikely foundation for genuine intelligence. Although with the proviso, I don't think people still really have a great handle on how brains actually work. So who knows? Maybe.
- [90:17]
I don't claim to be a cognitive scientist. Ask other people about that. Now, reasoning models, I think that's acknowledging the fact that larger language models aren't reasoning, they're trying to. As I understand it, have some kind of loop where they send something back to critique itself and try and do things more step by step, step.
- [90:38]
And those are supposed to be achieving compelling results using vast computational power on kinds of tests that ChatGPT wasn't good at.
- [90:51]
So that's what I know about the current situation, but I've never used one of these things. It does seem that the clearest evidence of them having an actually copy economic value so far is in programming. And that makes sense because computer code is so much more structured than natural language.
- [91:12]
But look, I think one of the other important things to remember is, and I know that you've had people on the show talking about this, that there is absolutely nothing new about automation.
- [91:23]
Lee Vincl
Yeah, exactly.
- [91:24]
Thomas Haigh
And that in many ways it predates the concept of automation. Right. I mean, I, I'm still partial to the classic idea that it's about substituting capital for labor. And if you go back, I start my History of Capitalism class with a week on colonial period and mercantilism and so on. And at the time of the American revolution, something like 90% of the population in the relevant colonies is working in agriculture. Some of them enslaved, some of them free, but 90% doing agriculture. And at this point, I believe we're down to something like 1.6% of the US population working in agriculture. So technological change, eliminating jobs on a massive scale is not really a new thing. And even white collar jobs being automated. And if you go back the 80s, there was one of those waves of concern about that Shoshana Zubov's in the Age of the Smart Machine.
- [92:27]
And the truth is, a lot of that happened. There's so much less.
- [92:32]
Employment doing routine clerical processes. We do our own data entry. Humans aren't routinely approving loans and credit cards anymore. Vast swathes of white collar unemployment have been automated. I think to a large extent. What's different here is suddenly this is the technology where people who make podcasts or write opinion columns or talk and express themselves for a living.
- [93:01]
Saying like.
- [93:03]
Oh my God, like the face eating leopards have come for me. Right.
- [93:10]
That it's much more concerning to the people who do that kind of work. So that they view this as some kind of unprecedented historical rupture. Whereas if anything, maybe it's. It's just more of the same that's been happening to other kinds of workers for centuries.
- [93:33]
Lee Vincl
So you're revising this book and you've still got some work to do.
- [93:39]
Do you know what's next for you?
- [93:40]
Thomas Haigh
Or.
- [93:40]
Lee Vincl
I mean, you still are so into this project that like Horizons are, you know, are this.
- [93:47]
Thomas Haigh
No, I actually have another. Another thing which unfortunately has been suffering from trying to do two things at the same time. So apologies to any contributors out there there who are listening to this. Something coming out of the Seagan project, Historicizing Digitality, which from many different viewpoints, which is a hybrid of a co authored volume and an edited volume. I weirdly have not published my dissertation which. Okay, yeah, yeah, I noticed that before I die. Although it's going to be a different thing at this point.
- [94:17]
Lee Vincl
Yeah, sure.
- [94:19]
Thomas Haigh
A book in that space.
- [94:21]
Lee Vincl
And what was it on?
- [94:23]
Thomas Haigh
It's on the institutional and professional aspects of the management of information systems in the 20th century US corporation. So it's kind of way back to this technology institution.
- [94:37]
Labor, looking at groups who've claimed essentially to have expertise in information processing. Starting out with scientific office managers in the 20s and then data processing management, information systems, chief information officers, Joanne Yates type stuff.
- [94:52]
Lee Vincl
Yeah.
- [94:54]
Thomas Haigh
Someone with rather more attention to professional identity.
- [95:01]
Lee Vincl
Okay, sure.
- [95:02]
Thomas Haigh
And the kind of like interest and agency of the groups of specialist workers involved. Right. Because Yates, it's a great book, but it's got a kind of top down rationalist view viewpoint. And I was trying to come more from, hey, let's do the labor history of these professional and technical groups. And I think still essentially someone that pretty much hasn't been done. So I'd still like to do that.
- [95:31]
So yeah, I got a bunch of other things that I should have been also doing. So my advice is don't try to do two books at the same time time.
- [95:42]
If you've got anything beyond what you said to say about what you liked or found interesting in the book because you actually I don't teach the history of technology. I haven't so much been keeping up with the broader literature. I do teach the history of capitalism.
- [95:57]
So from that viewpoint of someone who is more immersed than me in current kind of history of technology and STS things, what do you think the book is doing that maybe other things on doing in the same way?
- [96:11]
Lee Vincl
Yeah, well, I mean for me I think it was. I mean this book is right up my alley in lots of ways in the framing and I think the branding.
- [96:21]
Angle really just did a lot of mental work for me that I've been looking for. And it goes back to that thing that I said about the John. I think it was John Lindsay who said that thing. Every time we use the term AI, we're mystifying things and really we should be looking at specific applications. And I think what you gave me was a nice history that showed how this term had kind of morphed and been applied to different things over time.
- [96:49]
And what's interesting about that is its success in sustaining and keeping coming back where so many of these other terms fall away. Right. I mean, I love young Seob Choi and Cyrus Modi and maybe it's just the two of them have this paper about how. How just as you were saying and you were pointing to that 84 conference in some ways that, you know, basically nanotechnology was just people using, applying this term to shit that had been like material science and other things because it was the hot term to apply to things and it was just redescription. And so I think we have a kind of sociological picture of that out there. But I think you've done a really nice job.
- [97:34]
Of, you know, of showing this weird term that just kind of sticks around and just keeps coming back for a variety of reasons. And I also just think I laughed out loud several times while reading this book. I think it's very funny and I think you really take the piss out of like just so many things throughout it.
- [97:53]
And you know, you have this very nice picture of like, you know, you say some interesting things in the last chapter about. Well, you have this funny chart I was going to talk to you about, but we're kind of out of time that was kind of comparing 20th and 21st century AI hype and these moments. And.
- [98:13]
You point out that a lot of the kind of critics out there of AI and big tech and the kind of tech clash literature these days, these criticism of bias and it's got again, this Winerian, all these bad values have been built into these systems and throughout it you're just showing that, you know, so many of these systems are not being applied in the real world. If we keep a kind of business history focused. Sure, there's spinoffs, they come to lay the foundations for things, there's effects out there, but they're not really being applied as systems in this way. So, you know, it raises questions about, you know, about the way we're talking about bias. So I just. Yeah, I mean, this book is directly up my alley in so many ways with my own concerns recently that I'm your perfect target market.
- [99:01]
Thomas Haigh
Great. Well, yeah, thanks, Lee. And one of the things I hope I can do with this being, as I said, a short history of AI in both senses. So a lot of people are interested in doing long histories of AI where they're like, the history of AI is really the history of statistics or bias algorithms or whatever. Alvier or the degradation of work or capitalism or whatever. And the thing is, the AI brand has been applied to so many different things that essentially to write any long history, you're taking some particular thing that AI has been to someone over the last 70 years, and you're then taking that analytically and you're saying, well, this is the real AI, and actually it doesn't start in 1955, it starts in ancient Greece or whatever.
- [99:50]
There's nothing wrong about doing that. I don't want in any way to sound like I'm trying to delegitimate that kind of work. And obviously scholars come from different backgrounds and it's natural and normal for them to connect AI to the things that they have expertise in. But I hope that at least the existence of this book is going to help people who are doing that to kind of see where the kind of AI that they're taking as their analytical definition to construct a particular long history fits within the, you know, the short history of AI as a brand that is being constantly redefined and mobilized by different actors to do different kinds of work.
- [100:31]
Lee Vincl
Ah, man, I couldn't have said it better. Thanks so much, Tom. This has been a lot of fun and I really do love this book and so thank you. Thank you for coming on.
- [100:39]
Thomas Haigh
Oh, thanks again. It's been a pleasure. Leap.
- [100:48]
Lee Vincl
I hope you enjoyed this episode of our podcast. You can reach us with questions, comments and suggestions@leevinselmail.com or by following me on Twitter tsnews or on YouTube @peoplesthings. Our podcast is distributed by the New Books Network, the leading platform for academic podcasts. So that you can find us wherever you get your podcasts. Peoples and things, like most things in this world, depends on the work of many people. I want to thank my brother Jake Vincl, for writing the music for the show. I want to thank my buddy Juliana Castro for designing the logos for the podcast. You can check out her work at julianacastro Castle.
- [101:41]
Joe Fort is the producer for the podcast and Mandy Lam is the production assistant. This podcast and other Peoples and Things programming are produced in affiliation with Virginia Tech Publishing and supported by the center for Humanities and the University Libraries at Virginia Tech. For information about other podcasts from Virginia Tech Publishing, this visit Publishing Vt Edu for the entire Peoples and Things team, I am Lee Vincl and most importantly, I want to thank you for listening. Thanks.
- [102:32]
Thomas Haigh
Sam.