Sam Arbesman on the magic of code, tools for thought, interdisciplinary ideas, and latent spaces (AC Ep5)

Code, ultimately, is this weird material that’s somewhere between the physical and the informational… it connects to all these different domains—science, the humanities, social sciences—really every aspect of our lives.

– Sam Arbesman

What you will learn

• Rekindling wonder through computing
• Code as a universal solvent of ideas
• Tools for thought and cognitive augmentation
• The human side of programming and AI
• Connecting art, science, and technology
• Uncovering latent knowledge with AI
• Choosing technologies that enrich humanity

Episode Resources

Books

• The Magic of Code

• As We May Think

• Undiscovered Public Knowledge

People

• Richard Powers

• Larry Lessig

• Vannevar Bush

• Don Swanson

• Steve Jobs

• Jonathan Haidt

Concepts and Technical Terms

• universal solvent

• latent spaces

• semantic networks

• AI (Artificial Intelligence)

• hypertext

• associative thinking

• network science

• big tech

• machine-readable law

Transcript

Ross Dawson: Sam, it is wonderful to have you on the show.

Sam Arbesman: Thank you so much. Great to be talking with you.

Ross: So you have a book coming out. When’s it coming out?

Sam: It comes out June 10. So, yeah, so it comes out June 10. The name of the book is The Magic of Code, and it’s about, basically, the wonders and weirdness of computing—kind of viewing computation and code and all the things around computers less as a branch of engineering and more as almost this humanistic liberal art.

When you think of it, it should not just talk about computer science, but should also connect to language and philosophy and biology and how we think, and all these different areas.

Ross: Yeah, and I think these things are often not seen in the biggest picture. Not just, all right, this is something that draws my phone or whatever, but it is an intrinsic part of thought, of the universe, of everything.

So I think you—indeed, code, in as many manifestations—does have magic, as you have revealed. And one of the things I love, love very much—just the title Magic—but also you talk about wonder.

I think when I look at the change, I see that humans are so quick to take things for granted, and that takes away from the wonder of what it is we have created. I mean, what do you see in that? How do we nurture that wonder, which nurtures us in turn?

Sam: Yeah. I mean, I completely agree that we are—I guess the positive way to think about it is—we adapt really quickly. But as a result, we kind of forget that there are these aspects of wonder and delight.

When I think about how we talk about technology more broadly, or certain aspects of computing, computation, it feels like we kind of have this sort of a broken conversation there, where we focus on it as an adversary, or we are worried about these technologies, or sometimes we’re just plain ignorant about them.

But when I think about my own experiences with computing growing up, it wasn’t just that. It was also—it was full of wonder and delight. I had, like, my early experiences—like my family’s first computer was the Commodore VIC-20—and kind of seeing that.

And then there was my first experience using a computer mouse with the early Mac and some of the early Macintoshes or earlier ones. And then my first programming experiences, and thinking about fractals and screensavers and SimCity and all these things.

These things were just really, really delightful and interesting. And in thinking about them, they drew together all these different domains. And my goal is to kind of try to rekindle that wonder.

I actually am reminded—I don’t think I mentioned this story in the book—but I’m reminded of a story related to my grandfather. So my grandfather, he lived to the age of 99. He was a lifelong fan of science fiction, and he read—he basically read science fiction since, like, the modern dawn of the genre.

Basically, I think he read Dune when it was serialized in a magazine. And I remember when the iPhone first came out, I went with my grandfather and my father. We went to the Apple Store, and we went to check it out. We were playing with the phone.

And my grandfather at one point says, “This is it. Like, this is the object I’ve been reading about all these years in science fiction.”

And we’ve gone from that moment to basically complaining about battery life or camera resolution. And it’s fair to want newer and better things, but we kind of have to take a beat and say, no, no—the things that we have created for ourselves are quite spectacular.

And so my book tries to rekindle that sense of wonder. And as part of that process, tries to show that it’s not just this kind of constant march of better camera resolution or whatever it is. It’s also this process of touching upon all these different areas that we think about—whether it’s the nature of life or art or all these other things.

And I think that, hopefully, is one way of kind of providing this healthier approach to technology, rekindling this wonder, and ultimately really trying to connect the human to the machine.

Ross: Yes, yes, because we have—what I always point out is that we are inventors, and we have created extraordinary things. We are the creators, and we have created things in our own image. We have a relationship with them, and that relationship is evolving.

These are human artifacts. Why they matter, and how they matter, is in relationship to us, which, of course, goes to— You, sorry, go on.

Sam: Oh no, I was just gonna agree with you. Yeah. I feel like, right, these are human artifacts, so therefore we should think about how can they make us the best versions of humans, or the best versions of ourselves, as opposed to sometimes the worst versions of ourselves.

Right? So there’s a sense of—we have to be kind of deliberate about this, but also remember, right, we are the ones who built these things. They’re not just kind of achieving escape velocity, and then we’re stuck with the way in which they make us feel or the way in which they make us act.

Ross: All right. Well, you’re going to come back in a moment, and I’m going to ask you precisely that—how do we let technology make us the best we can be?

But sort of on the way there, there are a couple of very interesting phrases you use in the book. “Connection machines”—these are connection machines. Also “universal solvent.” You use this phrase both at the beginning and the end of the book.

So what do you mean by “universal solvent”? In what way is code a universal solvent? What does that mean?

Sam: Yeah, so the idea is—it’s by analogy with water. Water is kind of a universal solvent; it is able to dissolve many, many different things within itself.

I think about computing and code and computation as this universal solvent for many aspects of our lives—kind of going back to what I was saying before, when we think about language. It turns out that thinking about code actually can provide insight into how to think about language.

If we want to think about certain ideas around how ancient mythological tales are transmitted from generation to generation—it turns out, maybe with a little bit of stretching, but you can actually connect it to code and computation and software as well.

And the same kind of thing with biology, or certain aspects of trying to understand reality through simulation. All these things have the potential to be dissolved within computing.

Now, it could be that maybe I’m just being overly optimistic with code, like, “Oh, code can do this, but no other thing can do that.” It could be that lots of other fields have the ability to connect.

Certainly, I love this kind of interdisciplinary clashing of different ideas. But I do think that the ideas of computation and computing—they are beyond just what we would maybe categorize as computer science or programming or software development or engineering.

When we think about these ideas—and it turns out there’s a lot of really deep ideas within the theory of computation, things like that—when we think about those ideas or the areas that they connect with, it really does impinge upon all these different domains: of science, of the humanities, of the social sciences, of really just every aspect of our lives.

And so that’s kind of what I’m talking about.

And then you also mentioned this kind of, like, this supreme connection machine. And so I quote this from—it was, I believe, the novelist Richard Powers. He’s talking about the power of the novel—like, certain novels can really, in the course of their plot and their story, connect so many different ideas.

And I really agree with that. But I also think that we can think the same thing about computing as well.

Ross: You know, if we think about physics as the various layers of science—where physics is the study of nature and the universe—and that is basically a set of equations. It is maths. And these are things which are essentially algorithms which we can express in code.

But this goes to the social layers of the algorithms that drive society. And I also recall Larry Lessig’s book Code, back from 25 years ago, with the sort of parallels between essentially the code as law and code as software.

In fact, a recent innovation in New Zealand has released machine-readable law—legislation basically embedding legislation in code—so that this can now be unambiguous and then read by machines, and so they can implicitly obey what they do.

So there’s a similar multiple facets of code, from social structures down to the nature of the universe.

Sam: I love that, yeah. And where I do think, yeah, there is something deep there, right? That when we think about—because code, ultimately, it is this very weird thing.

We think of it as kind of text, like on a screen, but it is only really code when it’s actually able to be run. And so it’s this kind of thought stuff—these words—but they’re very precise, and they also are then able to act in the world.

And so it’s kind of this weird material that’s somewhere between the physical and the informational. It’s definitely more informational, but it kind of hinges on the real world. And in that way, it has this kind of at least somewhat unique property.

And as a result, I think it can connect to all these other different domains.

Ross: So the three major sections of your book—in the middle one is Thought. So, of course, we can have code as a manifestation of thought. We can have code which shapes thought.

And one of the chapters is titled Tools for Thought, which has certainly been a lot of what we’ve looked at in this podcast over a long period of time.

So, let’s start to dig into that. At a high level, what do you describe as—what do you see as—tools for thought?

Sam: Yeah, I mean, so tools for thought—I mean, certainly, there’s a whole domain of software within this kind of thing.

And I actually think that there’s a really long history within this, and this is one of the things I also like thinking about, and I do a lot in the book as well, which is kind of try to understand the deeper history of these technologies—trying to kind of understand where they’ve come from, what are the intellectual threads.

Because one of the other interesting things that I’ve noticed is that a lot of interesting trends now—whether it’s democratizing software development or tools for thought or certain cutting-edge things in simulation—these things are not that new.

It turns out most of these ideas were present, if not at the very advent of the modern digital computer, then they were at least around relatively soon after. But it was the kind of thing where these ideas maybe were forgotten, or they just took some time to really develop.

And so, like, for example, one of the classic beginnings of tools for thought—well, I’ll take a step back. The way to kind of think about tools for thought is probably the best way to think about it is in the context of the classic Steve Jobs line, “the bicycle for the mind.”

And so the idea behind this is—I think he talked about it in the 1970s, at least initially—I think it was based on a Scientific American article he read in the ’70s, where there was a chart of, I guess, like the energy efficiency for mobility for different animals.

And I think it was, like, the albatross was really efficient, or whatever it was, and some other ones were not so efficient. And humans were pretty mediocre.

But then things changed—if you put a human on a bicycle, suddenly they were much, much more energy efficient, and they were able to be extremely mobile without using nearly as much energy.

And his argument is that in the same way that a bicycle provides this efficiency and power for mobility for humans, computers can be these bicycles for the mind—kind of allowing us to do this stuff of thought that much more efficiently.

Ross: Well, but I guess the thing is, though, is that—yeah, that’s, it’s a nice concept. I think, yeah, 

Sam: Oh yeah, it’s very popular. 

Ross: The question is, how?

Sam: Yes, yeah. So, how does it, how does it work?

So the classic place—and I actually discuss even a deeper prehistory—but like, the classic place where people start a lot of this is with Vannevar Bush, his essay in The Atlantic, I think in 1945, As We May Think.

And within it—he’s discussing a lot of different things in this article—but within it, he describes this idea of a tool called the Memex, which is essentially a thought experiment. And the way to think about it is, it’s kind of like a desk pseudo-computer that involves, I think, microfilm and projections.

But basically, he’s describing a personalized version of the web, where you can connect together different bits of information and articles and things you’re reading and traverse all of this information. And he kind of had this idea for the web—or at least, if you squint a lot. It was not a reality; there was not the technology really quite there yet, although he describes it using the current cutting-edge technology of microfilm or whatever it was.

And then people kind of proceeded with lots of different things around hypertext or whatever. But in terms of one of the basic ideas there, in terms of what is that tool for thought—it is ultimately the idea of being able to stitch together and interconnect lots of different kinds of information.

Because right now—or I wouldn’t say right now—in the early days of computing, I think a lot of people thought about computers from the perspective of just either managing large amounts of information or being able to step through things in a linear fashion.

And there was this other trend saying, no, no—things should be interconnected, and it should be able to be accessed non-linearly, or based on similar topics, or based on, ultimately, the way in which our brains operate. Because our brains are very associative. Like, we associate lots of different things. You’ll say one thing, it’ll spark a whole bunch of different ideas in my mind, and I’ll go off in multiple different directions and get excited about lots of different things.

And we should have a way, ultimately, of using computers that enhances that kind of ability—that associative ability. Sometimes maybe complement it, so it’ll make things a little bit more linear when I want to go very associative.

But I think that’s ultimately the kinds of tools for thought that people have talked about.

But then there’s other ones as well. Like, using kind of more visual methods to allow you to manipulate information, or see or visualize or see things in a different way that allows you to actually think different thoughts.

Because ultimately, one of the nice things about showing your work or writing things down on paper is it allows you to have some spatial representation of the ideas that you’re exploring, or write all the things down that maybe you can’t immediately remember in your short-term memory.

And ultimately, what it comes down to is: humans are limited creatures. Our memories are not great. We’re distractible. We associate things really well, but it’s not always nearly as systematic as we want.

And the idea is—can a computer, as a tool for thought, augment all these things? Make the way in which we think better, as well as offset all the limitations that we have?

Because we’re pretty bad when it comes to certain types of thinking. And so I think that is kind of the grand vision.

And I can talk about how certain trends with AI are kind of helping actually cash a lot of these promissory notes that people have tried to do for many, many years.

But I think that’s kind of one broad way of thinking about how to think of this broad space of tools for thought—which is recognizing humans are finite, and how can we do what we want to do already better, which is think.

And to be clear, I don’t want computers to act as sort of a substitute for thought. I enjoy thinking. I think that the process of thought itself is a very rewarding thing. And so I want these kinds of tools to allow me to feel like the best version of the thinking Sam—as opposed to, “Oh no, this kind of thing can think for me. I don’t have to do that.”

Ross: So you mentioned—you start off from looking around the sense of how it is you can support or augment the implicit semantic networks of our thinking.

These are broad ideas where, essentially, we do think in semantic networks of various kinds. And there are ways in which technology can support it.

So I suppose, coming to the present, as you say, AI has been able to bring some of these to fruition. So what specifically have you seen, or do you see emerging around how AI tools can support us in specifically that richer, more associative or complementary type of prostheses?

Sam: Yeah, so one basic feature of AI is this idea of being able to embed huge amounts of information in these kind of latent spaces, where there are some massively high-dimensional representations of articles or essays or paragraphs—or just information in general.

And the locations of those different things often are based on proximity in some sort of high-dimensional semantic space.

And so the way I think about this is—well before a lot of these current AI advances, there was this information scientist by the name of Don Swanson. And I think he wrote this paper—I think it was like the mid-1980s—it was called…

Oh, and I’m blanking on it, give me a moment. Oh—it was called Undiscovered Public Knowledge. And the idea behind it is: imagine some scientific paper somewhere in the vast scientific literature that says “A implies B.”

Then somewhere else in the literature—could be in the same subfield, could be in a totally different field—there’s another paper that says “B implies C.” And so, if you were to read both papers and combine them, you would know that perhaps “A implies C” by virtue of combining these two papers together.

But because the scientific literature is so vast, no one has actually ever read both of these papers. And so there is this knowledge that is kind of out there, but it’s undiscovered—this kind of undiscovered public knowledge.

He was not content to leave this as a thought experiment. He actually used the cutting-edge technology of the day, which was—I think—keyword searches and online medical databases. Or I don’t know if it was even online at the time.

And he was actually able to find some interesting medical results. I think he published them in a medical journal, which is kind of exciting.

This is kind of a very rudimentary thing of saying, “Okay, can we find relationships between things that are not otherwise connected?” Now, in this case, it required keyword searches, and it was pretty limited.

Once you eliminate some of those barriers, the ability to stitch together knowledge that might otherwise never be connected is enormously powerful and completely available.

And I think AI, through this kind of idea of embedding information within latent spaces, allows for this kind of thing.

So the way I think about this is—if you know the specific terms, maybe you can find those specific papers you need. But oftentimes, people are not specifying things in the exact same way.

Certainly, if they are in different domains and different fields, there are jargon barriers that you might have to overcome.

For example, back when I was a postdoc—I worked in the general field of network science—and I was part of this interdisciplinary email list. I feel like every week, someone would email and say, “Oh, how do I do this specific network metric?”

And someone else would invariably email back and say, “Oh, this has been known for 30 years in physics or sociology,” or whatever it was.

And it was because people just didn’t even know what to search for. They couldn’t find the information that was already there.

And with these much more fuzzy latent spaces, a lot of these jargon barriers are just entirely eliminated.

And so I think we now have an unbelievable possibility for being able to stitch together all this information—which will potentially create new hypotheses that can be tested in science, new ideas that could be developed—because these different fields are stitched together.

Yeah, there’s so many things. And so that is certainly one area that I think a lot about.

Ross: Yeah, so just one—I mean, in that domain, absolutely, there’s extraordinary potential to, as you say, reveal the latent connections between knowledge—complementary knowledge—which is from our vast knowledge we’ve created as humanity.

There are many more connections between those to explore, which will come to fruition.

This does come to the humans-plus-AI piece, where, on one level, the AI can surface all of these connections which might not have been evident, but then come to the fore. So that is now a critical part of the scientific process.

I mean, arguably, a lot of science is collecting what was already there before, and now we’re able to supercharge that.

So in this humans-plus-AI world, where’s the role of the human there?

Sam: So that’s a good question. I mean, I would say, I’m hesitant to say that there’s any specific task that only a human can do forever. It seems to be—any time you say, “Oh, only humans can do this,” we are invariably proven wrong, sometimes almost instantly.

So I kind of say this a lot with a lot of humility. That being said, I do think in the near term, there is a great deal of space for humans to act in this almost managerial role—specifically in terms of taste.

Like, what are the interesting areas to focus on? What are the kinds of questions that are important?

And then, once you aim this enormously powerful tool in that direction, then it kind of goes off, and it’s merciless in connecting things and providing hypotheses and suggestions and ideas and potential discoveries and things to work on.

But knowing the kinds of questions and the kinds of things that are important or that will unlock new avenues—it seems right now (maybe this will no longer be the case soon), but at least right now, I still think there’s an important role for humans to provide that sense of taste or aim, in terms of the directions that we should be focusing on.

Ross: So going back to that question we touched on before—how do we as humans be the best we possibly can be?

Now that we have—well, I suppose this is more a general, broader question—but also now that we have extraordinary tools, including ones of code in various guises, to assist us, how do we be the best we can be?

Sam: Yeah, I think that is the singular question of this age, in this moment.

And in truth, I think we should always be asking these questions about, okay, being the best versions of ourselves. How do we create meaning and purpose and things like that?

I do think a lot of the recent advances with AI are sharpening a lot of these kinds of things.

Going back to what I was saying before—at many moments throughout history, we’ve said, “Oh, humans are distinct from animals in certain ways,” and then we realized, “Oh, maybe animals can actually do some of those kinds of things.”

And now, we are increasingly doing the same kind of thing with AI—saying, “Oh, AI can maybe recommend things to purchase, but it can never write crappy poetry,” and guess what? Oh, it actually can write pretty mediocre poetry too.

So for me, I kind of view it as—by analogy, there’s this idea, somewhat disparagingly, within theology, of how you define the idea of God. Some people will say, “Oh, it’s simply anything that science cannot explain yet.”

This is called the “God of the gaps.”

And of course, science then proceeds forward, explaining various things in astronomy, cosmology, evolution, all these different areas. And suddenly, if you ascribe to this idea, your conception of God gets narrower and narrower and might eventually vanish entirely.

And I feel like we are doing the same kind of thing when it comes to how we think about AI and humanity. Like, “Oh, here are the things that AI can do, but these are the things that humans can do that AI can never do.”

And suddenly, that list gets shorter and shorter.

So for me, it’s less about what is uniquely human—because that uniqueness is sort of a moving target—and more about what is quintessentially human.

What are the things—and this goes back to exactly your question—what are the things that we truly want to be focusing on? What are the things that really make us feel truly human—like the best versions of ourselves?

And those answers can be very different for many people. Maybe you want to spend your time gardening, or spending time with your family, or whatever it is.

But certainly, one aspect of this—related to tools for thought—is the idea that I do think that certain aspects of thought and thinking are a quintessentially human activity.

Not necessarily unique, because it seems as if AI can actually do, if not real thought, then a very accurate simulacrum of thought.

But this is something that does feel quintessentially human—that we actually want to be doing ourselves, as opposed to outsourcing entirely.

So I think, as a society, we have to say, “Okay, what are the things that we do want to spend our time doing?” and then make sure that our technologies are giving us that space to do those kinds of things.

And I don’t have all the answers of what that kind of computational world will look like exactly, or even how to bend the entire world of big tech toward those ends. I think that is a very large and complicated issue.

But I do think that these kinds of questions—the ones you asked me and the ones I’m talking about—these are the kinds of questions we need to really be asking as a society.

You’re seeing hints of that, even separate from AI, in terms of how we’re thinking about smartphone usage—especially smartphone usage among children.

Like, Jonathan Haidt has been talking about these things over the past several years, and really caused—at least in the United States—kind of a national conversation around, “Okay, when should we be giving phones to children? Should we be giving them phones? What kinds of childhoods do we want our children to have?”

And I feel like that’s the same kind of conversation we should be having more broadly for technology: What are the lives we want to have?

If so, how can we pick and choose the kinds of technologies we want?

And I do think—even though some of these things are out of our hands, in the sense that I cannot unilaterally say, “Oh, large social media giant, change the way your algorithm operates”—they’re not going to listen to me.

But I can still say, “Oh, in the absence of you doing the kinds of things that I want, I don’t have to play your game. I don’t have to actually use social media.”

So there is still some element of agency in terms of picking and choosing the kinds of technologies you want.

Now, it’s always easier said than done, because a lot of these things have mechanisms built in to make you use them in a certain way that is sometimes against your better judgment and the better angels of our nature.

But I still think it is worth trying for those kinds of things.

So anyway, that’s a long way of saying I feel like we need to have these conversations. I don’t necessarily have all the answers, but I do think that the more we talk about what are the kinds of things that make us feel quintessentially human, then hopefully we can start picking and choosing the kinds of technologies that work for that.

So, like, if we love art, what are the technologies that allow us to make better art—as opposed to just creating sort of, I don’t know, AI slop, or whatever people talk about?

Depending on the specific topic you’re focusing on, there’s lots of practicalities. But I do think we need to be having this conversation.

Ross: So just rounding out, in terms of looking at the ideas in your book—sort of very wide-ranging—what is your advice, or what are your suggestions for people in terms of anything that they could do which will enhance themselves or make them better versions of themselves, or more better suited to the world in which we are living?

Sam: That is a great question.

And I think I would say it’s related to kind of just being deliberate—whether it’s being deliberate in the technologies you adopt or being deliberate in terms of the kinds of things that you want to be spending your time on.

And it’s even beyond technology. It’s more about, okay, what is the—it involves saying, “Okay, what are the kinds of things I want to do, or the kind of life I want to live?” And then pick and choose technology, and the kinds of technology, that really feel like they enhance those kinds of things as opposed to diminish them.

Because, I mean, as much as I talk about computation as this universal solvent that touches upon lots of different things—computing, it is not all of life.

As much as I think there is the need for reigniting wonder and things like that, not everything should be computational. I think that’s fine—to have spaces where we are a little bit more deliberate about that.

But going back to the sense of wonder, I also think ultimately it is about trying to find ways of rekindling that wonder when we use certain aspects of our technologies.

Like, if we feel like, “Oh, my entire technological life is spent in this, I don’t know, fairly bland world of enterprise software and social media,” there’s not much wonder there. There’s maybe anger or rage or various other kind of extreme emotions, but there’s usually not delight and wonder.

And so I would say, on the practical sense, probably a good rule of thumb for the kinds of technologies that are worth adopting are the ones that spark that sense of wonder and delight.

Because if they do that, then they’re probably at least directionally correct in terms of the kinds of things that are maybe a little bit more humane or in line with our humanity.

Ross: Fantastic. So where can people go to find out more about your work and your book?

Sam: So my website—it’s just my last name, Arbesman. So arbesman.net is my website. And on there, you can read about the book.

I actually made a little website for this new book The Magic of Code. It’s just themagicofcode.com. So if you go to that, you can find out more about the book.

And if you go on arbesman.net, you can also find links to subscribe to my newsletter and various other sources of my writing.

Ross: Fantastic. Loved the book, Sam. Wonderful to have a conversation with you. Thanks so much.

Sam: Thank you so much. This was wonderful. I really appreciate it.