52. Samuel Arbesman: Decay, Wonder and Where Knowledge Lives

Samuel Arbesman is a complexity scientist and writer. He is passionate about bringing together seemingly unrelated ideas from science and technology. Samuel works with companies and founders that recognize that the future happens at these boundaries, in such areas as open science, tools for thought, managing massive complexity, artificial intelligence, and infusing computation into everything from biology to manufacturing.

Samuel’s scientific research examines such areas as scientific discovery and network science. His writing has appeared in The New York Times, The Wall Street Journal, and The Atlantic, and he was previously a contributing writer for Wired. Samuel is the award-winning author of Overcomplicated: Technology at the Limits of Comprehension and The Half-Life of Facts. His most recent book is The Magic of Code.

In addition, Samuel is a Senior Fellow of The Silicon Flatirons Center at The University of Colorado, and a Research Fellow at The Long Now Foundation. Previously, Samuel was a Senior Scholar in Research and Policy at The Ewing Marion Kauffman Foundation and a Research Fellow in the Department of Health Care Policy at Harvard Medical School. He completed a PhD in computational biology at Cornell University and earned a BA in computer science and biology at Brandeis University.

Transcript

Intro 0:00

What's it worth? Uh-uh. Uh-uh. Uh-uh. What's it worth? Uh-uh. Uh-uh.

Simon Hill 0:09

Welcome back everyone to the Total Innovation Podcast. As always, I'm your host, Simon Hill. Here is the uncomfortable thought to begin with. A good portion of what you know already is out of date. Not because you were careless, but because the facts have a half-life. The things you learned, the certainties of your field, the numbers you quote with confidence, they decay slowly and quietly, at a rate you can almost measure, and nobody sends you a notification when they do finally become irrelevant. Today's guest has built a career on that idea and on the larger argument that follows from it. If knowledge expires, and if the systems we have built have grown so complex that no single person can hold them in their head, then the expert is no longer the safest person in the room. A topic many of you will know I hold dear to my own heart. The answer to a hard problem is far more likely to arrive from somebody or somewhere that we never thought to look. And that is a thesis I've spent much of my own life testing in the real world. And today's guest has spent his time proving it, not just on the page, that might be cruel, but across three remarkable books, from the decay of facts to the limits of what we can comprehend, and to the strange and beautiful logic of code itself. He is a complexity complexity scientist, easy to say, a writer, the scientist in residence at Lux Capital, and one of the most enjoyable minds you'll meet on the boundary between science, technology, and the things that we don't yet quite understand. And with that, I'm delighted to welcome Samuel Arbersman to the Total Innovation Podcast. Welcome, Sam. Thank you so much. Yeah, that's a very kind introduction. I appreciate it. Thank you. I work hard on those intros. People tend to love them, but they're all they're all true, right? This is this is you, or at least the you that I saw as I researched uh and got to know you in the build-up to this. Um let's start then with um with where your work first caught my attention. And I, you know, I do love the book, The Half-Life of Facts, for all of its, you know, it captures you at the title, which is always a good start for a good book as well. For someone hearing the phrase for the first time, what do you mean when you say that a fact has a measurable rate of decay?

Samuel Arbesman 2:19

Yeah, that is a great question to start with. Um, yeah, the way I kind of think about it is, and if you if you take the idea of just the half-life, or so I mean, which is by analogy with like radio uh radioactive materials and radioactivity, um, in the realm of radioactivity, uh, if I gave you a single atom of uranium, let's say, or some other radioactive material, you can't actually predict when it's going to decay. It might decay in the next fraction of a second. You might have to wait, I don't know, millions of years or whatever it is, depending on the material. Um, but it's really it's very unpredictable. But if I gave you many, many of those atoms, if I gave you an entire chunk of uranium, for example, suddenly then that unpredictability actually becomes remarkably predictable. You can actually say how long it will take for half of the material to decay. You can actually, and you can draw this nice curve, um, and you can actually calculate this half-life, which means like, and so even though I can't predict which specific atoms in that half are going to decay, I know overall the shape of the decay. And the idea behind the half-life of facts is that there is a similar kind of pattern to how we think about the like the change in knowledge, which is even though it might be very difficult for me to say, oh, this fact is going to be overturned in the next six months, or we're going to learn some new discovery in this in this field in some amount of time, there is an overall shape to how knowledge grows and changes. And so the idea is that, yes, it's on the on the one hand, it's true that maybe some of the things you learned um in school when you were younger or that were in your textbooks are no longer true. And that can feel a little overwhelming. Um, but underneath all of that, there are these regularities and there's almost this like mathematical pattern to how knowledge grows and changes, how errors are rooted out, um, and how knowledge actually decays over time. And so that was kind of the the analogy, the hopefully productive analogy in terms of how to think about the half-life of facts and the half-life of knowledge.

Simon Hill 4:14

Yeah, I love it. And uh as I said, it kind of grips you at the point through. And that half-life analogy in in that you that you drew on there is, you know, is obviously where people's minds go, but and it in a way it's a good anchor. But uh it's almost more profound than some of that, some of this thinking as you as you start to explore it as well. Um often when we do this, like it's good to do it through examples. So can you share some some of the some of your favorite examples that you uncovered of maybe facts that flipped or that people are most surprised to learn that you know that that maybe were once thought of oppositely or others?

Samuel Arbesman 4:50

Yeah, I mean, so and one example I think I it might even be like in the very beginning of the book that I give is um so my grandfather, um, he was uh he was a dentist. Uh he lived to the age of 99. Um, and he uh when he was in dental school, I believe it was in in dental school, he actually learned the wrong number of human chromosomes that are in a in a cell. Um he actually learned 48 instead of 46. Um, and it turns out that there was this period of time, um I think it was about like 20, 30 years or so, between when we could like actually image the cell and actually like count these, but not necessarily count them accurately. And so this wrong number kind of made it into the textbooks and was something that was not just at the frontier of knowledge, but was being taught in medical and dental schools and part of training. And it wasn't until I think the mid-1950s that a better imaging technique was developed and people actually re-counted and said, Oh, wait a second, there's actually only 46 chromosomes in a human cell instead of 48, which is kind of wild because you think, okay, that it's a pretty like basic thing. But I mean, so there's that, and actually related to medicine, I feel like medicine is one of these areas where you see so much of this kind of thing happening because there's new techniques and uh new and new technologies that are used or new treatments. Um, and so uh my father, who's a he's a retired dermatologist, um, he told me this story that when he was in medical school, a professor of his used the same exam, like with the same like multiple choice questions and answers. And one year, one of the choices on one of the questions was correct. And the next year, a different quite a different answer in the same multiple choice thing was correct as well. Uh, or a different one had become correct. And so there's just kind of this constant churn. Um, similarly, of course, I mean, there's like the well-known example of like Pluto being demoted in, I guess, 2006 or so. Um, but the truth is that even wasn't the first time that we had had a similar kind of thing. So it turns out in the 1800s, when uh when the when the asteroids of the asteroid belt were first being discovered, the first few that were discovered were much larger than the other asteroids. And I believe there was actually a little period of time where the like Ceres and some of the other large asteroids were actually considered to be planets, and there were charts of the solar system where those were listed as planets. And so, in the same way that I mean several generations of students uh learned Pluto as a planet and then it kind of got demoted, and they had to kind of contend with this change. Um, there was a similar kind of thing where I think a generation of students in the 1800s also learned the wrong planets. And of course, there's things around, I mean, what dinosaurs look like and how we think about it, are they are they kind of these like slow gray-green reptilian monsters that are cold-blooded, or now that are they actually warm-blooded and kind of fast and colorful and have feathers, and there's all these different things um that have changed over time. And so there's this constant churning of all these different facts and and bits of knowledge.

Simon Hill 7:52

Yeah, I love the Pluto example because I I have two kids, so nine and seven. And uh I was my parents are those parents that have still retained lots of things from when I was younger. And so I was back at my parents' home and I found this old um book that I had as a kid about space and everything. And yeah, in there we have Pluto as and the kids are like, that's not a planet. I'm like, what do you mean? I I learned from them that it's not a planet anymore. I'm like, all right, yeah, I wasn't paying that much attention for the that number of years, clearly.

Samuel Arbesman 8:21

So the interesting thing with and with that one is I think a lot of kids and um, and especially even like college students, like they know something is up about Pluto because oftentimes the teachers who taught them about the solar system still kind of have this like residual, like because they're a little bit older, they probably learned about Pluto as a planet, and they kind of still have this like residual feeling of like, oh yeah, like it still probably should have been a planet, but then it got demoted. And so, so many, many kids, if you talk to them and you ask them about Pluto, like some are like, oh yeah, it's like a it has like some other category categorization or whatever it is, but other ones are like, there's something up about Pluto, like we don't quite know what's going on there because there's kind of this like this almost like residual effect of the previous generations that learned it one way, who are still the educators. And so they so when they teach the the younger generation, they kind of include some other kind of yeah, complexity there as well, which is yeah, and so you can kind of see that even when knowledge changes, it takes so long for things to be constant, like to be updated and to really allow the the novel information to be able to be integrated.

Simon Hill 9:30

Yeah, I I think I mean you and I both you know exist in in these sort of you know frontier spaces in some way and looking at them. And in the book, you draw a distinction between the concepts I think of fast facts and slow facts, and these ones in the middle you call messo facts, if I'm pronouncing that correctly, as well. The the ones that change just slowly enough that we never quite notice. Um and I was thinking about this, you know, a lot of the teaching from business perspective is and I don't know whether we know whether they're all still facts or not, but the books have been around, you know, we're still we're still learning books from 60s, 70s, 80s, which you know may not feel that long ago for you and I, but they are quite a while ago now. Um, as a 70s baby, uh you know, I don't want it to be that long ago, but it's long enough ago, right? And uh and like the T and so how do we I guess how do we know and what is the concept of these fast and slow and messo facts in terms of you know uh I guess risks as much as anything, right? Because we're we're putting a lot of faith in in the concepts of a fact, particularly in what we're what we're training to lots of people. Yeah, some people are deeply critical and cynical, maybe that's what we call entrepreneurs, but still like what's the danger that sits behind it.

Samuel Arbesman 10:45

Yeah, I mean, when I think about like the rate at which information changes, I mean, right, there's like there's like the two odds the two extremes of like things that are kind of changing very, very rapidly, like I don't know, um, what the stock market closed at or what the weather might be be like tomorrow. Um and you kind of know that like no matter what um what the weather was like a week ago, you still have to look at the new weather prediction because you know that doesn't necessarily provide that much information. And so we're very good at understanding, okay, when things are kind of changing rapidly, we have to constantly update um and uh and actually just look things up. Um but then you have the opposite extreme, like things that you kind of learn once, and you you you don't have to worry about them changing, like the, I don't know, the number of continents or the the number of fingers on a human hand. Like I and maybe we'll like define continent a little bit differently, but like the actual physical shape of the contents is not really changing um for for any appreciable amount of time. Um and so those things you can kind of learn once, you don't have to worry about them. Um, but then in between, there's this very large range of knowledge that often change. And I talk about it as like this kind of like the middle scale, like the mesoscale, of knowledge that changes on the order of decades or on the order of a human lifetime. And the problem with that kind of information is we often learn it the same way we learn the facts that that we think might never change. Um, and then um, and we kind of learn them once, and then the way our educational systems often operates is like when you're young, you're kind of this generalist and you're learning lots and lots of different things about lots and lots of different areas. Um, and then as you as you grow and specialize, you you learn more and more about less and less. And you might be updating information in your own field, but by and large, outside of those areas, you're not learning all the things that kind of change at that mesoscale, um, even though they are changing. And then oftentimes you don't realize things have changed until you're confronted by the next generation who then says, Oh, guess what? Like Pluto's been demoted, or dinosaurs look completely different, or whatever it is. And you're like, What, like, what happened here? Um, and the thing is, a lot of that knowledge is changing steadily, but you don't perceive it that way. And then you're kind of like blindsided. And so I think actually one of the stories I mentioned in the book is about like some like financier or hedge fund manager who's like is giving this like a number of years ago, but it was like giving some talk and said, like, oh, like um, like this is true. Like, as we as we all know, like there's only four billion people on the planet. And of course, like like that isn't that has not been true for for quite some time. Um, but oftentimes you learn that fact. And so I think I might have learned like five or six billion people, and of course, now that is also out of date. Um, but until you are like unless you are very deliberately trying to update that knowledge, um, because of the way we learn those kinds of things and just the way that our educational system is structured, we often, yeah, just kind of have the, I don't know, half remembered facts that we learned in a read in a magazine or information that we learned when we're young. And even though these things are subtly changing, uh, we don't perceive it that way. We might perceive it in this kind of much more like stepwise fashion when we're kind of just confronted by some big change, or even worse, we just persist in the outdated knowledge and then make decisions based on those kinds of things. Um, and yeah, and so that's the reason why it can be um I don't know, I don't know if dangerous is too strong of a term, but kind of like these are like the riskiest category of uh of facts to kind of handle because they change slowly and steadily by and large, but we often don't update them because we kind of just learn them like once when you're when we're younger.

Simon Hill 14:24

Yeah, and there's there's a lot in this, right? And I think also just to set some context, you wrote this book a while ago, and people may, you know, like, why are we talking about it now? And I think partly because I discovered it quite recently and and wanted to talk about it, also because your bodies of work kind of flow off the back of it, and but it is also just a profoundly, you know, important observation, I think, as well. There's real science underneath this this concept, um, but also there's a weirdness about the things that we say because once we've established them as new facts, if you like, then it's you know the human psyche is often just like, well, of course that's now true, right? And of course it was wrong before. Um, so how how useful is it as a as a as a concept, or how maybe, maybe the other way around is how should people be thinking about the utility or the application of the knowledge that all facts have a half-life, but we don't quite know how how how long that half-life is, and it may be difficult to know when that half-life has been achieved or not.

Samuel Arbesman 15:25

Yeah, I mean, I I think I mean there's there's a number of different ways of kind of responding to this information. And and and to be clear, like I mean there's the both like the half-life, but then there's also just more kind of larger regularities to how knowledge grows and changes over time. And I think um for many people, um, if they're kind of confronted with the idea of like, oh, scientific knowledge is changing and requires updating, um, then they kind of then there's almost this like, I don't know, scientific nihilism where it's like, oh, like if things are constantly changing, we can't know anything, and they kind of just like I don't know, throw their hands up in the air and say, I it is what it is. Now, of course, that is not really how knowledge works. And I would say part of it, and actually it's encapsulated by this great quote by um uh the science fiction writer Isaac Asimov. Like so back in the day, I think someone wrote an wrote in a letter to him and said, uh, like, we used to think the earth was flat and we were wrong. Um, and then then we learned more and we thought it was like a perfect sphere. Um, turns out that's not entirely true. It's more kind of this like oblate sphere, this kind of crushed sphere. Um, and so therefore, like, how can we know anything if all these things are constantly being changed? And so um Asimov's response was if you think that thinking the earth is flat is just as thinking, is just as wrong as thinking that the earth is like a perfect sphere, then your view is wrongger than both of them put together. And the idea is that like, yes, we are overturning our knowledge, um, but it's kind of getting closer and closer to hopefully a true understanding of the world. And we're kind of like sort of like asymptotically approaching what might be a true understanding of the world. Um, and in fact, if you actually like measure out like the amount of error in like the earth is flat versus what what what what it really is versus like spherical, you can see that we've actually been like reducing the amount of error over time. And so, and I think that's like the kind of way to think about those, which is and by and large, at the core of our knowledge, um, we like those things are not changing. Sometimes they do change, but by and large, the core is kind of pretty well established. But then at the edges, we are changing what we're learning as as we have like better, better tools and technologies and imaging systems or whatever it is. Um, and the uh the way I think people kind of outside of the field of science perceive this, though, is that everything is changing. Because oftentimes when you're learning, like when you're reading in science journalism about article or about scientific advances, that's where the most exciting things are happening. And so that's where uh people would say, oh, like in nutrition, like we thought this thing was was healthy. Now we thought maybe it's carcinogenic and kind of goes back and forth. And we don't, and people will say, Oh, how do we know anything? And the answer is those scientists are all working at the frontier. At the frontier is where there is the most turmoil. Um, but I think the idea behind this is like like if we can kind of like internalize the idea that the frontier is where where the most exciting things are happening, um, and that's where there's going to be the most churn, that will kind of provide sort of a bounds on kind of our uncertainty. But I would say more broadly, it requires almost even just like internalizing the idea of what science actually is. Because like science, it's not just a body of facts or knowledge, it's really this rigorous means of querying the world around us. And and and so therefore, what that means though is of course things are going to be changing. And you kind of have to, and so science is always like in terms of like the knowledge, like the the facts are always in a draft form. Um, and actually, I even um one of my professors from graduate school he told me this great story. Uh, and I've mentioned it a lot, but I I love mentioning this story, where he was um, he was teaching a course, I think around Tuesday, uh, and lectured on some topic. Then the next day, he actually read a paper that invalidated everything he had taught in the lecture the day before. So he uh so then the next class, I think like the following day, like on a Thursday, he came into his into his class and said to his students, like, remember what I taught you on Tuesday? It's wrong. And if that bothers you, you need to get out of science. And I think there's that idea that like everything is constantly like in draft form. Yeah, we are we are doing our best, and therefore you shouldn't you shouldn't adhere to some sort of nihilism or anything like that. But holding things a little bit loosely and recognizing that things are in a draft form is a critical part of science. And I think increasingly, as all of us are kind of just confronted with knowledge, like knowledge, changing knowledge in our own fields or in just kind of things we read about in science or whatever it is, that kind of mentality, sort of the scientist mentality, not the scientist mentality of, oh, here's all the body of science, but the science I have idea of like, like there's a rigorous means of querying the world and things are constantly in draft form. I think that's actually really important for everyone to internalize. It's hard. I mean, and even among scientists, okay, it's one of these things where, yeah, you can talk a good game about this, but like well, when it's your own scientific scientific research um being invalidated or being overturned, then oftentimes scientists will like fight tooth and nail to like to avoid those things being overturned. Um, but I think by and large, kind of having that perspective, um, whether um just for all the information that you have in your head, um, I think is actually the most powerful one. Kind of like it almost means like we kind of all need to be scientists to a certain degree.

Simon Hill 20:33

Yeah, everything's in kind of permanent beta hypothesis mode, I guess, is is part of the part of the thing. Exactly on some of that. There's I want to talk a little bit around identity and and expertise as well. So there's a there's a there's a group of folk, and I would put myself in them, um, and it's a finding at the heart of everything that I do that around seven we work on wickedly complex problems, right? And about 70% of those problems that we throw out to a sort of a broad network of people, human and synthetic, are solved by someone from outside of the domain of that problem space, right? And and and the other, the other, the other, the broader corpus of that um of that research or the other sort of group of people that exist around this would put the argument forward that in those kind of domain areas or those kind of problem types, uh, the specialist is the least useful person in the room, right? Like you want to find other perspectives and other lenses, and and the the most likely person to solve it is the non-domain expert, right? And there's quite a lot of empirical evidence to sit behind this that generalists outperform specialists in that sort of work. Reading your work, I think that I think you might be in that camp as well, right? It doesn't feel like that would necessarily be surprising research to you. Um, maybe one, you can clarify whether I'm by reading or not, but if a domains expert's facts are quietly decriticating, and if the outsider has structural advantage, I'm going to blend a number of questions here. Like, what does that mean for? identity and expertise in in you know in in the world that we live in because we place a lot of value as humans on becoming increasingly specialist and expert on on specific cases particularly in science yeah I mean so I would say I mean I feel like it's there it's a little complicated there like on the one hand like I do not want to besmirch expertise I actually think like an expertise is a very important role.

Samuel Arbesman 22:22

That being said I mean as someone who kind of comes from like a very interdisciplinary field um I mean interdisciplinarity and more generalist approaches like they hold a very like dear place in my heart and like I and and I actually think like oftentimes we kind of undervalue those kinds of things. And I and I agree with you that when we are dealing with problems like these kind of the wicked problems you kind of mentioned or just like engaging with sort with I guess I mean the complexity of the world around us the complexity of the world around us is inherently interdisciplinary and requires lots and lots of different mental lenses and mental models and frameworks and ways of understanding the world. And so expertise is valuable for maybe being able to provide insight into one specific part of it. But oftentimes the role of the generalist will be okay how do I stitch together all these different things and I and I would say experts are probably pretty good at avoiding some of this kind of decay because if they know they're very little subspecialty, maybe they can kind of stay up to date on those those specific things might be very hard. But at the same time though, when it comes to solving these problems and when you have right like an expert is not going to be able to overcome jargon barriers or things like that um where oftentimes very relevant ideas and concepts or um mathematical frameworks or or or ideas um are that there are there are important things in other fields but if you don't even know the terminology of what to look for, you'll just never you'll just never know them. And so so for me like someone who is a generalist um or is more comfortable overcoming those jargon barriers is going to be really really well positioned for being able to kind of essentially run like an import export business for ideas and say, oh like here's the thing you're working on the like you as expert are thinking about it this way but guess what? Here's these other areas that are entirely relevant. And of course like you see this in like in the realm of science um there's there's many situations where people have like reinvented mathematical models or other kinds of things like many, many times. Like I there was one paper um this is it's been a while so I'm not sure I remember exactly but like I feel like they looked at the extent to which like some mathematical framework was like reinvented like eight or 10 times. And and I actually remember like when in um in my own field um back when I was doing a postdoc, um kind of like of network science and complex systems, I I remember being part of this um this mailing list. And and I feel like once a week someone would say oh like how do you measure this specific thing? And then of course like someone would write in and say oh this has been a solve problem for like 30 years in whatever my field is and and it was just this wild thing of like not even realizing this actually another another example of um of this in in in my postdoc as well is so a friend and I, we were working on um in a certain research product and we needed some sort of like data clustering technique and we couldn't quite find what we were looking for. And so we initially were like oh we'll just kind of make up like we'll like invent our own technique. It won't be so great but it'll be fine. And then I said like why don't we just talk to like the statistician down the hall and and I feel like within like 30 to 60 seconds he gave us the exact thing we needed. And it was one of these things where like we didn't even know what terminology to use. Now it could be with certain AI tools, overcoming jargon barriers is going to be is going to be a lot more easy and surmountable. But I still think that more kind of generalist perspective is really really important in being able to draw and kind of like yeah that import export business I mentioned before of like draw ideas from one field and apply it to another one and kind of act as that sort of um interstitial role where experts they do provide a lot of value, but they're not going to be able to do that kind of thing because you need to now draw from so many different domains in order to solve very, very interdisciplinary problems.

Simon Hill 26:17

Yeah and for fear of not besmirching experts or brothers as you put it I'm not walking you into a trap there. I think the uh in and maybe this leans nicely into into the maybe the the sort of subject matter I think of your second book, right, in terms of some of this the sort of complexity I think you've overcomplicated was the was the word or the title that you used that and I very much you know anyone working in the broad innovation systems world will be somewhat familiar with this in that you know everything has become so complicated and so multimodal, multidisciplinary I often use the example of you know like take something as simple as as a as a pencil, right? You know, there probably aren't that many people on planet earth who could tell you how to get to this pencil from from from from the raw material in the ground upwards right um and that's a basic pencil right there's lots more complexity from that. And so maybe that's part of the argument we're really making here right and that and that builds into that you want to just talk a little bit about overcomplicated and what it was what it was saying and how you think about the world through that lens.

Samuel Arbesman 27:21

Yeah I mean so yeah the basic the basic argument of the book is that um we are increasingly building systems um like sociotechnological systems um uh that are increasingly complex but often but not just like more complicated and more complex but increasingly so complex that they are not understandable to the people who themselves built these systems these technological systems or the experts who work with them on a daily basis. And going to back to what you're saying in terms of like construction and how we build these things, when we are building a very large complex technology or an industrial process or whatever it is, there are many amazing returns to specialization, like allowing like each person to be responsible for like little parts. And the problem is though is when all these things are connected and connected together, there is no single human being who understands all the different aspects of it. And the thing is that of course like I mean so the book came out in 2016 well before kind of the current like AI boom. And I feel like I had to kind of make this argument um and provide lots and lots of compelling examples to kind of like really bring it out. And now I almost you don't even need any examples. You're just like oh yeah look at these AI systems. No one understands what they're doing. They kind of just you pour a lot of data in um maybe the original algorithm is understandable but the combination of algorithm and training and data they work but we don't really understand how. So like it's almost this like like trivial kind of thing. But the truth is we don't need to go as far as these massive data centers or AI uh AI systems. I mean if you look at like I don't know a desktop computer or a laptop like these things are also incredibly complicated and no one person really understands all of the details um in all of their complexity. And so so the book kind of looks at what are the forces that lead us inexorably towards this kind of complexity when we build systems. And so it has to do with like increasing features and functionality, increasing interoperability and interconnection, um, having to contend with the complexity of the world around us. Like for example, and if you want to make like a software app that works for your the that like is a like some sort of like calendar application, um you can't just say, oh, there's 365 days in the year done. And of course because like there's leap years and those things are weird and then there's like different time zones and suddenly you're dealing with and contending with like the weird messiness of time itself and the world around us. And which means that the software you're building has to be necessarily complex. And so and through the those those kind of forces and of course I'm sure there's other ones as well you end up with systems that get more and more complicated and increasingly incomprehensible because to be honest I mean our brains are not like we do not evolve to contend with things that have millions of interconnecting parts. We can't really hold all those things together. I mean we can really only hold like a few bits of information in our minds at any one point. And so so I kind of look at like what are those like what are the reasons behind this incomprehensibility? But then also what do we do? I mean do we just kind of and going back to like the have line of facts of like oh yeah we can't like things are changing like let's just throw our hands up in the air in despair do we do the same kind of thing when it comes to incomprehensible technologies? And the answer is no. Like there is a wide gap between total understanding and total ignorance. And the and the question becomes what are the techniques we can use to query and interrogate our systems to give us a little bit more understanding and kind of move us along that continuum from ignorance to better understanding and and some of them are like if our technologies are increasingly rivaling biology in their complexity, maybe we can actually take a page from biologists and how they actually interrogate biological systems or ecology um ecologies and try to use those kinds of techniques to understand our systems better. Or understanding various specific like subcomponents or like what are the philosophical approaches we need to understand things or like I mean like exposing certain aspects of of our technologies. Because right now, I mean most people don't even realize the extent to which these technologies are incredibly incredibly incredibly complicated because we've been shielded from that. So like for example um when the Apple Watch first came out I remember reading some article in um I think it was like the styles section of the Wall Street Journal. And it was an article about like whether or not people are still gonna wear mechanical watches. And the answer is yeah people are going to still wear them. But they interviewed this one guy who was a fan of mechanical watches and and he had this line where he said like oh yeah of course I want a mechanical watch. Like I think of like the complexity of the watch of the mechanical watch as opposed to a smartwatch which is just a chip and like I mean just a chip but like it's like it's orders of magnitude more complex than the mechanical watch. But we've been shielded from it. And so having ways of allowing people to see the level of complexity and incomprehensibility of these systems, I think can also kind of give us a better sense of of these kinds of things. And so yeah for me I mean maybe maybe this is just because I'm sort of like temperamentally optimistic or happy but I do think even in the face of these vastly complicated technologies, there are ways of better understanding them. We maybe need to have a certain level of humility about the about how much we can understand them, but there are ways of kind of allowing us to better understand these technologies.

Simon Hill 32:30

Yeah and I and I think um and again I'm gonna I think it's somewhat and you know segues quite nicely into your latest book um on on this topic as well. But if I if I may for a second and I think the analogy the fact you are overcomplicated in a pre-AI era quite considerable pre-AI era and then we sort of draw some of the analogies perhaps to the pace of of of change and you know this in some senses very simple UX at the front end of it with all the complexity behind it and this argument of like these systems are so complicated. I was with you know one of the founders of of one of the large um uh models recently it's overcomplicated for them they don't understand you know there's so many parameters it surprises them on a daily basis what what it can do um it was anthropic and like clawed code surprised them right like it wasn't a thing that necessarily now became huge and every day something new surprises them and so the people building the things I think that's somewhere where some of the I guess concern and worry and if I lean into your optimism I'm an optimist as well right I believe very much in in you know positive futures and that that that we will you know whilst there's always a dark side there's always a very light side right and much more glass half full than half empty your latest book I think makes the case for wonder rather than fear about code and and I guess the broader technology landscape as well in the context I just said I feel like there's a the opposite is true for perhaps many people there's certainly this sort of more fear-based narrative I think around the direction of travel as to where we're going. And so if we take your concepts of wonder versus anxiety what what what does wonder buy versus the anxiety does not at the moment and why do you place that argument?

Samuel Arbesman 34:17

Yeah I mean so um I mean certainly when I think about I yeah the current technological moment right there is a lot of fear and anxiety um there's ignorance and there's fear in the face of that ignorance like there's a lot a lot of these kind of like very negative emotions kind of all bound up together. And when I think about though like my own childhood and kind of like early experiences of computing I mean there there really wasn't any of that. It was much more around like it was like full of wonder and delight. It also didn't feel like computation was just kind of this branch of engineering. It was all this like very humanistic liberal art pursuit of like connecting to like language and philosophy and biology. But like I my my own childhood like it was like like my family's first computer was the Commodore Fic 20 and then we had like like some early Macintoshes and like I remember like like there was like SimCity and fractals and screensavers and like all this kind of like very exciting weirdly interdisciplinary kind of stuff. And for me I and I definitely want to rekindle that sense of wonder. It can be a very it can be like a cautious wonder and delight because I and there are many concerns and I think a lot of the a lot of the concerns about these kinds of systems are actually quite valid. But I think when we have a sense of wonder and delight it's actually it's a much healthier way to kind of approach computing like it doesn't feel as adversarial of like oh it's like humanity versus the machine or whatever it is. Like there are there are ways of like working together. And in fact when we look at like complex technologies oftentimes like when or like these AI systems when we are confronted with them we're off we often kind of zoom off to two extremes either fear in the face of the unknown and all the unanticipated consequences or this almost like undue reverential awe of like almost like this like worshipful state of like oh my God like the the mind of Google or AI or whatever it is is amazing. And the truth is like ne not only are neither of these extremes good they're also not productive because they cut off questioning if you're constantly afraid and worried you can't necessarily examine these technologies as they are which is kind of they're kind of messy things. And the same thing with like this undue reverential awe it's like these systems and they're built by imperfect beings. They're built by humans. And like and and so for me kind of a humble sense of wonder for me like not only does it cut like it it doesn't cut off questioning but it also kind of is probably it's probably a more much more productive way of actually interrogating the systems of like oh these are interesting um yeah maybe we can be concerned about these things but let's keep on learning like just kind of having a certain sense of curiosity and playfulness. And so like for example the I mean there was the um the this classic meme several years ago of um like how do we kind of think about the um like these AI systems as sort of like the like like the Shogoth with the um with like the smiley face mask where so like the Shogoth is like this like hideous Lovecraftian monster and then like oh it has a little smiley face and like we kind of interact with the smiley face and we don't we kind of forget that it's kind of this like crazy thing underneath underneath the hood. I'm not sure if like Lovecraftian monster is the right analogy but if you say oh this thing is really really deeply weird and yeah we're kind of interacting it with the chat inter ch interface let's try to peek under the hood and kind of see what's going on there and find all the weird edge cases and messiness. And so for me, I mean maybe that's um maybe for some people that would be horrifying of like, oh I actually want to kind of understand these things and like realize they're actually much more complex and weird weirder than they are. For me, exploring edge cases and glitches or unexpectedness that's the window into learning about these kinds of things. And for me that's also a sense of wonder. Like so when I see um like glitches and failures especially when they're not like very dangerous ones, but like glitches and failures that are like kind of weird things that an AI system does or gaming systems that have some kind of delightful, unexpected behavior, not only are these windows into how the system actually operates, like and sometimes glitches are actually the way like a bug is kind of the way in which you narrow the gap between how you thought the system operates and how it actually does operate. But they also then can kind of provide this like sense of like oh delightful insight into what how these systems work. And so for me that kind of like curiosity, wonder, playfulness playing with these bugs and glitches and this kind of thing that is a very productive but also um it feels productive means of understanding these systems but also feels a lot healthier than kind of the extreme either negative or or or kind of like worshipful um attitudes that that too many people have because um they just don't feel yeah they they don't feel as healthy um and they kind of like they kind of put humanity very much in opposition um to to our technologies as opposed to recognizing no these things ultimately these systems they are built by people for people and we kind of have to find the way in which we can kind of make them work for us in the best way. And and I think wonder is kind of the right attitude uh towards figuring out okay how can we kind of make these things um be the most exciting for us.

Simon Hill 39:27

Yeah I think and it it I completely agree actually but it's that it's that level of uncertainty of having not solid firm ground under our feet therefore right and going well there there are shifting sands here and and interestingly in an agentic world that's kind of built on the truisms of one zeros as well right it's like the the the ai doesn't under understand necessarily the construct of half-lives and of of truth and fiction in many ways. It's a one zero one zero binary in in a number of senses and yet in the argument of of half-life and other things then one zero doesn't really exist or if it does it doesn't exist for very long in or it doesn't it exists for indeterminate amounts of time and we don't know which of those are the other determinate and indeterminate times that that could be both a positive or a negative and I guess you've got to take it on face value and run with the opportunity that gives but for many people the idea therefore that that the ground underneath us is nowhere near as solid as we think it might be on all assumptions, right? They're not one zero as they are sort of you know they're bugs or features and we don't quite know which one they are yet is a quite abstract and B quite quite unsettling I guess for you know or it might be unsettling for certain people.

Samuel Arbesman 40:40

Yeah no I I it definitely can be unsettling and I think um I I think partly is kind of just trying to get people to recognize that like we've always been to a certain degree on kind of these shifting sands of like levels of understanding or what the current like state of knowledge is. And if we can kind of approach that sort of constant change or lack of complete understanding with a with a sense of curiosity and humility as opposed to fear then we'll still I mean we still have to confront the world as it is but and and confront all this change. But if we can do it in a little bit more optimistic and positive way, then I think we'll be able to laugh like we'll be able to handle it for much longer than if we're just constantly worried at every single moment.

Simon Hill 41:28

Yeah I'm gonna ask you a couple of questions around AI and the half-lives now and I I was before we came on on to recording today I was talking to you about one of these experiments I've just been running um with an academic institution I do some work with. And uh one of the parts of that experiment uh was more of a less by design in this case but it kind of led me to this thought for this discussion we used a model that stopped being trained at the end of 2024. Right. So it doesn't know anything absolutely after 2024. And in the concepts of of of half true of half life of truths and everything that gives us a window of two and a bit years for things that we know became untrue to test whether the model could predict that truth or untruth or not right I haven't actually done that necessarily but I could do that. It's quite an interesting experiment to go, you know, and my because my question is going to be do we think in the age of AI that and it's gonna be too big a question it's going to be it depends massively I guess context for prefer pre pre pre predicting what you're going to say is that is AI gonna shrink or potentially extend some of these half-lives right are we because it may be looking at it from a slightly more contrained through that some somewhat specialist lens even though I know we can build all kinds of personas but it's built on a trained data set right for a variety of reasons. Or is it you know is there any evidence that you've seen yeah and maybe you haven't looked at it that closely that maybe it's going to help accelerate the the half-life experience of things or are they reasonably set in stone do you think? Yeah we've been through industrial revolutions and digitizations and transformations already and you know maybe we've got quicker maybe we haven't maybe we can't know right yeah I mean to be honest like I'm not sure I have like a very clear answer or like like um of like which way it's going to kind of go.

Samuel Arbesman 43:17

I would say I'm more interested in the way in which these AI tools which I have been like you've poured in all of human knowledge into these things or we'll we'll say a very large subset of human knowledge. It can hopefully I mean we were talking earlier about sort of like the import-export business of of these ideas and overcoming jargon barriers. I think that will rapidly collapse the like the amount of time ideas can be kind of in some subs some sub specialty or some specific field without without um like diffusing into the kind of the larger body of knowledge. I think AI will kind of help allow things to hopefully um spread and be useful. much, much more rapidly. Um and maybe I'm being like overly optimistic or overly naive, but that might not necessarily shrink the half-life of knowledge, but kind of just but accelerate the speed with which ideas that are already like that already exist can be actually be productively used in other fields. So I think that will be a boon for for science and innovation. But yeah, how it kind of changes some of these kind of half-lifes, I'm not sure. I mean I can definitely also see kind of like the negative version of um even when things are overturned if uh if people are kind of I mean living in their own like little like filter bubbles or whatever it is or kind of AI like just kind of like reinforcing whatever they whatever they think um yeah then they kind of constantly kind of have um outdated knowledge. I don't think the AI systems are doing that. I know this is now I'm just like speculating kind of um yeah I mean this is definitely sort of predictive stuff right yeah I don't know I so I don't know but I but I do think in terms of like helping with that import export business it's going to be very valuable.

unknown 45:01

Yeah.

Simon Hill 45:01

I don't as I said I think it's interesting. I'm some semi curious to go and even run that experiment now on something you know like it's quite good to have something that's very time bounded in terms of what it absolutely knows, right? And sort of retest that theory. So um I mean last couple of questions then so um I I'm kind of interested so you're you're increasingly involved in sort of research and academia and cataloguing new kinds of knowledge creating institutions and we've spoken already about you know some of this the the education that comes out and the knowledge that comes out of quite our institutions are sort of deigned to to protect settled knowledge. Yet that knowledge keeps expiring so what might you know in the in the world we're living in now and everything else and there's a big question this to end with but what might the institutions of the next century look like um and how do we rethink those perhaps?

Samuel Arbesman 45:51

Yeah oh that's an that's an interesting question. I mean I definitely I mean certainly I mean the institutions we have and like the the forms of research that we do right now um they shouldn't necessarily be like thrown away but I definitely think we can do a lot more. Like when I kind of think about where research and innovation happens, I mean people will say oh it can happen in like corporate industry labs, it can happen in universities, um, it can happen sometimes in like certain types of startups. But the truth is like those are just three points in some high dimensional space of potential organizational forms. And we should actually be exploring this high dimensional space a lot more. And so I mean and gratifyingly over the past few years there's been a lot of interesting new organizational or research like non-traditional research institutions, ones that maybe um are much more like like radically interdisciplinary maybe are focused a little bit more on kind of longer term kind of research and and I definitely think those kinds of things are really really important. For me, um I would certainly love kind of like a certain amount of like more open-endedness or kind of playfulness when we kind of think about ideas. But ultimately I I'm kind of agnostic as to what the form of these institutions are going to take. I kind of just I feel like we just need to run the experiment of actually trying to populate this high dimensional space much more. Which ones are going to win out, I don't know. But I think we need to do that because like for so long we've only had a relatively small set of institutional structures and and we need to actually try this and it could very well be and people talk about like how there's like a Cambrian explosion of new research institutions. Of course like when you have a very large explosion of new forms in evolutionary biology oftentimes it can be per it can be followed by an extinction event. And there might be that but hopefully on the other side we will have learned about new institutional forms that are actually things that we should be be using more of. And so um yeah so I'm I I I'm excited to see kind of where it goes but I'm also fairly agnostic as to what are the specific forms because I just think we need to try um like almost like rec recombining all the different features of the things that we realize could be useful and just trying out like exploring this like a combinatorial space of of research institutions.

Simon Hill 48:03

Yeah I was gonna say it feels like a combinatorial recombinatorial you know big experiment at the moment to see as you said like and there probably will be a big extinction event and maybe it reverts back to norm again and that was always the best version or I have a sneaky feeling that we're we're entering a different paradigm I think from from that angle now. We've spoken at length now and time is ticking upon us. So as much as I have a whole load of other stuff I want to go on I'm gonna ask you my one final question I ask everybody in this podcast which is you've written three amazing books. I highly advocate people to go pick them up I don't know whether they were ever planned to sort of follow each other sequentially in logic or that just happened through life but they do beautifully flow on from each other. And so I'll give you the benefit of strategic planning on that one over over over thank you I appreciate that but but I like to ask others you're you've got a great bookcase behind you and and you know we always learn something interesting. So is there one book that you know you would recommend to others that you know something new, something different perhaps that has either inspired you on your journey or surprised you um you know within the broad topics of you know knowledge, innovation, et cetera, et cetera?

Samuel Arbesman 49:10

Yeah I mean so I and we've talked a little bit about kind of like complex systems and uh simulation I think I even mentioned like when we're talking about like how like wonder and computing I mentioned like how I grew up with SimCity and things like that. There's actually this great book um came out several years ago called Building SimCity by Haim Gingold um who's also a friend. He's fantastic um and it's about the the creation of the computer game SimCity but also about the the deep like intellectual prehistory. So not just like and it includes like the nature of the game itself and the nature of the the the the computer company Maxis which created it and like the history of that but also this prehistory of like cellular automata and systems dynamics and like all these like deep ideas that were kind of recombined in this very novel and exciting way. And and for me I think about I mean is SimCity a very detailed model of a sit of a city? No. But can it actually lead you to understand many, many things about urban planning or even just simply to grapple with the unanticipated consequences and feedback of a complex system that kind of does things in a way that you don't you don't anticipate? Yeah it's really good at all these kinds of things. And so I I love thinking about these kind of like simulation toys these kind of like software with this kind of like little like merc like a microcosm within it. And um yeah and this book is a really good window into kind of many many ideas that kind of go into this kind of thing.

Simon Hill 50:35

I love this question and I I agree and I think SimCity I don't know if it's hugely underrated but I feel like it needs to just permeate every generation. It's such an important as you said simulation and concept for the system level of analysis so uh brilliant and a book I haven't read so now I must go and pick it up I always commit to doing that as well. Sam, thank you so much. Um you have taken us through three books without us going too deep I've pulled you back in your mind and consciousness over a decade or more and from from the first book as well but I know it applies a lot to the work you're doing today and and the work that you're doing across your your sort of portfolio career at this point in time. It's been an honor having you on the podcast thank you thank you so much for uh for your time for the details for the sharing of the stories um in a detailed and very charismatic way as well so much appreciated thank you.

Samuel Arbesman 51:23

Oh thank you so much this is a lot of fun I really appreciate it.

Simon Hill 51:25

Great um where can people find more about you?

Samuel Arbesman 51:28

Like do you have a website or whatever yeah I mean so my website um so arbisman.net just my last name dot net um there's lots of links to uh recent articles and things and you can sign up for I have I have a newsletter um I as I think many people do nowadays and so yeah you can sign up for there and then that will you'll you'll be able to subscribe to my my writings and musings about many of these ideas.

Simon Hill 51:50

Yeah they're great writings and great musings. So thank thank you very much to everybody as always thank you very much. I know there's many podcasts out there for you to listen to we appreciate all of our listeners all of our subscribers this has been another incredible episode there's many more gone before please check those out and we have many more exciting ones coming as well um next up is a topic looking at failure and some incredible stories of failure over the years from another Sam actually so for today thank you very much Sam thank you everybody as always I've been Simon Hill this has been the Total Innovation Podcast.

Intro 52:20

Goodbye what's it worth uh uh uh uh uh uh what's it worth uh uh uh uh