Apple Podcasts | Google Podcasts | Spotify

Rhys Lindmark: Hello listeners. Today, I'm excited to chat with Cesar Hidalgo. Cesar is in the Chilean-Spanish-American scholar in economic complexity, data visualization, and applied artificial intelligence. He's written some great books that we’ll chat about today, like Why Information Grows and How Humans Judge Machines. Cesar, thanks for being on the show, and welcome.

Cesar Hidalgo: Oh, it's my pleasure to be here.

Rhys: Yeah, we're excited to dive in. And yes, he's never chatted about this before the show, but there are 3 big things that we kind of want to dive into today. Cesar is a great visual thinker. So I want to kind of understand that because I would like to be a better one and a natural one, but I think you do a great job of it and then also to dive into your books. But let's start with that visual thinking side of things. As I was watching one of your talks last night, you had shown, there are just so many amazing examples of this visual thinking that you do.

Whether it's graphing things in two and three dimensions, or whether it is your project data wheel, I think, is that the name of it, where you, yeah, then like the Journal of Economic Complexity. You just do this amazing job. Let me give one other example for our listeners, which I think was really cool. You gave this talk at Google with Talks at Google for How Humans Judge Machines. At the end of the talk you put up a picture on screen of all the pages of your book and you highlighted the different pages that you had covered in that talk and you're like, hey, this is what we covered today, but there's also all this other stuff and it was just a cool way to see the visual.

So tell me a little bit more about how you think in visuals and how that ends up being portrayed to others?

Cesar: So in some way, my main concern is always to communicate. My fear is the fear of not being understood and I tried to fight that fear in different ways. I tried to fight that fear by trying to be as simple as possible when I write, but also I realized that writing has some inherent limitations and the main limitation of writing is that writing is linear, when you speak, when you ride, when you watch a movie.

You cannot avoid having one thing come before the other. There is this linear component to narratives, and that is great if you're able to linearize your thoughts properly. But in reality, it's a very difficult process because the way that our mind works, or at least the way that it feels that it works, it's sometimes in parallel. Sometimes you get hit by stimuli and you start thinking about multiple possibilities. So I like to use visuals when I try to communicate things that have that property that it's more something in parallel than something linear.

When you want to look at multiple things at the same time and compare them, and I think that's what visuals are able to do well is to tap into this other mode of thinking in which you have comparisons from sizes, to shapes, to positions and so forth that are very cumbersome to communicate in writing. Writing is not bad at describing geometries or systems or networks and all those things. Sometimes those diagrams that are able to show things simultaneously do that job much better. So I think both my writing and my user visualization stem from the same fear is the fear of not being understood. But when it comes to writing, I try to make sure that I always say what I'm going to use before I use it. When it comes to visuals, I tried to use it when I try to communicate something that involves that parallel type of thinking.

Rhys: Yeah, I love it. It's funny because for you and for me to some extent being in this field of complex systems and networks and stuff like that, those things are so much better to convey visually like a feedback loop or network graph or whatever. It's like, hey, you could say every little bit of the feedback loop, but it's kind of easier for the viewer to just check it out and be like, oh, OK, this is how it works, here's the loop here, here's the loop there, and they kind of get it more intuitively.

So I think that it's funny 'cause I think your work itself probably lends itself well to visualization. What in your mind makes something like a great data visualization?

Cesar: I think at the end, of course, it has to have a strong take-home message, and sometimes visualizations can be very simple. But the power that they have is the topic that they're representing or the aspect of the topic that they highlight. I think that's one part of it. The other part of it, which I think is really important, is that it has to be memorable too. That's kind of challenging because a lot of the time the visualizations that are very memorable have to be a little more unique and it's hard to be unique without introducing things that might be decorative or unnecessary. But sometimes you can have something as simple as a bar chart with a very powerful point that you're trying to make.

But it's not memorable and might be forgotten. So I think you want to hit both, you want to hit a core important message, but you also want to try to create something that is memorable because at the end that's what we were trying to create. We’re not trying just to create impressions, but we're trying to create long-lasting impressions.

Rhys: Yeah, I love that. It's like, OK, if you give a talk or whatever and then people only get, oh, that's kind of a cool visual and then it just leaves their mind immediately. It's like you didn't really do your job or whatever, and so I guess when you think about that fear, there's both the fear of communicating and or of being understood. There's this also bonus double fear, which would be the fear of only being understood for a day instead of for the ideas to go into the mind forever or something.

Cesar: Of course.

Rhys: Yeah, this makes me think too like, for you, a lot of your work it comes across, I mean, you've worked at the MediaLab and you’re now into Toulouse. Your work goes in a lot of different directions in a very interdisciplinary way.

What is the through-line in your mind that ties all of your work together?

Cesar: So there's a few but to me basically then the thing that connects my work is to understand the geography of how knowledge is being produced and how knowledge diffuses and how that impacts society in numerous ways. So for instance, if you look at the work that I've done on economic complexity and relatedness, that it's all about how we can use administrative records such as custom data, like for imports and exports or employment data for industries and occupations, or patent data or research paper data to understand the geography of knowledge and its dynamics.

But if you look, for example, at the work that we did with Pantheon. Pantheon is a website that basically helps visualize a data set that looks at cultural exports. It looks at the famous people produced by a country rather than the products that the country exports, and in that case, we have to, for example, create that data set using NLP and apply that on Wikipedia to structure the data. Then we use that data to study the effect of, for instance, technologies in the creation of knowledge.

We found very strong effects with the interaction of printing, radio, film, television and how those technologies change the type of culture that is produced at a time period and remembered until today. We also use that data set to explore the role of languages and translations in the diffusion of knowledge and information across the world. In that context, I think from that core we start exploring different ideas. So, for instance, how humans judge machines started exploring also the idea, well, as we develop knowledge and we create a more complex society in which the technology that we create starts to transcend the idea of just being a tool and starts becoming a member of teams. How do people react to those and how do we accept them into our groups?

That's a little like what the book explores by comparing how people judge machines compared to humans doing the same thing. So to me, that's kind of like the core, but at the same time I'm never, I like exploring new stuff, I'm a person that is very open to spin. So if I find a new idea, if I find a new collaborator that I'm having fun talking about and working with and there's something that we can do, I do believe in the journey itself too. You know, not just on the destination. So I would deviate and a lot of the things that sometimes I find have been more gratifying have come from those deviations.

Rhys: Yeah, that's interesting. I think that I mean, well, some of the work you've done on geography of knowledge was really cool. Just to highlight one piece of that for the listeners. You can look at how many people are cited in a Wikipedia page, like how many Wikipedia pages there are for people, and before the printing press, there was not very many. Then after the printing press, oh, wow, now, we as a society remember the people more. Is that kind of one of the takeaways? So it’s like getting in there, right?

Cesar: So, for example, prior to printing, you find that most of the people that we have in our record like Wikipedia, and we use our criteria in which we focus only on people that are present in many languages to try to find people that have a level of global fame. It's people involved in religious activities or people that basically went or on a political position where there's a king or a monarchy or some sort of a governor or person of power at the time. After printing, what is interesting is that not only the number of people that will member jumps, meaning that our collective memory doubles very quickly but also the new people that we remember are people in different categories. So the interaction of printing was the start of society remembering composers and artists and astronomers and mathematicians.

As printing develops and we go from printing books to starting to print journals, a process that takes a couple of hundred years to take place actually. Then the sciences start to bloom when you start seeing sciences grow by dividing the natural philosophy sciences into physics, chemistry, biology, and so forth. Then all of a sudden you have the interaction of film and radio and that completely shifts the arts from the playwrights and the composers to the performers to the actors to the musicians to the singers. It resonates very strongly with this idea from Marshall McLuhan that the medium is the message, which is that as you develop new technologies to communicate things, there are certain messages that are better adapted to those mediums and those are the ones that spreading further and further so.

Actors existed at the time of the Greeks. Actors existed, of course, that Diamond Shakespeare, but famous actors only existed when there was a medium that was able to capture the performance, when there was a medium that was only able to capture the play, it was the playwright, the one that became famous. That is really interesting because it tells us that these technologies are truly transformative they shape our society, they really show that there are these breaks in our history that are the result of changes in communication technology.

Rhys: Yeah, I love that. To double click on that for a second. This is a very macro question, but good luck. From a medium is the message perspective. What things are optimized for the Internet?

Cesar: The Internet is weird because it is such like multimedia. I know that sounds like a really ‘90s word, but it's true. You know that. You have audio. You have text and so forth. The web, nevertheless, that's a lot of text like when you look at a medium like Twitter or Facebook and so forth even though there is video. A lot of it is like little headlines and so forth. In our data set, we start to see a few things, some of which might sound ridiculous, but I think they show that at least the technique makes sense.

For instance, with the rise of the Internet, we're starting to see in data set of course, like famous YouTubers. But for instance, you also see like famous pornstars there is a rise and whether you like it or not, the Internet and porn had kind of some sort of connected history, like a lot of the compression algorithms and everything we're developed originally in the context of that industry. Also, I think the Internet has given some sort of revival to a different type of politics that we're seeing across the world.

I say that today we're in this sort of weird Twitterocracy because of the fact that the politicians have direct access to their audiences. Vis-à-vis 30 years ago when it had to be through interviews or through a TV channel developed our politics that is much more untapped, much more aggressive. In the United States, we basically saw that when Trump was removed from Twitter that was kind of real, like Twitter sometimes in a way an action that seemed to be more decisive on changing the mood and the landscape than the impeachment attempts that didn't go through. So it's kind of weird and we're seeing these pop up all over the world.

So I do think we're in this weird space in which we have a technology that allows people to have direct instant fame. The fame is not mediated through like work or through people other than you writing about it and that has generated kind of a very strange dynamic on people that become famous by sharing opinions, sharing conspiracies, people that also will become famous because they're good at explaining difficult things and they provide a good service. So it's hard to judge when you have such a mixed bag, but I think definitely it's not the same as what we had 20, 30 years ago.

Rhys: It is funny. Yeah, as multimedia. All these multimedia people on the information superhighway. It's crazy out there, you know. But I agree with a lot of what you're saying here, which is the disintermediation and the ability to speak directly to your audience is really crazy. It also makes me think something else that you talk about in Why Information Grows is this leveling up of networks as if we think about the various information processors in the world and whether it's our bodies as an information processor and reminds us an information processor and then we go from a person bite up to a firm bite where you have all the things that we can keep in our head are not enough and so we get into these firms with lower transaction costs and higher trust in order to process information better.

Then you can kind of zoom out again another level to, oh, what about these networks of firms or whatever? So I think that's something that I want to get your take on is on the Internet. Do you see any kind of new institutional forms like that showing up? Or how do you think about that side of things?

Cesar: I do see attempts and I'm interested in not only understanding them but participating in their creation because I think from the beginning the Internet has always had a component of a democratic utopia in some way. I remember when I started to use the Internet like in ‘94, ‘95, and a lot of the allure of the Internet was like this is a space that is in some sense outside of the traditional establishments where you can create your own server if you want. You can create your own email server if you want. You can create your newsletter and it was an Internet that was basically based on protocols. It was much more decentralized than the Internet of today.

But as the Internet developed commercially that platform that had been generated in some way got cleverly privatized by companies that dominated key applications, search, social, mail, maps, and so forth, and the Internet transforming to an Internet of protocols into an Internet of platforms or aggregators which are technically different that now control a large amount of the traffic. So they're sure there's a lot of websites on the Internet, but big websites, the big platforms explain a lot of people's traffic.

A lot of the thing goes there and if you're not getting traffic from search or from social, you're not getting traffic. It's kind of like a tough world out there now. So in that context, I think the Internet became a little more centralized and there's power or those platforms when it comes to the decisions that they make about how they operate, who gets to participate, what forms of participation are allowed, and so forth. At the margin of that, though, there are still people that I think have stuck to that utopia of the Internet as a medium for democracy.

During this semester at the Center for Collective Learning, I run a seminar series. I interviewed people. I invited people to participate in the seminar like Audrey Tang, who's the digital Minister of Taiwan. The people from the Five Star Movement in Italy that they have this platform you saw that they use. Some people from Spain that have created the City Madrid, which was a big participatory budgeting initiative, hundreds of millions of euros and a little bit of what I see is that there are a lot of isolated efforts and their attempts to create these new institutions.

On the one hand, I think where I live in underdeveloped, when it comes to like the HCI of those institutions, I think a lot of the friction sometimes come from bad user interface, and that's something that in my book we tried to explore now doing good user interfaces is very costly because you have to make a lot of decisions a priori that the users are posteriori. They don't appreciate much because for them the result is that it's easy. It feels that it was easy to make but I always tell people like a BMW is easy to drive but it's not easy to build.

So you have a little bit of that. I think there's a lot of ideas and unfortunately, when you go to the more official channels a lot of those efforts, or the builder sympathetic to those efforts, they think that they can create those institutions of technologies by decree or buy a green on some course, values and ideas and from there to an actual implementation that is engaging, useful, fair. So for this such a long road that we need people working on the other side of trying and erring and exploring and building and there's few. But they're not zero, they're growing, and it's also something that I see that is very much alive among younger generations that are digital natives each time we know more and more and that tend to believe that the Internet is kind of not just another medium but an integral part of society that should be, of course, part of the way that we make collective decisions, let’s say democracy.

Rhys: Yeah, I love that. It's like if you have some Gen Z kid who grows up and they're doing their thing, they're part of a Minecraft group where everybody built co builds things together and they're part of these Discord servers, and they're like doing all these things that are very like, and they’re remixing on TikTok or whatever and then they grow up and they turn 18 or 21 or whatever, and they're like, oh, now, I get to vote and I get to give one bit of information every four years. They're like, is that really how this works? So yeah, I think that they get disappointed by that.

Let me ask you one other question on this. Then we can switch topics. I agree with your version here, which is OK. There's this like democratic utopia like bottom-up Gov tech, Civ tech thing this great way to kind of process information better as a society, especially around this kind of governmental side of things. The other side here is these new kinds of institutions, and I'm not sure if institution is the right term for them, but something like a hashtag or a movement. Whether it's #MeToo or #BlackLivesMatter or something like #MAGA or #StopTheSteal where people are kind of they come in because there's lower friction.

You can understand as part of these networks and then you get that like thing that new kind of being that new shared myth can kind of help propel folks forward. There's like that side of things, and there's also like these outside things like you're probably part of many. You're probably part of complexity Twitter, or economic complexity Twitter, or AI Twitter or whatever, and that's kind of like a new social. It's like a new group that didn't exist before. It's like a network or something like that, and so I'm curious about these new kinds of, in the past, we've had firms and nation-States and markets and stuff, and I just think that there might be some like network native kinds of institutions that will pop up that use lower transaction costs and higher trust with the kind of new medium. Does that make any sense? Do you have any thoughts on like network first?

Cesar: Yeah, I do believe, of course, that there is some sort of online geography that doesn't map immediately to the physical geography. So we participate in groups that are tightly needs in terms of relationships but can be quite distributed in terms of their physical location. But the technology that we have right now I think are great at some things like, of course, there is the ability to express yourself, but then they're so bad at other things like trying to aggregate those preferences, doing some minimum form of participation, doing proper deliberation and debate that as those movements grow and the only way for them to survive is to find a very simplified topic or languages and in some sense that's kind of like how the larger idea that has more neurons gets killed.

At night, I've been reading my daughter Animal Farm from George Orwell. I read that book when I was a teenager and I've been reading again and honestly, it's such a fantastic book. Basically, as the revolution moves on in the farm, Napoleon and Snowball, which were like the two pigs that are running the revolution in the beginning, have this set of rules about the principles of animalism. They soon discovered that there's some animals that could learn to read and write, but other animals that could learn the alphabet and their other animals like the sheep, that were really stupid and. So they had to start to dumb down those principles and it ends up being two legs bad, four legs good.

It starts in the book with the speech of May or the pig that makes the big speech in Chapter 1 and dies in Chapter 2. It starts like this more nuance. Of course, revolutionary idea then gets done down as it needs to be spread within Animal Farm Society and I think, of course, all was doing a satyr of what he observed happen in reality. I think Twitter has changed the world. Facebook has changed the world but that mechanism seems to be still in place.

Rhys: Yeah, I totally agree with that and I think on two points, one is as you say, it's like we're so good at like everybody can express themselves and kind of yell about whatever they want to. But then how we actually aggregate that or curate it is so bad. We have this beautiful, obviously, it's like a firehose of information. But then when you actually try to get that down and have it be helpful for coordination or aggregating information, all these things. There's some beautiful things like Wikipedia that do it in a really good way. But all things considered, I super agree that we're very we're just at the beginning there and we're only at the place right now where because of our tools it kind of gets cut down into a very small space that doesn't have much nuance. So, well, curious about how that evolves over time. Let's chat about How Humans Judge Machines now, tell us a little bit more about the main thesis of the book, and we'll kind of dive into a little bit more about it.

Cesar: So basically the book was motivated by the fact that a lot of people were talking about the defects of AI in society and to me it felt like that conversation was incomplete because a lot of those judgments that were being shared or communicated didn't have proper counterfactuals. So like in reality I'm a big believer that things always have to be understood in some context. They have to be understood in comparison to something. So people would say, well this machine is really harmful or this is bad to side but is it fair?

But wait, those same tasks are now being executed by humans. How are the humans doing it? I think the human make the same mistakes. Do we judge them equally? So because of that, I decided to create a small team. I did those experiments when I was at MIT. I had two post-docs who were social psychologists and we created a list of a little bit over 80 experiments in which we had a scenario in which a machine or a human perform the same action. Then we have people react to that scenario and we use our clinical trial type of study design in which we randomly assign people to the machine condition or the human condition.

So to give a super vanilla example, imagine that there is an excavator that is digging up a site for a building, The site contains a grave and the excavator digs it up. Well, there were 200 people that thought that was the action of a machine and there were several people, 200 people that thought that was the action of a human and those people are all independent. They're not deliberating, they're just giving us the reactions. We look at those averages, and we see, OK, did people find the action of the machine to be more harmful than when the same action was performed by human? Or did they find it more morally incorrect? Do they feel more identified with the human action, thinking that they would have done the same if they would have been in that situation?

We find differences across the board in many experiments. In some, we don't find differences. At the end of the book what we do is we put all of the data from all the experiments together to try to understand if is it just like a preference for humans over machines or if there's something deeper going on. We find that there's something much deeper going on, which is that people are judging humans and machines through different moral philosophies. So we tend to judge humans with a moral philosophy that would be called more Kantian. Basically, it's more about your motivation. More about your intentions and hence we tend to forgive humans in accidental scenarios.

But we tend to judge machines through a different moral philosophy that is utilitarian, that is more consequentialist. So we only care about the outcome. We don't care about what the machine was trying to do or what it was designed to try to do. It's only whether it fails or not. So we have these different moral philosophies and that helps explain other differences that we observe in the scenario. For instance, people are unforgiving of machines. In scenarios that are clearly the fault of an exogenous factor, we had scenarios where there was a tsunami that basically destroyed a town or a car that has to swerve to avoid a falling tree so that exogenous trigger people still tend to be less forgiving to machines in accidental scenarios because they're judging it through a different moral philosophy, that it's consequentialist rather than cantion.

Rhys: Yeah, I love that. I think that makes a lot of sense. It's like, hey, we're like, oh there's a human over there. Oh, they didn't really intend for that to happen. They're just trying their best or whatever, and I can understand the mistake versus the machine. It's like, well, the machine messed up. I think it's not like the machine was trying its best or whatever like that's a weird frame to take. Do you have an instinct of it like you got this result, this normative result that was like, OK, humans, we judge each other based on this Kantian intention thing and we judge humans based off this utilitarian outcome thing. Do you think that is right or if you were to, if you think about the outcomes of this, are you going to try to, if possible, shift how we think and how we judge machines? Or if you could, how would you do that? What would you want the goal to be?

Cesar: Yeah, I don't think that per se that those things are right or wrong. I think they can be used rightly or wrongly. So let me give you some examples. So in Chapter 7 of the book, after having done all of the data analysis, I take the liberty to speculate a little bit, and I'm talking about implications, and one of the things that I talk about is, OK, so beyond humans and machines, we have organizations, organizations are another type of structure that also makes decisions, execute actions, but it's not like judging an individual when they're judging the Coca Cola company or the US government or the US Army. It's very different than judging an individual.

For instance, when it comes to government approval, government approval is basically a synonym of presidential approval in countries that have a presidential list type of regime and that is sort of a little bit weird because in some way the organizations are so large that a lot of the time the outcomes might be very different from the intentions of the leaders or the people involved. So to me, it seems that we tend to transport our intention-based form of moral judgment organizations. When those organizations are more political and they have leaders to which where ascribe a lot of responsibility.

So my speculation is that we judge something like the US government using a similar judgment that we do for humans is like, no, this policy proposal it's wrong because I don't trust Trump or because I don't trust Biden. So basically I'm interpreted from the perspective of the intentions that I associated to that person or tool. That party I don't trust they’re conservatives. I don't trust the Liberals, but what we're thinking about a company in the private sector where we all know who the CEO is and everything. We tend to use a more utilitarian approach, in part because it is easier to measure the quality of the product. So hey, like Disney movies are really good man, they do a fantastic job in animation, storytelling, they are tight.

Like Apple makes great products and whether it is Tim Cook, the CEO or not, in some sense you don't see the product to be like, it's because, you know, like Microsoft doesn't do such a nice operating system because they’re not trying to do it right. No, it's like the outcome at the end is hard and it's also more directly measurable for you as a consumer while measuring whether they're getting good government services or not. It's a little more difficult and that's why we might be using these different modes. So I go a little bit into that. So now the question is, should we be judging governance by their intentions or should we judge them by the outcomes?

I think that's a very deep question because intentions are much easier to fake and communicate and outcomes required capacity. So sure you can have someone that comes and says, oh, yeah, I'm here for everybody that is poor and desperate and it tries to touch all your sensibilities. But when push comes to show and they have to do something about it, they don't have the capacity and you can have someone that is more of a policy wonk that knows all of the constraints and how difficult is to do something and it will say, well, actually, we kind of just give you free everything for everyone, but maybe this is what we can do and that person politically might have a lower chance of becoming elected or moving up their political ladder because they are going more into the technicalities instead of just trying to talk to people intentions and so forth.

We might end up then with lower quality governments because the institutions get captured, but those that are are able to touch our sensitivities and not necessarily by those that are able to achieve the outcomes.

Rhys: Yeah, it's kind of a sad like a classic government thing or it's like they're signaling blah blah blah or they're saying blah blah they're making these pledges and then, in the end, it's what is actually happening and it is especially with something like the Internet where it's just like you can state directly to your people and kind of a populist way. Oh, we're going to be amazing for all of you. Then, OK, what actually happens on the policy side? So it does seem like it would be nicer if we judged folks based off of, you judge governments based on their outcomes, and this gets into things like prediction markets and stuff like that where you would be able to they say and make pledges and then might get incentivize or distance and device based on either prediction market outcomes or things like that might be one OK way to do this.

Thinking about the moral side of things here and it was cool that in your book you use these hate moral foundations a lot in terms of how humans judge machines. Were there any big differences you saw in kind of, there's like the care and harm-based side of things, and some of the more individual side? Then there's these more kind of collectivist or group ones, like sanctity or loyalty or authority. Were there any kind of big differences between those two that you saw in the data?

Cesar: So yeah. That's an interesting question. On the one hand, we find, for moral dilemmas that involve harm. People tend to immediately be very unforgiving towards machines. So when it comes to harm, the machine bad, human good. When it comes to fairness is a little more complicated because when people look at a situation that it's unfair and involves a human. They tend to judge that human very harshly. They consider those things to be very bad. They consider those things to be very intentional, and in those cases machines sometimes get a little bit of a break. Time a little bit but a little bit of a break.

So let me give an example. We had one scenario in which there was a human or a machine that would write songs for artists, you know that would kind of write songs for singers and later a journalist discovers that the algorithm or this songwriter had been plagiarizing songs from lesser-known artists. In that case, people basically see the machine is like, OK, this is like a machine learning case go bad and they tend to forgive it compared to the human that plagiarized. In that case, it's sort of like, hey, what you were doing, you know it's bad while in the case of harm you don't see the same.

We also found, for example, in some royalty examples, we find that people also tended to judge humans more harshly. We have a scenario that is borrowed from Haidt’s work in which there is a cleaner or a robot that uses a flag to clean a bathroom floor. So that's sort of like a very complex type of moral transgression. You have to understand that this flag from this country is bad to these people if it's used that way and so forth. In that case, people judge the action of the human has less morally correct. But we did other scenarios that try to explore transgression that also touch upon patriotism, nationalism in some ways. So for instance, we had someone interrupting a national anthem on a sporting event. In that case, we find in a very small effect almost it doesn't matter and then, for example, we had another one in which we had someone like playing the wrong national anthem for our country in an international sports event, in that case, there was no difference like people didn't see the action of the machine as were as that of the human and they were actually not very upset at it.

So there is a lot of variation to explore. I think still, but definitely, the different moral dimensions did affect the way that these judgments behave and in the case of harm, furnace, pretty consistent effect. In the case of loyalty, authority, purity, not so much, and also it is much harder to construct the scenarios in which someone is betrayed by an algorithm. So they're harder to make at least sound plausible and I think that's part of one of the things that we struggle with.

Rhys: Yeah, that's interesting. It just makes me think too of like the long term view here where things like loyalty, authority, sanctity kind of were created within society and humans as a result of our kind of groupishness desire to create these both tribal groups and then city-states and these larger entities where it was like, oh it is good to be loyal. It is good to be to defer to authority in some ways, and I don't know it makes me think about the long-term future here with machines.

Yeah, it's harder to kind of loop them into those more complex scenarios and that like, maybe, as machines continue to do more and more that maybe some sense of the groupishness or the moral capital there will kind of start to diminish and we'll just become like utilitarians all the time. So I'm not sure about that.

Well, one final question I have for you with How Human Judge Machines is, I noticed that you're giving away the PDF for free on the website, but you're also selling the book through MIT Press, and the book, by the way, these beautiful, these illustrations talking about the visual side of things. It's really, really cool there. How were you able to, you know, giving away a book for free as a PDF and also selling it usually doesn't work for the publisher. So how was MIT Press, OK with that, or why did you choose to do that?

Cesar: I had to pay the publisher the cost of the first run. So basically, I said I want to do the book honestly. It was a bit complicated because I wanted to get the book out quickly because I have done a lot of experiments. I could write fast if I had the chance and basically what I was telling them is, look, I want to have this book done by the end of 2019.

Then I pay for the run because the only thing that I'm sort of getting from you guys is the peer review service like the bull display review by anonymous peer reviews outside. I have to respond, so I respond to all of that and I do that. Unfortunately, even though I talked with the publisher and stuff like that and that was kind of the book was quite delayed in the time frame that I wanted. It should have been out before COVID.

But, in any case, the deal was, they were interested in the book. It was approved through peer review. They made me an offer. I told them I want to also have it available online and they say no, we cannot do this because this costs money and we have to make the money and I said no, I understand, but then how much money and then they gave me a number. And then I pay that, and then they have to kind of agree to that.

Rhys: That's funny, yeah, you're like, OK, it costs a certain amount of money, how much? And then if it comes from me or from someone else like that it can come to you. That's funny. That's cool that you, I mean that is really putting your money where your mouth is. Was it like a desire to have it be open source or was it the speed that you wanted it or why would you pay money to give your book for folks up to free on the Internet?

Cesar: I think there's a few things one is, of course, I do like making it accessible, and in particular like to me. One of the things is that I grew up in Chile where books are expensive. They almost never get there. Some books would simply just don't get there. So it's a world that I grew up buying MATLAB on the sidewalk as well as the video games and all of that stuff.

So to me, I always want to think about what I'm doing. I think kind of like beyond the US and European audience, and I think like hey, maybe there is like an 18-year-old in Indonesia that like this topics, might be interested because he saw a talk of mine on YouTube or something and if he wants to read the book, I don't want him to get like a pirate version. If he can get like the legit version and I can make that happen, you know, that's good. I wanted also to see if by doing this I could get the book to spread more broadly and to reach a large audience.

At the same time, I think the book has been well-received, but I don't think it has the impact that I expected. It has been a hard year to promote books because there's a lot of books out there. There are no speaking events. Everything is online. So that has been complicated, but I think that the fact that people like what it's out there, a lot of people are interested in, it means that little by little the word of mouth is going to hopefully make its job and the book, hopefully, is going to become a reference for people that say, hey, I want to know how people judge humans and machines. In this scenario, let's see if among these 80 something I find one that I can use as a story that I can put in my journalistic article, or that I can side on my paper, et cetera.

Rhys: Yeah, I like that. I think I commend you for doing it. I think it's a really cool idea and I haven't seen that many. You know, sometimes people will do a sliding scale if they sell it through like Gumroad or something, but this is I think a pretty unique thing where it is going through a big publisher and you've kind of paid the publisher in order to have a PDF copy online. Yeah, so I think that's cool. As we begin to wrap up mode here. Let's talk about Why Information Grows, a book that you wrote in 2013 or something like when it’s first released?

Cesar: It was published in 2015.

Rhys: Yeah, 2015. OK, yeah, I think it's a delightful book and I especially enjoyed some of the talks around and I think that we've been chatting around it in the first part of our conversation here. But one thing just kind of in general that I want to ask you about is how do you think about information storage and information processing of a, like human and social systems or these informational systems?

Cesar: So in summary, what the book tries to do is two things. On the one hand, it tries to generalize the idea of computational information by describing it at multiple scales and in multiple systems. So like information is a very fundamental physical property is found in thermodynamic systems is related to entropy, which is a thermodynamic state variable, and then years later, we discovered in the theory of communication from Shannon and Nyquist and eventually we now understand that information is related to the way in which we would encode something in physical order.

Whether it's little magnetic domains in a hard drive. Where it is the sound waves that we transmit as we speak across a room and there is sort of like this very fundamental physical quantity that we are producing in great quantities in society. Biology also produces a lot of it, like DNA in a way is a molecule that is very rich in information and that to do so it requires certain cost like you need to in some way expend energy and exclude entropy to produce information. It's kind of like running a little bit like a refrigerator, you know, that you're running kind of like a little bit thermodynamics backward to create order to keep it organized and we do that all day inside our bodies and we do it in our society as well.

So one of the things that the book tried to do is to say, OK, look, this is how people think about information when you're thinking about like physics like spin glasses and molecules and very simple things. Go a little bit through biology, but it moves quickly into society and into the idea that a lot of our economy is based on creating these packets of information. But packets of information that are not just there to communicate meaning but to communicate the practical uses of knowledge. So when we build a video camera, a projector, a telephone, a jacket, a pair of shoes, a dishwasher.

These are also things that we create by going against thermodynamics were kind of like locally reducing entropy to create those objects. So a dishwasher is in somewhere like a molecule of DNA. Thermodynamically speaking, and by doing those things we can communicate the practical uses of our knowledge. We can communicate the ability to do things that seem like magic, to fly, to communicate a long distances, to travel really fast across space, to produce vast amounts of food, all of those things. The other thing that the book does beyond connecting this idea across scales and systems is to say, well, when we're at the human and particularly the social scale, there are certain limitations that we need to consider that involve the accumulation of the knowledge that we need to produce that information and these limitations are there because knowledge is not like just one thing.

It's not like a liquid that we can accumulate more of it. It's something that is extremely complex. It’s very specific, you know, very unique, and it has a lot of like the specific parts, like a giant alphabet order or a gigantic periodic table and to accumulate all of those different pieces that we need to do something, we need to accumulate the knowledge in networks of people, because simply individuals cannot learn enough. Then I use that as a point of departure to start to explain, well, how do we form networks and how large networks can be, and how that depends on transaction cost, trust, and other things and the ability to create these large networks. It's key to be able to accumulate large models of knowledge and to then create these complex things that few people can create and that can make your geography richer.

So then it goes from there to explain the differences in wealth as differences in this capacity to accumulate knowledge in social and professional networks. That's what the book is about, is about the geography of knowledge but it's starting from the physics of information.

Rhys: Yeah, I love that and it’s cool, you know, that connecting it across disciplines and scales is a beautiful way to do it. I think there's a lot more interesting stuff to be, I just love that frame of thinking about these networks of humans as ways to kind of process information better in order to create this knowledge that can't just be stored in human brain. So I think it's fascinating, interesting stuff there. As the final question here, Cesar, so for listeners definitely check out How Humans Judge Machines or Why Information Grows and again, How Human Judge Machines just check out some of the visuals for it online. It's really cool what you were able to do there.

Is there anything else that you kind of recommend to our listeners, either place to find you on Twitter or anything like that?

Cesar: Yeah, of course, I'm not that hard to find online. But, yeah, I would say to visit the Center for Collective Learning. This is my new group that I started. It's a continuation of the collective learning group that I had at MIT, but now we're growing into a center and we’re starting to explore new ideas. Some of them continue the work that we've done on economic complexity and geography of knowledge. But we're also interested on detail democracy, and that's I think where you're going to probably see a lot of things coming out later this year. I'm doing the next one.

Rhys: That sounds amazing. That's even just good for me to know. If you're interested in collective learning more generally, either on the economic complexity side or on this kind of Civic tech, Gov tech like networks of folks processing information. Boom, check out Cesar’s place

Cesar: Thank you.

Rhys: So thank you again, Caesar, and thank you, listeners, for coming by and goodbye everybody.

Cesar: Goodbye.