History of Science and Technology Q&A (October 25, 2023)

Hello, everyone. Welcome to another episode of Q and A about history of science and technology. And I see bunch of different interesting questions saved up here. Gosh, where to start? There's a question here from M. Rudo about can I talk about the history of Quicksort or of Huffman coding? Well, these are sort of two classic kinds of algorithm ideas, very different kinds of ideas that both originated, I think, at the end of the 1950s. So that kind of time frame. Let's talk about what they are. So sorting is. You've got, let's say you've got a bunch of index cards and you want to put them in alphabetical order. How do you do it? I mean, these days you just say. Well, you just click on the column header of your. On your computer and you say sort these things. But how do you actually do it? Well, back in the day, people often did have to sort things manually in order. And I remember doing that when I was a kid, needing to take things that were, for example, on index cards and sort them. And there are kind of ways that one does that by hand that are kind of feasible for a person to do. And then there are quite different ways that one can imagine having a computer do those things. And I think the most common kinds of things one does by hand, you kind of lay them out, you start flipping pairs of cards. So you say, these two aren't in order, let's put them in a different order. You're kind of doing what in computer science gets called a bubble sort, where you're just sort of taking neighboring cards and you're saying, are these in order? No, let's put them in the right order and so on. Well, this problem of putting cards in order was something that showed up in all kinds of places. A lot of the early sort of tabulating machines were all about sort of counting things from cards, doing things with cards. That was all stuff happening in the early part of the 20th century, before there were electronic computers. I don't know the precise motivation for the invention of QuickSort, but QuickSort is an algorithm which is kind of a. A randomized algorithm. It's an algorithm where you're almost, you're picking things at random, you're breaking the list in two, you're sorting the sub lists, then you're sort of merging those things together. And it's sort of an interesting idea that you can get somewhere by kind of making a random choice and then going on from there. This idea of let's make a random choice to see how things Work out. This is the idea also of the so called Monte Carlo method, which was used in World War II for the Manhattan Project for using, for studying things like neutron diffusion. You try to work out what will happen when all these neutrons are running around in a piece of fissile material. And you say, well, let's just see what would happen if we were to randomly pick which direction these neutrons go. And we have enough neutrons that what will happen in actual neutrons, where we have lots of them that are sort of moving at random versus in our simulation, will agree. Actually, this idea that one could sort of estimate things by just doing random choices, that turns out to be a pretty old idea. I know that I happened to find work by Kelvin from the, I think 1880s or so in which he was trying to estimate things related to thermodynamics and essentially did this idea of just let's try a lot of different paths. And he talked about, I think he used a deck of playing cards or something and shuffled them and picked things. And I suppose this idea, that kind of randomness as a way to work things out, that's something that existed in Kelvin's work at that time. That was a much earlier use of the Monte Carlo method that I knew existed. I thought that was a method that really only come into existence a little bit before electronic computers in kind of the 1940s type time frame. I mean, there's other work that was done on thinking about kind of the effects of randomness. Back Even in the 1600s, the Bernoullis and so on were studying games of chance and trying to make, and trying to sort of develop the early ideas of probability theory. And again, it was even though you've got probabilities and random things happening, you can make definite overall statements. But using that as a way to actually calculate an exact result, that was something that I think didn't really emerge. I mean again, Kelvin and what he did in the Monte Carlo method was sort of doing that you can sort of estimate things by just taking a bunch of random choices and looking at their average. But quicksort is yet a different idea where you're just sort of picking something at random and then getting an exact result of how the list is sorted, even though the algorithm by which you get there involves randomness. Now, there are plenty of other algorithms in modern times that involve all sorts of randomness. There are random ways to estimate whether numbers are prime, where you're kind of getting the final result right. There are ways of trying to figure out things where you're sort of trying to find a path through a graph or trying to prove a theorem and so on, where in the end you get sort of an exact result, but where you pick something at random to do it. I think it was probably 1959-1960ish timeframe that quicksort was invented by a chap called Tony Hoare, who's still around and who I'd been meaning to have an interaction with. Actually. I believe he was working at a company called Elliot Brothers, which was a company that I'm more familiar with than I might be, because it's a company that ended up getting into the computer business in England and actually made a computer, the Elliot 903, which was the first computer that I used back in 1973 or so. Elliot Brothers actually was a scientific instrument maker that had existed probably since the beginning of the 20th century, making all sorts of things. They probably made slide rules and stuff like that. But anyway, they got into computers and one of the people who worked there was a chap called Tony Hoare, who was sort of a mathematically inclined person who ended up inventing this quicksort method. The. So that's the story of that Huffman coding. Boy, you guys are really testing my memory. Sort of. The origin of that story had to do with the question of when you're sending messages, how much redundancy is there in that message. So when you have a piece of English text, it's a pretty good guess that if you have a Q, you're going to follow it with a U. So you don't really need that U after the Q. And actually, more than that, if you take a piece of English text and you start just blacking out letters, one can still pretty much read that text. When you black out up to about half the letters, you can kind of guess what the words were just by seeing what the pattern of letters was, even though a bunch of them have been blacked out. So what happened during Second World War? People were very interested in that question because it had to do with if you're trying to do cryptanalysis, if you're trying to sort of break a code, you want to know, if you start filling in, if you get sort of a candidate for what the unbroken code is, is that plausible or not? And how much do you need to know of the. In order to be able to figure out whether you are successfully breaking the code? How much if what you get, you say, I've got a key for the code, you get A bunch of sequence of letters. That sequence of letters couldn't possibly be English because the kind of statistics of the letters are wrong. Then you say, oh, that's not the right key. Let me go on to the next key. And so an idea of cryptanalysis is you keep going until you get something. You don't know what the actual message was, but you can ask whether what you get out is statistically plausible to be the message. And that idea and this question of sort of how much data do you need? If you're going to be able to figure out a key of a certain length, how much text do you need to even be able to know that you've got that key? Right? And how much could you if you didn't? And so that turned into During World War II, Claude Shannon was working at Bell Labs and was working on a speech encipherment. There's sort of a mysterious meeting that happened at Bell Labs, which was then based in New York City, between Alan Turing and Claude Shannon in the middle of World War II, supposedly to talk about kind of the, I think it was called Delilah, the speech encipherment system that was used, I think a few times. I think Churchill and Roosevelt had a chat over the encrypted phone system that had been set up at that time that was kind of an analog encryption system. But somehow Claude Shannon and Alan Turing met up. Alan Turing was sort of famous later, after the work was not secret anymore for having worked at Bletchley park and having invented kind of a statistical method for, for thinking about how to do cryptanalysis. And sort of that's what he did. And meanwhile, Claude Shannon, I think during World War II, invented sort of similar ideas. And then after the war ended, I think 1947, 1948 or something, Shannon published a very famous paper that was sort of the Origin of Information Theory. And kind of the idea of that paper is to talk about how do you measure the redundancy of something like English, how do you make mathematical this sort of observation that you can blot out some letters and still roughly know what the words were? And for example, another example that Shannon actually used in his original paper is when you make a crossword puzzle, if kind of any possible sequence of letters is a possible word in a language, then you can expect that you can make a crossword puzzle where every single place in the two dimensional array of letters is filled in. Because after all, any sequence of letters makes a word. On the other hand, if you have a language that's very redundant, so very Few sequences of letters make words. You'll end up with a very sparse crossword puzzle. And English is sort of a critical case because it's about half redundant. And that turns out that if you want to make a crossword puzzle in D dimensions, you need a redundancy of about 1d to make the crossword puzzle work. So, so, in other words, if every Q had to be followed by U, A, E, D, something or other, then with very little redundancy, with very little choice, you wouldn't be able to make a crossword puzzle where much was filled in, because only very, very few sequences of letters would make valid words. But, so Shannon invented this kind of method for estimating kind of what was the true redundancy of a message or a signal. And that was very relevant because people were interested in, if you were going to transmit something through phone lines or something like this, could you compress the message so that you didn't have to use as much bandwidth, as many bits to transmit down your copper wire for your phone line and. And saving the phone company money because they didn't have to send as much data and so on? And so Shannon kind of figured out that if you have an Alphabet, for example, and each letter in the Alphabet has a different probability of occurring, and then each block of letters, like pairs of letters, triples of letters, and so on, you know the probabilities of all these things occurring, you can estimate the probability. You can estimate the redundancy of messages by looking at those probabilities. There's famous formula sum of P log p, which gives you where P is the probabilities and the sum is over the Alphabet or over the possibilities for what the blocks of letters A, B, a, C, a, D, A, et cetera, can be. And so Shannon had this way of estimating what the effective redundancy of a message was. And that told one something about how much you could compress the message. Now, people who already sort of intuitively know these things, for example, Morse code, invented in, I guess, the 1830s, was something where it was making use of the fact that in English, E is the most common letter, followed by t, followed by sequence of other letters. And when you encoded that in Morse code, an E is a dot, a T is a dash, and that's about as much of Morse code as I know. And S is a dash, dash, dash, I think. So you're using the shortest code word effectively for the most common letter. And so, on average, the typical word will be able to be transmitted in fewer symbols because you're using the shortest code word for the most common letters. And so what Shannon's idea about information theory told one was what in English, for example, what in principle the amount of compression you should be able to get is. And so then the question was, well, how do you actually achieve that compression? And how do you. And sort of the notion which was what led to Huffman codes was, was this idea that you take blocks of letters and you represent blocks of letters by code words of various kinds. So in the simplest case, something like Morse code, you're taking an individual letter and you're representing it by a code word that is like, let's say a dash is a zero and a dash is a one and a dot is a zero. You're representing E by just a single zero, T by a single one, and then S by 1, 1, 1, and so on. So the idea of Huffman coding is to do that for an arbitrary Alphabet and for arbitrary messages and to use the probabilities with which different blocks occur in messages to determine which one should be allocated the kind of shortest code word and so on. And it's just a systematic way that's based on kind of making trees of things to, to develop that. And it was sort of one of the first practical ways to do something like, to make use of Shannon's information theory idea, which had been used sort of implicitly in things like Morse code, much earlier than that in making compressed versions of messages. Now, in the years since then, people have come up with better and better ways to do compression. For example, one very, very famous one is the one that's used in zip, so called LZW compression, Lempel Ziv compression, that is so called a dictionary approach to compression. So the idea there is you just see what are the complete words that occur, and then you say, okay, we're going to have a dictionary and we're going to have, essentially we're going to say the word and as a one, the word the is a two, the word rhinoceros is a 714, et cetera. And we're going to kind of just watch in the text that we produce, we're going to see, well, what actual words appear, what actual sequences of characters appear. And for those sequences of characters, we're going to give them, we're going to assign them an ID where the more common a word appears, the shorter its ID will be, the earlier its ID will be. And kind of that can be done sort of recursively where you're saying you assign code words for sequences of code words and so on. And that's what builds up LZW compression, which is what's used for Zip and GIF and TIFF and so on. Actually, TIFF uses. TIFF has many different compression methods that are part of it. IMAGE COMPRESSION SYSTEM Originally, when fax machines were a thing, there was one of the issues with a fax machine is you're sending an image of a page and most of that page is white. And so you don't really want to use a lot of bits to send something which is purely white. So a big part of the compression that was done there is so called run length encoding where you just say if you've got a sequence of pixels or a sequence of letters or characters, digits that are all the same, you just say, I've got 100 ones or a thousand ones and all you send is 1000 0. And then you might say two ones followed by five zeros and so on. And that's another form of compression. But there's been, over the course of time, there have been a sequence of different kinds of compression methods that have been invented. Famous one in recent times are things called polar codes which are quite unintuitive. They take quite a while to kind of think through what's really going on in this coding method. And that's something that is popular in thinking about. I think it's a standard coding method for 5G telecommunications. Now, it is worth saying that when we talk about compression, compression is a. So there are two basic forms of compression. One is lossless compression, where you are using the statistics of the message to send it with fewer bits than were actually in the original message, but where you lose nothing, where you can reconstruct the exact initial message. And the other is lossy compression, where you reconstruct the thing for the purposes of human perception, for example, but you don't necessarily actually get every single bit in the original input. And that's what's relevant. For example, most kinds of image compression. JPEG is a form of lossy compression. MP3, MP4 are forms of lossy compression. That's sort of a separate branch, which is sort of irrelevant to what we're talking about here. But it's worth realizing that when we talk about compression, when you say, well, the sort of methods of compression that Huffman coding was an early version of are just using statistics of messages. They don't really use anything about the meaning of the messages. It's a quite separate way of thinking about compression to say, well, let me know exactly what this message means. And let me for example, write a program that will reproduce this message. So, for example, PI, infinite number of digits in PI, 3.14159 et cetera. But yet you can summarize all those digits by just saying it's PI or by giving an algorithm for generating digits. So that's a very dramatic form of compression, sort of algorithmic compression rather than the sort of statistical compression. Huffman codes were the first step on the path towards this kind of statistical compression idea based on Shannon information. The. Okay. Mrudo is commenting that TIFF can also be lossless. Yes, TIFF is a big mess because it is a sort of meta encoding. It's kind of often when people say we've got a standard, it's called blah. And you say, great, then we can implement that standard and then we'll be able to import and export files in that standard. You look at the standard and the standard is 500 pages long and you know that it's got all sorts of substandards and so on. And then you discover that, oh, there are these programs that supposedly do this and that, but actually they only do a subpart of the standard. So when you say, okay, we're going to export svg, for example, and yes, some program we export to, it'll be able to handle that just fine. Or in machine learning, export Onx, for example, on an X, you know, it'll be just fine. But then you discover that actually this vaunted standard, people only actually implement little tiny pieces of the standard. And so a lot of the sort of interoperability isn't there. But TIFF is one of these meta standards. Gosh, let's see. Oh, Pruts is asking about 5G. Let me not go into too much technical detail. I mean, I don't know all of the technical detail, but I do know a decent level of it for how 5G works. But yes, 5G is a very digital kind of thing. In fact, all of the. The by the time we even got to 3G cell phones, one's really dealing with digital data. And the method of sort of transmitting all that data, it's usually CDMA code division multiplexing was what was used for 3G. And that's something where you're really dealing with digital bits of data and you're sort of putting together kind of in a cell phone. Cell. Right. What's happening is you've got. I don't know whether it's still 666 cell phones per cell. It always used to be that. And you've got to transmit from the cell phone tower to every one of those different cell phones. And how do you do that? Well, one thing you could do is allocate different frequencies for every cell phone, but that's not how it's actually done. 1. Well, not as most commonly done. One way of doing it is something that at first seems very bizarre, which is every cell phone gets a different random sequence of ones and zeros. They're linear feedback shift register sequences. A different sequence that it's allocated and it's kind of decoding itself based on that sequence. It's decoding the kind of signal that comes in based on that sequence. And there are these different sequences are sort of randomly different from each other. And so every particular cell phone gets to decode the message based on sort of fishing out its template from this stream of stuff that's coming from the cell phone tower. And each one is sort of fishing out a different kind of template. But the templates are sort of random with respect to each other. And so each cell phone just fishes out the message for it and not for the other things, other cell phones. But it's an extremely digital all ones and zeros kind of story. The yes, somebody's commenting on SIP being another one of these non standard standards. That's an everything standard and a nothing standard. In the end, that's a standard for voice over IP type communication. Let's see, Jimmy is asking about logographic languages and compression. I think in logographic languages, well, you know, by the time you've got definite characters, you just say, I'm going to allocate, you know, that weird looking character to this number or this sequence of bits and then you're off and running again. I mean, in Unicode, the kind of standard for the encoding of glyphs, character like things in languages, there are numbers essentially for the glyphs of all the languages of the world. Originally, unicode was just 64,000 different possibilities. And that was in the early 90s when Unicode was being developed. We were a bit involved in that. There was sort of a lot of arguments about, in fact, Wolfram language was Mathematica were kind of some of the very first pieces of software to support Unicode. And we actually contributed some characters to Unicode in the math section of Unicode. But back in the early days of Unicode there was a lot of concern about, well, which languages are really going to fit in unicode. We've got 64,000 characters and for example, Korean with its Hangul characters and a sort of block of four Type characters was sort of a big concern. Is it going to fit in unicode? Is the 30,000 ideograms of Chinese going to fit in Unicode? And those things fit. But some languages didn't make it into Unicode. For example, the cuneiform of ancient Babylon did not make it into the original Unicode. And nor did a bunch of other things, and certainly nor did only the very basic emoticons made it into Unicode. Now, that was at a time before the whole emoji explosion happened. But there wouldn't have been room in Unicode. But you could have represented, you could represent, I think maybe some Egyptian hieroglyphs within the original Unicode. I'm not sure. But anyway, what happened after that time, that was sort of plain zero, as in there were just 16 bits used to represent each character then. And a big pain in the neck for software makers like us to deal with sort of different lengths representing different kinds of characters. Back before Unicode, everybody always represented characters using seven bit or eight bit sequences, which means you get either 128 or 250, 56 possible characters. And then there was a whole complicated zoo of different ISO standards for different kinds of characters. So, for example, ISO Latin 1 was the standard that was used for standard English. And it didn't contain accented characters that appear in other European languages. It didn't contain Greek characters, it didn't contain all these different kinds of things. And so, for example, what would happen is, if you wanted on a computer, you had a file and it had Greek characters in it, it was all Greek, for example, then you would just say those eight bit sequences in that file. Instead of the file being encoded in isolatin 1, for example, it was encoded in ISO Greek 1, which means that the bit sequence, I don't know what it is, you know, the bit sequence that represents 60 or something would be an alpha. I'm making that because I don't know isogreak1, I don't know it's particular encoding. But in any case, then when that file, you've got those eight bit sequences in that file, each eight bit sequence represents a character. Then you say, I'm using ISO Greek encoding. So then I know how to actually render that on the screen. And that was a bit of a mess because you had to know for every file how is it encoded. And that's something you see in email headers and in webpage headers and things is what is the character encoding. And in Wolfram language we have character encoding options and so on. And that's kind of a way that that is specified. And so, for example, I think there was isolatin 1 through 5 and different European alphabets would get into those different isolatin encodings. I think Isolatin I does contain a couple of funky characters that come from kind of Old English or Norse or something like this. I think it has a thorn and an F in it, I think that are not quite that fit in the 256 bit thing and snuck their way into isolatin 1, I believe. But anyway, that all kind of got cleaned up in Unicode. 16 bits accounted for most of the current languages of the world, although some left got left out, as I recall. I think Kamir made it in and I think maybe Georgian, I think maybe didn't, and maybe I'm wrong about that. I remember sort of very heated discussions about which languages made it in and which did not. But that all got solved because it was like, okay, let's have an extra bit and let's have not just plain zero Unicode, but have plane one, plane two and so on, which were represented by extra bits that then allowed you to shift in a whole other 16 bits worth another 64,000 characters. And so now one can have kind of all the cuneiform, which is sort of, okay, hasn't really been used for a couple of thousand years in kind of everyday life. Although it's of interest if you're looking at ancient Babylonian documents and tablets and so on. And as well as that, you can fit all the emoji you want. And that's been sort of a growth area. There's a comment from Nightmare here saying the future is going to consist of languages that are just emojis. Yeah, well, I find that really amusing that it's kind of like history does repeat itself. A good reason to study history is, yes, we had the hieroglyphic languages and now we're back because we've got emoji and even these determiners in the hieroglyphic languages that say this character means this thing is a place, or this character means this thing is a person. We're kind of almost back to that with kind of emoji language back again now, 4,000 years later. So it's kind of fun to see how that works out. Let's see. Well, Eli asks, when did the study of economics form? I don't know the history of economics as well as I should. It depends what kind of phenomena of economics, what it's really talking about. I mean, I think people would usually say Adam Smith and his wealth of nations book was Kind of an early kind of attempt to understand kind of how value works and to get sort of an idea of what's really going on when we sort of think about economic activity. But certainly people have been exposed to economic phenomena and had an intuitive sense of them, certainly in antiquity. I mean, it's sort of an interesting thing that in the Roman Empire there were just published prices for things, and those lasted for long periods of time. I mean, a shark tooth or something, there was a price, and there were tablets that existed that were just the price for X. Now, that's an interesting thing because that sort of denies the possibility of inflation. And I don't know exactly how sort of the Roman central bankers, as they were printing up money, how that really quite worked and how that was sort of kept under control. Because what's been a phenomenon of modern times is the phenomenon of inflation, where the effective value of money goes down and goods get progressively more expensive denominated in the currency. I think people certainly started making. What was it in ancient. Well, I mean, people say Croesus invented money, but I think that's more kind of a mythical thing. But the idea of coinage is very old. You know, minting coins out of metal is something that certainly existed. Well, certainly in ancient Greek, Persian times and maybe even before that. And in different parts of the world, different kinds of things were used for money. I mean, in Polynesia, it was cowrie shells and so on. I mean, sort of anything that is rare and where you can sort of have some control over how it works is something which was a candidate for making kind of symbolic money. I mean, it's sort of a famous thing that for a long time people really didn't understand the distinction between the value of gold, as in how much you could buy and sell gold for, and the coinage that was made of gold or made of silver. And I think it's a sort of famous thing from Marco polo in the 1400s, I guess, when he went to China, China had invented paper money. And Marco Polo found that incredibly bizarre. How can something that really doesn't have sort of, in his view, tangible value, like a piece of gold that you could melt down and use, and it's real value in gold, so to speak, was how could you have a system of money where the money was just paper? And of course, the answer to that is, well, you have to believe in the government. You have to believe that there is some sort of force behind that money that guarantees its value. Now, that same thing came up again in cryptocurrencies. And Bitcoin and so on. Of, well, do you really need sort of a thing behind the money to establish its value? Or is there a way in which something can have a value abstractly? And I think this is still sort of a question in economics. My own view tends to be that as soon as you have enough of a network of transactions, you've kind of established that you can't change that value because it would change too much of this network. It's kind of a thing a little bit like what happens in statistical mechanics. In kind of you have this equilibrium where you've established sort of the temperature of the gas or something, and that establishes something that you can't kind of tweak with by making small changes. And so that's, in a sense, by having enough of a network of transactions, you can have a meaningful notion of value, even though there isn't an ultimate thing down there that you say is truly valuable. I mean, a common idea in economics is that in the end, value is associated with some kind of utility to people. Now, gold, I don't know. You have to kind of squiggle around a bit to argue for the utility of gold because really, gold is mostly valuable because people think it's valuable, and it's also scarce. And if there was, you know, if somebody suddenly captured a golden asteroid, I'm told that no such thing can actually exist because of the way that elements form in the solar system. But somebody brings a golden asteroid to Earth and it's worth a quintillion dollars, let's say. Then what does that mean for the price of gold? I mean, this is what happened, for example, the price of silver. Back in several kinds of disasters with silver, people were like, the price of silver, it's going up, it's going up, it's going up. And eventually people had silver spoons in their houses and they decided, look, the price of silver is so high, we might as well melt down our silver spoons and use that as kind of currency like silver. And that led to all kinds of trouble with the price of silver, which then sort of crashed as a result. But this question about when economics as kind of a science really formed and people would argue, how scientific is it? Now, that's a big question of to what extent is economics really a descriptive field, that it's just talking about the statistics of what happens with the flow of value and money and so on, and to what extent is it a proscriptive field where you start from sort of some axioms about human behavior and deduce what has to happen in the world. And I think there's still a tension between these things. That's something actually I'm interested in thinking about. And I think maybe there's some ideas from our physics project that are relevant for thinking about this in a new kind of way. But the kind of. I would say here's a funny story. Perhaps the school I went to, high school. I went to school called Eton in England. It's a very old school from the. Founded in what, the 1450s or something. Anyway, there was lots of distinguished alumni of that school, particularly the part of the school that is the, the so called college which is just 70 kids at any given time. So about a dozen or so per year. That's kind of the. Was an interesting group of kids. They've done interesting. The cohort that I was in have done many interesting things. But there were lots of well known people in sort of previous cohorts of that particular strangely selected group of British kids. But a story that was told when I was at Eton, this is in the early 1970s. There was this day where so called Threepany Day. And the idea was that the original founder, who was Henry the should know the sixth I think had put a. You know, had given some kind of sort of funding to the school and it said the, the. The people, the scholars at the school should get every year either threepence or a pig. And threepence, three old pennies. They still was. I was there just after British currency had been decimalized, but they still had, they'd had, at least in my time, a supply of old threepences which they used to give out on this day. But the thing was by the time I was there, the price of a pig was very different from a threepence. Threepence was 3 old pence, which is 3 200/40 of a pound which would be 1 80th of a pound which is in dollars, you know now. Well, a cent or two, a few cents. In fact, I happened to just convert some British pounds to dollars. And I was sad to see that the British pound is falling with respect to the dollar. And it's like a. I don't remember. It's almost a dollar, it's almost one to one. When I was a kid, for example, these currencies, and this is another kind of economic phenomenon, these currencies had fixed conversion ratios so you could memorize there were $2.80 to the pound. And it wasn't something where you had to go up to that Airport money exchange thing. And look at the display which was changing every moment, so to speak, telling you what the current market price for that currency was. But any case, back to my story about threepences. So clearly by the 1970s, threepence and a pig had very different values. But it turns out earlier folk who were at the same school, and this was a story that at least I was told, there were two young lads, a chap called Harold Macmillan and a chap called John Maynard Keynes, spelt K E Y N E S and they were students at Eton in the early part of the 20th century, not quite sure when. And apparently on their sort of threepany day, this Harold Macmillan chap was asking the John Maynard Keynes chap, you know, how come you can't buy a pig for threepence anymore? And so this Keynes chap explained that the value of money didn't stay constant while Keynes became a famous economist. Harold Macmillan eventually became Prime Minister of England. And that was. So it was at that time, at least I don't think it was well known. Even in the early part of the 20th century, the notion of inflation was not well understood. It was kind of something that you had to be sort of a future economist, so to speak, to, to have an idea of how that worked. And I think there were sort of attempts to make kind of big theories of economics. Like, for example, Oscar Morgenstern was an economist at Princeton who worked with John von Neumann. This was in the 1940s and wrote this book called Theory of Games and Economic Behavior, which was in a very characteristic John von Neumann kind of way, wheel in a bunch of somewhat advanced math and then imagine that you're just solving a field. And so in that case, he thought he was solving the field of economics didn't really work that way. I mean, game theory is interesting in economics, but it's certainly not the be all and end all of understanding how economic behavior works. So that's a little bit on economics. There's more to say and a little bit more that I know about that history. But it's kind of like even kind of getting systematic data on what was happening economically. People weren't necessarily getting that. I mean, there started to be stock exchanges. That was a thing that was going, I think, in the notion of kind of a market that involved trading commodities and so on. That was a thing. Insurance became a thing. That was a thing. I think in the 1600s that started to develop. There was a lot of sort of high tech put into kind of financial market making. I mean, A famous story is that in Venice, for example, there were all these merchants who were trading and it really mattered whose ship was going to come in, whether you were trading this or that thing. If there was just about to be a ship that had made it in and it was carrying such and such a thing, then the price of that would go down or whatever else. And so one of the things, there's this big tower in Venice that's still there. And the thing was, people would run up that tower and go and see which ships were coming over the horizon so that they could be the first to trade on that information. And famously, Galileo licensed the telescope to the merchants of Venice for all terrestrial uses. And they were using telescopes to go and figure out whether that was around 1608, I think, 1607, maybe, to find out what ship was coming in. Who could be the first merchant to figure that out? And, you know, Galileo and things in those days were very litigious, I suppose, in Venice and so on. And, you know, there was a contract, it said all terrestrial uses. And so Galileo was like. And this story may be partly apocryphal, but it's kind of an interesting story. I've never really checked into whether the details are correct. So I said, well, forget about terrestrial uses. I'm going to turn the telescope to the sky and see what I see there, and I'm allowed to do that. And then, of course, he discovered the moons of Jupiter and so on, and lots of exciting things happened as a result of that. Let's see, there was a question here from Prab about the history of double entry bookkeeping. I mean, double entry bookkeeping is usually attributed to Luca Pacioli, friend of Leonardo da Vinci's in Italy. I don't know the. I think it was sort of a very. This whole idea that sort of mathematicians were consultants to merchants was certainly an established idea. I mean, for example, Fibonacci in the early 1200s when he brought back Hindu Arabic numerals to the west, he was based in Pisa in Italy. And that was sort of a look, merchants, you can do better than working things out with Roman numerals. You can use this positional notation of Hindu Arabic. And by the way, I'll teach you how to do that. And he had this book called Liber Abaci, the Book of the Abacus, that was about kind of how to do calculations and so on, and how to think about arithmetic in this Hindu Arabic positional form. It also famously, at the very end of the book, it has as kind of an exercise example, these rabbits in an enclosure and there's rabbits and every year they have baby rabbits and so on. And that was the Fibonacci series, defined in around 1205, I think, A.D. by the way, the Fibonacci series was also, I think independently discovered in India much earlier than that, as a result of an attempt to enumerate the different forms of prosody, the different ways in which a piece of sung poetry could be presented. Is it long, short, short, short, long, long, et cetera. What are the different forms that there can be. And there was some kind of rule that you couldn't have, I don't know, two shorts in a row, probably two longs in a row actually. And then when you enumerate the number of possibilities that gives you Fibonacci numbers. And that was an even much earlier use of Fibonacci numbers than Fibonacci with his rabbits. But the point was Fibonacci was first and foremost somebody who was educating the merchants about how to do their calculations. And that was a long tradition and Luca Pacioli was part of that tradition of saying, how would you organize your accounts? You're a merchant, how would you organize your accounts? And that was something that people were interested in for a long time, I think lots of, until computers. Basically it was really a non trivial thing to keep the accounts and do all these calculations. And in some parts of the world, particularly in Asia, abacus were a big thing. In other places it was really just people doing calculations. And for example, Ramanujan, famous Indian mathematician at the beginning of the 20th century, he was sort of employed at first as a clerk at the port of Madras, now Chennai, is that right? Yes, that's right. Who was responsible for basically doing the numerical, doing the calculations to keep the accounts at that time those things weren't as easy without computers, but there's a long tradition of mathematicians in the service of merchants. And there were a series of books that were sort of the famous book of the time that was like, I've got a better scheme for doing this arithmetic. And I think those books were used sort of the open source approach to selling services. And you know, those books were, were printed after printing started up and then you could go get these books and they cost a certain amount of money. You made something as an author from, from selling those books, but I think you made more by selling your services of let me come in and do a, do a kind of a, a seminar or a consulting job for showing you how to, how to do Your kind of calculations for your business. I mean, it is kind of funny that in a sense, nothing changes. You know, there's still the kind of. Lots of companies that are in the business of coming into your business and telling you how to set up all your transactions and your databases and so on. And it's obviously the volumes are higher today than they were back in the day. And back in the day you were teaching the merchants how to do their calculations. And there were lots of kind of mysterious methods. There was a chap called Richard Record who was the first person who wrote known arithmetic book in English. And he had the method of. Well, I'm sure in English it was called the method of Z's, the method of Zs, where he had some whole elaborate calculation thing where you followed numbers in some pattern that looked like a Z. And anyway, there were all these things and later on there was the whole. The method of threes, the method of sixes. These were all ideas for how to think about calculation. And it is sort of ironic that when it comes to modern math education, there are these fads which would cause it to be the case that if somebody explained to me how some elementary school kid was being taught arithmetic, it would probably have terms in it that I don't know. Even though I think I more or less know how to do arithmetic by this point. Took me a long time to learn. I taught computers how to do it long before I could fluently do it myself. I would say when it comes to learning multiplication tables and doing long multiplication or long division and so on, I'm not a big expert at doing that by hand. I certainly put more effort teaching computers to do those things. But any case that was in modern kind of math education, they keep on being these kind of ideas. We can explain it like this, we can explain it like that. This is nothing new. This has been going on for hundreds of years with different kind of methods that at different times to later folk seem quite obscure. I kind of almost think it would be fun to inventory all of those different methods and sort of import them into modern math classrooms and so on. Just, you know, you get to teach math history at the same time that you. That you teach the actual. The methods of math. And it's sort of, I think, relevant to know that these organizations of calculation are historical. They're things that happen to be that way. They are not necessarily that way as a feature of the nature of numbers and the nature of arithmetic. They just happen to be that way because that's a way that somebody Thought it was a good way for people to do it. And for example, it might not be the way computers do it. For instance, the standard long multiplication method. I remember when one of my kids was supposedly learning that and was like, why should I learn this? Computers can do it. And I didn't really disagree. But actually we had the opportunity at that time to try and figure out, well, maybe there's a reason to learn it, because it's kind of fun to know what the computer is doing inside. And it was like, well, what is the computer doing inside? How does Mathematica and Morphem language actually do long multiplication? And turns out it's not what you think. It's not the standard so called schoolbook method, which is which if you have N digits in each number, takes N squared steps. It's something much more efficient than that. But it's something deeply non human. It eventually uses Fourier transforms and things. And eventually if you have a very, very long number, it takes a time that's sort of N log log log N or something. So a time very, very close to just looking at every digit. And it certainly doesn't need N squared time where you're looking at every digit with every other digit type thing. So it's, you know, the, in that case, the sort of, the human historical approach is not the one that's actually used. Stephen Jones is asking what were the early tabulating machines like, such as ones that IBM sold? Well, There were tabulating machines in the late 1800s, maybe even earlier than that, and they made use of cards. Cards were invented, I believe by Jacquard, Marie Jacquard, at the beginning of the 1800s. And they were invented for looms, for weaving. And kind of the idea was the threads are being put in the piece of fabric in a particular way. What pattern will the threads be put in? And these punch cards where you had holes punched in them. There was a mechanical device that would basically, if there was a hole, then put a thread in this place. If there wasn't a hole, then don't put a thread in this place. That's what led in the 1840s when Ada Lovelace was writing about Babbage's design for the Analytical Engine. The Analytical Engine was intended to have its programs stored on punch cards of the kinds that were used with looms. And Ada Lovelace wrote about this and has this lovely statement where she says the Analytical Engine will weave algebraical patterns as a loom weaves patterns of birds and flowers. So it was kind of this idea that Algebraical patterns. That was her term for programs effectively. And that was kind of a notion of you using a card to represent this program. Now, cards were used more directly as a way to keep census data. Governments like to do censuses partly because they do things like organize voting that way and because they like to collect taxes based on those kinds of information. And one of the ideas there was you have these cards and you would punch out holes in these cards, depending on whether somebody, like one card per person, okay, who has this attribute, who is between 50 and 60 years old, let's say, okay, punch that hole out of the card, who is this age, punch another hole out of the card, and then you can figure out how many. If you have a big stack of cards, you can. Let's see, how does this work? Actually, you basically are pulling out. You're kind of seeing, well, which cards have this piece punched out of them, and then you're sort of lifting up the cards that have those pieces punched out. But now that I'm thinking about it, I'm not quite sure how that mechanically worked. I should know that. But sort of a famous. The early sort of instantiation of IBM was the tabulating company. A chap called Herman Hollerith was responsible for inventing at least an innovation. I'm not sure exactly at what stage of tabulating machines. And that was famously used for the US census in the early part of the 1900s, maybe the 1910 census, I'm not sure. And that was sort of an early contract that IBM had with the US Government. It still has lots of business with the US Government, but that was an early contract that it had. And as various countries got to be sort of more efficient and informationally oriented, they had lots of tabulating machines used for sort of saying things about their citizenry and so on. I mean, eventually got used for all kinds of terrible purposes in the 1930s. And possibly it's sort of the early version of endless microdata collection that's happening all the time these days, happening sometimes in some places for commercial purposes, in some places for government purposes. And it's kind of, again, it sort of all started with the tabulating machine. Well, you could say it all started with the original sort of documentation of people and things in ancient Babylon, kind of when the first cities were being created. That was a time when sort of there was a desire to get things organized, write things down, have definite records, have definite descriptions of who owned what and so on. And that was kind of how that went. Let's see. Who you asks, do you think future historians will have a harder time passing through all the information available in the last 50 years compared to the last century or even two? What is the best historical record for research? Books, images, video, et cetera? You know, I've done a bunch of history work and I have to say, giving a bit of a shout out to my friend Brewster Kahle, who runs the Internet Archive. That's really the thing that transforms historical research, particularly for the 1800s, because the sort of all the documents from that time have been scanned and ocr, you can search them all the newspapers, all these kinds of things. It's actually more readily available from that time than from later times, because when things are in copyright, you can't make them as widely available. But the things that are out of copyright and have been scanned, they are just really a gold mine for historical research. It's sort of remarkable to me that I can, you know, there's the name of a person and I can just do a search and I can find out that, oh, yeah, they were mentioned in this obscure newspaper sometime in 1822. And there are lots of connections that I've been able to make by just doing those kinds of searches. So I think that's an important thing. The kind of scanning of everything and the digitization of everything is kind of a critical thing now when it comes to things like movies and so on. Well, we've now got to the point where we have pretty good kind of transcription of the dialogue in movies, so that becomes searchable. We have, for better or worse, increasingly good kind of things like face recognition and searchability of that in things like movies. So once being able to get a lot of kind of searchable stuff in things like movies, I would say that the. There's really a question of more of. Well, there's stuff that's on the Web. There's some number of billions of pages. There's probably about, I don't know, between 5 and 10 billion pages of stuff, textual material on the web. There's maybe 100 billion. If you include all the archive material. There's even more if you include the things that are in the deeper web behind sort of corporate firewalls and things like this. That's a lot of data. And I think sort of threading that together to do history is absolutely doable. I mean, I've done a bunch of that myself. I've written both about other people's history and about my own history. In the case of my own history, I have pretty good archives, so I Have I really have the information? I mean, I have for 35 years now. I have really pretty much everything. Everything I've electronically touched, I have, down to keystrokes and so on. And so it's kind of fun. I can say, and I've done this a number of times recently when I'm looking back at discoveries I made, let's say in the 1980s or 1990s, I can say things like at 2:51am and 32 seconds, I discovered this thing because I actually have that timestamp, because I have that digital data to see what actually happened. Now then the question is, well, how do you tell what is the arc of history? How do you tell all those individual things and knit them together into sort of the overall arc of history? I suspect LLMs make that easier. Certainly LLMs can do an okay job at summarizing. Like we have an LLM running right now, which is going to make a summary of all the yakking I've been doing here. It won't be perfect. It might make some things up, but it's going to get better at that kind of summarization, I think. I think that's one of the things that we'll see incremental improvement done. And I think that that's maybe a very good way to kind of tease out the arc of history from a collection of disparate documents. Now, as a practical matter, you know, one's really talking about sort of embeddings and vector databases and all this kind of thing, because you can't just sort of put that a trillion words into the kind of say, okay, LLM, here's a trillion words. Digest this. You have to do that either as a matter of training or as something that is just really searching for an individual thing. But in any case, I think it will be. The automated historian is, I think, going to be more of a thing. I don't know how good it will be. I mean, I've certainly noticed in my own history and trying to piece together histories of things, history, usually there is a narrative, really. Usually there's a reason things happened. It isn't just the case that somebody just suddenly, one day, boom, they had this idea they did something. There was a whole trail of things that you can sort of figure out that led them to that moment, led them to that thing. People rarely. And in the world, usually there's a buildup before things happen. It's not just a suddenly this happened to happen. And I think part of the story of history ends up being finding that narrative finding, that sort of arc of history that goes beyond just the this happened, this happened, this happened. And I think that's something that is lots of work. But the very fact that there is all this data makes it much more realistic to knit together those arcs of history and get them right. I mean, I certainly know it. Not to say that isn't a lot of work. I don't know how much the LLMs will be able to help. But that is the thing you get now in terms of there are other issues about the historical record, what's preserved, how. Like for example, one example, medical records, for instance. There are ones from wars and like there are ones From World War II, there are ones from the Gulf War. Turns out the ones from the Gulf War are in much poorer shape than the ones from World War II. The ones from World War II are on paper and they're sitting in some archive somewhere and people can go and look at them. It's kind of a nuisance to fish all that stuff out, but it does exist. The ones from the Gulf War, apparently, you know, they were all on computers. And when the computers came back from the war, somebody said, well, let's recycle these computers. And they erased all the data. So that's a danger with digital data. That it is, in a sense. It's sort of easier to store physical stuff maybe if you don't lose it, if the place doesn't burn down and so on, maybe than to store digital stuff. I mean, my own experience of storing digital stuff is, well, don't, don't be too clever with the formats. Point one and point two, keep everything in active storage. You know, I just recently I had a floppy disk, an 8 inch floppy disk, very obscure kind of floppy disk. And somebody was kind enough to the Computer History Museum actually to be able to actually get most of the data, some of it had not survived from that 8 inch floppy disk. But that was a, a tiresome thing to do. I mean, I got that floppy disk in 1982 and by now it's really unreadable except at a history, a computer history museum. So that's a reason. But if I'd kept that data, and in fact I had kept that data as it turned out in online form, then that data as it sort of gets transferred from online storage to online storage and so on, that's something that you can expect to preserve into the future. Well, unless the disks get blown up or whatever else. And perhaps it is the case nowadays with sort of distributed cloud storage and this. And that and the other, that one can be fairly confident that that data will survive. The next question is, will that data be exposed or will that data be encrypted? I had the experience, my first big computer system, SMP, that I built in 1979-1981. In a piece of modernity, I encrypted the source code and of course everybody forgot the key. And it was decrypted about nearly 10 years ago now by cryptanalysis. But that was a case where, yes, you have everything in digital form. It could still not be accessible because it could be encrypted. And I suspect the encryption we have today will last a long time. I don't believe it's all going to be broken by quantum computers, partly because I don't really believe in quantum computers as such. And I think that the combinatorics are on one side in the case of encryption. And so there will be things where yes, it exists, we have the data, we have the bits, but it's encrypted. And there'll be also cases where there's a sort of more and more complicated set of questions about sort of digital privacy kinds of things and what can be sort of put together to form the net of history and what cannot. You know, the Census Bureau, for example, in the US has the rule that after 70 years a census becomes public and it's sort of carefully protected before then, but then 70 years later it's public. And I guess that's four score years. And 10 was thought to be, you know, kind of, it's all over by that time. It's kind of like things where people have 99 year leases for land. It's like everybody who made that deal is dead by the time that the thing blows up in one's face type thing. But this question of what will be exposed and when is an interesting question which certainly will affect what future historians can do. But I do think that the raw material is much more easily available. And the fact that there's such a high volume of it, as we can use more and more computational tools, we more and more become able to do this sort of computational history. I mean, I myself have done a fair amount of that. I think there's all sorts of interesting methodology around really using computers and computation. I mean, Wolfram language is a great tool for doing this kind of the best, I think, because you get to sort of do all this kind of multimodal work between images and numerical data and text and kind of use machine learning, pick out, recognize dates and documents, things like this, make timelines, make sort of networks of who was talking to who and so on. It's really, really cool. And there's much more that can be done in those directions. Let's see. 420 asks. Doesn't copyright law allow flexibility with people who want to share their works online? Well, yeah, you can absolutely say. You can say in, I'm just giving this to the world, I'm putting this in the public domain, or I'm using the Creative Commons, some standardized license to say it's available, anybody can use it. I do that with lots of stuff I produce. It's like anybody can use it, but you've got to say where it came from. For instance, the copyright was. The berne Convention of 1900 was the thing that sort of really sort of organized copyright. I mean, there was sort of, there were vestiges of copyright earlier than that, but that was the thing that was really the sort of international agreement of there being a copyright. It always used to be written in words, copy space, right? You know, I think it's a constant of history, that word smoosh. I'm always telling this to our DQA document quality assurance folk when they try and correct words like website. Back years ago, they were like, website is two words. And I said, it won't stay two words. It's going to end up as one word. And you know, this is a principle of smooshing in words. And it happened to copyright as well. But I think the US even today is not part of this Berne Convention. But there are kind of this notion that you get a certain unique right in your works is a notion that, well, let's see, patents, which were a different kind of thing. They originated in the 1600s, I believe, in England at first, and then were sort of enshrined even in the US Constitution. But that's really for inventions as opposed to sort of creative works. I think the sort of one thing that was sort of entangled with copyright, I suppose, was the right to print things. And, you know, certain places had a certain. Had licenses to print certain kinds of things. I mean, I think this has been true in the more totalitarian countries. This kind of idea of, you know, does the government give you a license to print this or that thing? Has been a big deal. It was a big deal in the Soviet Union, in modern times, in China, for example, it's kind of like, can you have this website? Do you have the government certification to have this website? I mean, different countries have handled this differently. In the US for example, if you wanted to own a television station up until recently and maybe even still today, there are all kinds of constraints on what you have to. All sorts of government constraints if you want to have a television station. Because kind of the idea is, well, gosh, that's something where you can really sort of have an effect on the country by having using your television station. It sort of raises all these questions about, well, you know, now that social media can have an effect, what should the government be doing about that? And, well, I'm not usually a big believer in all sorts of complicated regulations, but I think the issues have come up again and again, and they come up in different countries in different ways. But sort of the origins of copyright had to do, I think, in part, with those kinds of questions. And, you know, for example, in England, I know that the license to print the Bible was only given to, I think, Oxford University Press and Cambridge University Press for a long time. And that was sort of a, you know, that was not quite the copyright on the Bible. It's not clear, you know, I'm not sure that King James quite had that copyright for the King James version of the Bible and the original. Very unclear who would have considered that they might have had the copyright to that. I think one can. But in any case, that was sort of a constraint imposed by the government on who could print these things. And I think that sort of led to an idea that there was a sort of constraint on what could be done. I think that if you had. I mean, like, for instance, I think Shakespeare's plays were printed, all sorts of people printed those, and it was like, whoever, you know, can get to it can do it, and then you would sell them. And I don't think that there was a, you know, I don't think Shakespeare could come and say, no, you can't do that. I have a right of, you know, a right associated with copying. But modern copyright law, basically 1900 is the time when that came in, and it really hasn't evolved that much since then. I mean, I think there were lots of questions about the copyrightability of computer code. One of the questions is, if you translate from one human language to another, does that kind of define a new work where it's kind of a derivative work? And there's all this set of definitions that I think really kind of sort of, I think largely condensed into this structure, sequence, organization, definition of whether there is whether this thing has a structure, a sequence, an organization. Another thing that came up, we were involved in this in the 1980s. The copyrightability of computer languages, that's different from the copyrightability of computer code. For a long time it used to be the case that if you translated from one computer language to another, then there wasn't any copyright anymore. I'm not sure what the current legal status of that is. That was a weird thing that I think was kind of a little bit funky and perhaps not a great understanding of what it meant to have a program and so on. But there are also questions like, can you copyright a. Well, here's a very basic question. When can you copyright a number? Now, you might say, that's ridiculous. How can you copyright a number? You couldn't copyright the number 17 because it's just a thing that's out there all over the place. But remember that any document can be reduced to a representation as a number, and that number is really just equivalent to the document. So to say I can copyright a novel is to say I can copyright this great big long number. So now the question is, well, how short a number can you copyright? And for instance, a haiku, short piece of poetry is something that surely you can copyright. And I have to say, in some various license management things of ours, we've made use of the fact that you can't legally sort of copy a thing that's transmitted that happens to be a haiku as a sort of a form of copyrightable, sort of a minimal copyrightable element. But, you know, a haiku is much shorter number. Now, if you have a number that generates for you this amazing sort of cellular automaton pattern, some rule that generates this amazing pattern, and that number is 14 digits long and generates this amazing pattern, there's a pretty good argument you can copyright that number. So just how low can it get? And then that sort of brings in the question of you've got this number and that number generates this interesting picture, and you can copyright, you copyright, but the picture is generated from the number by a purely automated process by a computer. Now, you may have written the algorithm by which that thing is generated, but there's an interesting set of questions about where is the piece of originality, creativity and so on, and how much does that matter for copyright? There was a very famous Supreme Court case in the US the feist case, I believe, which had to do with the copyrightability of a phone book. And the question was, a phone book is clearly. And it was arranged alphabetically, so there was no creativity in the phone book. It was, however, an original document. Somebody had to make it, and it had never been made before. And in that particular case, it was ruled that the phone book was copyrightable, that it was sufficient. You did not have to have kind of a creative spark to have something which was validly copyrightable. And that led to all kinds of issues about the copyrightability of data and so on. We've been quite involved in those kinds of things. And there are other different rules in the US as compared to the European Union for those types of things. But these are sort of questions that have been the emerging sort of stories of copyright law. Now, another feature of copyright law was this idea of copyright libraries of a place where to have a copyright, you have to deposit a copy of the work in that library. You know, the Bodleian Library, I guess. I think there are two copyright libraries in England. I think it's both the British Library and the Bodleian Library. In the us, it's the Library of Congress. And, you know, in some sort of ways of setting this up, you had to actually send a copy of your thing there for it to be validly copyrighted. Now, that's not the case. You can just put that copyright symbol and assert that it's copyrighted. And, you know, it depends whether you actually validly were the originator of it, whether you can validly claim copyright. But the. The sort of. The thing is, if it's sitting in the Library of Congress and you put it there on a certain date, and somebody comes along and says, I did this, too. And that date when they did it is when they said they did it, is after the date when you deposited in the Library of Congress, then the Library of Congress is kind of your blockchain. It's kind of like, yes, it really existed at that time. It really was put in the library at that time. And that's how that works. But there's sort of very early in the history of computer software, it was the idea that programs were copyrightable. That was one thing. The idea that the object code, the compiled version, the executable, was copyrightable, that became a thing very early. I have to say, I was sort of interestingly involved in this one. I was working at Caltech in the early 1980s and building software. There was this. Well, this is a long story, which perhaps not to tell right now, but there was a whole sort of meltdown about intellectual property in which the bylaws of the university, as the person who was sort of responsible for this stuff, was keen to point out to me, the bylaws said copyrightable material is owned by its authors, which meant he didn't have to try and work out what to do with this intellectual property that we produced, which was software. And it's like, well, copyrightable material is owned by its authors. Copyright software is copyrightable. Therefore you own this. Don't bother us, just go start your own company. Which is what I did. That was my first company actually later that turned into a giant conflagration of various kinds that ended with me. Otherwise I might still be a professor at Caltech all these years later if it hadn't been for that particular. It hadn't been for me quitting in the end. But in any case, that was sort of a, you know, at that time. Again, that was early 1980s. The concept that software was copyrightable was a thing. It wasn't super well known, but it was known. And you know, in modern times there's sort of a big question about the copyrightability of for example, generative AI works. I think that's been a question that we've certainly addressed in the past. When we made our Wolfram Tones algorithmic music generation site, we had a very tried to do a real analysis of that question of who owns what when it's kind of a creator of the algorithm for doing something. A person who pushes the buttons. How many buttons do you have to push before you really were doing something creative? Is that the criterion? Is it that you were doing something merely original? If you search through a large number of possible interesting cellular automaton patterns. We made some NFTs from cellular automaton patterns and did kind of real time search through those things. And we're like picking them in real time. The act of picking is surely a thing that creates a kind of a copyrightability potential to the thing. Just like somebody who takes a photograph even though the tree was right there and they didn't do anything with the tree, they just picked the angle and decided to take the photograph of the tree. So in any case, That's that. Story. I guess there's a question here. Maybe I need to go soon here from Eggy. How did legal structures evolve with the creation of the Internet? Were completely new structures built because of it? A lot of different things came up with the Internet. I mean a lot of things from domain name registration, which was a. That was something like who can claim this domain? Who can claim the domain? You know, phish.com, how do you register that? How do you set up a registrar for those kinds of things? That one maybe was similar to the concept of registering the name of a company where there can only be one of it in a given state, for example. And that is different from trademarks which are established through use and so on. That was just a. Let me register this. But in terms of the Internet, I mean, there's just a boatload of issues about, oh, all kinds of things from, you know, do people, when you have the kind of the fiber optic wires of the Internet, you know, do they. Does everybody have to carry everybody's traffic? Can you have net neutrality? Do you force net neutrality where sort of everybody has to carry everybody's traffic equally? That was something that was certainly implied by the sort of early Internet protocols, But it's something that can certainly be debated. Those are a few of the things. But there's just an incredible number of issues that come up of different kinds of privacy, kinds of things and kinds of things about, well, about sort of material and its copyrightability. Look, one of the big ones right now is you put something out on the web and you intend it for people to read it. Then an AI comes through and reads it and trains itself from what you put on the web. For example, it sees the picture that you painted on the web and it trains itself from that picture and 5 billion others, and then it produces pictures. And some of those pictures have possibly something which is very reminiscent of the picture you painted. How does that sort of. How does your copyright to your picture flow through what the AI saw and so on? Now, we don't imagine that. We think that, for example, when a human reads something and that thing that they read just goes into their brain and it wobbles around and they sort of are informed by having read it. I mean, certainly I've spent my life producing things which I hope inform people when they read them. But when people sort of make use of what they learned from what they read, there's no thread that the law tells them they have to put back to what I wrote. I mean, it's always nice if they say, oh, I learned this from what that chap wrote type thing, but it's not as such. Well, with some constraints, which are a different story about sort of the kind of the goodwill, you know, the kind of, you know, who gets to say they discovered what and things like this. But in any case, on the other hand, if you quote verbatim this big chunk of stuff, then at some length of thing, there's the idea that, well, that's a copyrighted thing and you kind of have to get permission, whatever you have to get the right to do that. There's this kind of idea that if It's a short enough quote. It's kind of a fair use of the thing, and you don't have to do that. And that's, for example, on the web. An example of that is thumbnails on the web. It sort of became established that when you have an image, that a thumbnail is something that you can fairly use on the web and you don't have to get elaborate permissions to do that. That's how image search works. You know, you see all these thumbnails and when you click it to get the big picture, you'll typically go to the website because that's, that's where it was supposed to be. But this question of, of what kind of what, what does the AI, when you are a provider of pictures, for example, and you provide them for human consumption on the web, what does that mean that AI can do with them? And of course, that wasn't imagined at the time when these pictures were put on the web. So it's now a question of sort of retrofitting that set of ideas into a framework that was really not, didn't imagine the possibility that there would be this kind of aggregate use of something like images by something like an AI. And I think that's a thing that's sort of just getting hashed out now is, for example, one thing. You make an image with an AI, how much do you own the copyright to that image? And how much does it matter whether you touched up the image, how much effort you put into making the image, how much does that all matter? That's one thing. And the other thing is if you've put material out on the web, to what extent can you say, well, you could say, this is for humans only. You could put a thing just like the robots. Txt file can say, search engines shouldn't come in here. They don't always obey the robots. Txt file. That's a matter for, I mean, that's a. Technologically, they're not blocked, you know, and I'm not even sure completely what the legal status of the robots. Txt file is. I'm sure that you can have kind of a license agreement associated with the website as a whole and that probably governs that use and whether you can include the material from the website in a search engine. But you know, the argument was made, well, it's good for you to have it in a search engine, otherwise nobody can find your website. And you know, in some cases, for example with news publications, it's like, well, is it good to have that in a Search engine or not? And, you know, I think Google in particular has done some rather nefarious hacks around many of these kinds of things. Sort of avoiding, you know, sort of dealing with those kinds of issues. I think that in this whole question of okay, it can be ground up by search engine, that's sort of okay if your robots txt file doesn't say otherwise. Well, what about if it's ground up by an AI, by an LLM dealing with language, or by an image generation system dealing with images? At what point is what, at what point is it okay and not okay? And there's some precedents here with music, for example, where there's all of this issue of mixed, you know, what on earth are they called? My gosh, mixes in music where people will take just a few notes from one song. And then there are cases where people have argued just that sequence of five notes is the heart of the song and it's subject to copyright. Even though it's only five notes. I don't know what the actual shortest is. I'm guessing three or four notes. And that feels like it's kind of copywriting the number 17. But it really isn't quite, because the number of combinatorial possibilities is quite large. And I don't know to what extent one takes it as a significant thing, how memorable the da da da da or whatever it is that sequence actually ends up being. But that's a thing where there's sort of this question of to what extent does the mix thing end up working in? How does that work? In the case of music, what was done was there's this. Oh, gosh, there are two of them. One is called ascap and the other's name, I've forgotten, I'm afraid. But there's this whole mechanism where you can play music, but somehow when you play that music, you are giving some small amount of money to the central fund, which then redistributes that money to the original producers of the music. And you don't have to go ask before you play some music at your store or something. You don't have to go get specific permission from the artist who made that piece of music. It goes through this kind of clearinghouse. And by the way, the same thing, there's an attempt to do the same with copyright. There's a company, a copyright clearing house, right, which, gosh, I haven't seen them being used for a long time now. But that's a place where, for example, if you're going to do something where you're giving out things for some class at some university or something, and you're making lots of copies, which would be potentially a violation of copyright. You can just say, well, I'm doing this and I'm doing it through the copyright clearing house. And I'm giving them, you know, I'm just telling them I'm doing it. And then you get a sort of aggregated bill for all the things that you did that had various kinds of. You don't have to independently negotiate. I mean, when, you know, it is a kind of a. A slightly. Slightly frictional thing that people, you know, people are reproducing, you know, very often. I mean, on a pretty much daily basis, people are interested in reproducing some picture that I made that I put in some place, and they want to print it in some book. And we get some requests saying, can we print this in a book? And we say, sure, with these conditions and so on. But it's kind of a thing where the publishing company has to kind of contact the individual author to get that permission. Whereas there is sort of this idea like what happens with music. We don't have to do that. Where there's this kind of centralized mechanism. And clearly, for the Internet, if you've got an AI that looked at 5 billion images, the thought of kind of going back to all of those image creators and sort of how that works is a very. It'd be very difficult to do that. And even more so that it's kind of like you've made an image of a cat. And the question is, of all of the 3,000 cat artists or even cat photographs that you got from the web, which one was the most contributive to that image that the AI produced? Which cat should be fed the most stylish cat food as compensation for the fact that its image was used by the AI by this weight in the neural net, this collection of 100 million weights in the neural net that all in some microscopic way encoded the particular whiskers of that particular cat. It's very hard to see how you disentangle that. It's a very interesting problem, actually, this kind of credit assignment to input data. I think it's a very difficult one. I think there's sort of a general question of is there a calculus of derivative works that is. I have thought about this actually, to some extent because I think it relates to some sort of core computational issues. If you are doing something like you're in a video game and you're designing for some reason, an example, I'VE used in the past is a dinosaur headdress. And you have the basic dinosaur headdress. And then you modify the dinosaur headdress. And the question is, the person who originally designed that thing, they own certain rights, copyright type rights, and then as you modify it, you are gradually pulling it away from that original form. And there's a question of to what extent can the original creator prevent you from doing that and say, it has to be this particular dinosaur headdress? I think in the EU there are some notions in the law of moral rights and so on, where the original producer of something can exert more sort of, you can't. If your picture is going to be used in some setting which you find horrifying, you can say you can't do that even though the person bought the picture. But that's not, I think, a concept in US law, for instance. And I think there's this sort of question of it. Can you imagine a sort of calculus of derivative works that establishes kind of how much is this person responsible for this final thing? And that obviously depends on, well, if you change that sort of the green eye of the dinosaur to a yellow eye, is that a critical thing or not? And sort of an interesting question, and it relates to core questions about meaning and so on. Can you assign different levels of significance to different aspects of the message, so to speak? It kind of relates to things we were talking about much earlier, about information theory and the encoding of messages. Anyway, you guys have taken me far, far afield here, but I hope you found that interesting. It's always useful to me to try and sort of crystallize thoughts about different kinds of things. And I'm sure I forgot and misaid many things as I was talking about these kind of legalistic issues, but hopefully people found that of interest. So thanks for. Thanks for joining me and talk to you another time. Bye for now.