History of Science and Technology Q&A (August 7, 2024)

Hello, everyone. Welcome to another episode of Q and A about history of science and technology. I must say that I am not in my natural habitat today, so if weird things go wrong. Sorry about that. All right, let me see. There are a bunch of questions that came in from our website. Website here. Let me see. There's one here from Marcus. What is the history of data visualization? Was the discipline only able to flourish recently with the introduction of computers? Or is there a deep and rich history of people creating pictures by hand to extract visual insights from abstract data? Well, let's see what I can tell you about that. First question is when did people start drawing diagrams that we now have, and I'm pretty sure were architectural plans that were drawn on stone tablets in the Babylonian era that still exist? Certainly we know that Euclid was drawing geometrical figures. And there are some fragmentary manuscripts, I think, that are more or less contemporary that have sort of drawn diagrams there. That's not data visualization. That's abstract things of circles and triangles and so on made abstractly for purposes of Euclid's elements and so on. But the thing that, but that was sort of diagrams got built then. Then the question is, for example, diagrams that show things happening as a function of time. I'm pretty sure the original person who got interested in doing that was a person called maybe Nicholas Aresme, maybe in around the 1200s or so started to be kind of the notion of drawing a diagram. Now, I have to say that must have been somewhat related to the invention of musical notation because that's, I think like 1064 or so, a person called Arezzo was at least said to have been responsible for kind of the idea of a stave and drawing musical notes on the stave. Now, by the time you have that, you have kind of a time sequence defined because you're drawing the notes in their time sequence. So I suppose the idea of showing where something is as a function of time was perhaps not so surprising. Now when one got to things like, well, by the time I'm trying to remember what diagrams, I mean, people were making diagrams of abstract things. Leibniz, Newton, people like this made diagrams to abstractly illustrate things in the same kind of style as Euclid, I think the question of when there was sort of started to be data being visualized, well, first there had to be kind of a, a notion of experimental data. I think the first kind of traditionally first identified instance of statistics, I.e. numerical data on the state, is a guy called John Graunt who had this rather grim book called Bills of Mortality, which was about during the plague, I'm guessing, the 1665 plague, maybe in London. It's kind of a weekly summary of how different people died and so on. It's quite grim reading actually. But that I think has some rudimentary kind of graphical presentation in it later on, the place where there really started to be kind of. Well, let's see, when do other things come in? There were two things that I know in the 1800s. One was Florence Nightingale particularly known for her kind of nursing kind of innovations around the time of the Crimean War and must be 1820s, 1830s, that kind of timeframe. But she presented data from those things and I think she invented, was it the bar chart or the pie chart, one or other of those two was invented by Florence Nightingale for the presentation of kind of medical data. Now, somewhat around the same time, maybe a little later, there's a chap called William Playfair who produced this book called, I think it was called the Statistical Breviary. And that's a book of diagrams about the economic and other conditions of different countries around the world at that time. And it's a very lovely book actually. It's. Those diagrams are definitely kind of modern, sort of high quality modern diagrams. And I think that the. And there are pictures of kind of the economic, you know, the gdp, I don't think they call it GDP quite yet. Various kinds of imports, exports, the amount of, you know, different kinds of things about currencies and so on, populations, all those kinds of things, and some somewhat more elaborate plots that were not just sort of a line as a function of time type thing. But so that was at that point by the time of William Playfair, there was pretty well developed sort of visual presentation of data by the mid-1800s or so. Then it's the next thing that comes about. Well, there's a. There's kind of an ugly period in data visualization history. The Nazi party in Germany was pretty big on making kind of sort of elegantly designed data visualizations proving who knows what. But that was some of the more striking, at least data visualizations which probably have been sort of don't get collected when people start talking about those things. But then in terms of sort of the arrival. Well, let's see, there were trying to think whether there was sort of data visualization that happened in the sort of electromechanical period of computers. And I don't think so so much, but you've got to remember that. Yeah, I don't Think that was really a thing. I think the time when visualization started to come in, there were two basic technologies and we're now talking 1950s to 1960s two technologies. One was pen plotters and the other was oscilloscopes. So a pen plotter, I mean like I used one of these in probably what 1975, 6 type time frame connected to an electronic desk calculator would be, which could run some rudimentary programs would be an XY pen plotter where essentially there were two motors and they would kind of move this pen in the X direction and Y direction and then you put a piece of paper there and the pen would draw this or that thing on a page. And certainly some early papers of mine, the diagrams, the original versions of the diagrams were drawn with a pen plotter. That was one technology. The other technology was oscilloscopes. And an oscilloscope is. Well it's kind of a cathode ray tube which means some people probably here know exactly what that is. But for those who don't, it was how televisions worked up until 25 years ago or so now when they kind of flip switch to the flat screen kind of semiconductor technology methodology. But before that time it was a cathode ray tube which meant it's a glass evacuated tube that has electrons produced from a hot wire at the back and then they're pulled by electric fields to the front they go and they're deflected by magnetic fields. No electric fields. I guess the they're deflected and what, what happened. So early television worked by scanning the screen every. Oh, I don't know what it was maybe 15, but somewhere between 15 and 30 times a second scanning the screen. And like in, in England there were 405 lines on the screen and then they went to, in probably late 1960s they went to 625 lines. I think in the US it was 550 lines. Maybe these were different standards for television transmission with names like PAL phase alternating line and gosh, what's the American one? Gosh, I've forgotten. There was another European one called CCAM S E C A M But these were different schemes for sort of painting, having. So what would happen is the electron beam would from the back of the television it's accelerated and to a few kilo electron volts of energy. And then the electrons would be sort of painted on the screen in the right place controlled by these electric fields, magnetic fields. And on the screen there was a phosphor which glowed wherever the electrons hit it. And that's what you would see and that's how you would see the television screen. Well, that same technology was used for electronic monitoring. That's how oscilloscopes worked. And so the idea of an oscilloscope was you would have this time based signal that was sort of constantly moving the dot from left to right across the screen. And then you would have another signal, some voltage coming from something you're trying to measure that would kind of have that dot sort of move up and down the screen in the Y direction. And so you would then get this sort of trace that was continually being painted across the screen. So you could see some electronic signal, something at a particular frequency or whatever else. Well, that technology kind of advanced, particularly a company called Tektronics was big. And doing this was in having these storage oscilloscopes where you would get. Sometimes early on it was just the phosphor didn't decay very quickly, so you'd still see the image for a while. Later on it worked in a more electronic way. But in any case, you would be able to get sort of paint more and more complicated things on the screen by just controlling a bit like in the pen plotter case where you're controlling with XY coordinates the actual position of the pen. In these oscilloscopes you were controlling the position of the electron dot. And then sort of the next level of technology was you could get a thermal printer that would print from the screen of the oscilloscope, so to speak. So, okay, so for example, I made a bunch of my early kind of data visualizations using, well, first pen plotters, then Tektronics scopes. And then sometimes it was just a question of printing things out on a printer and just representing things with stars and x's and so on. Kind of amusingly, the first, well, the system called SMP that I built starting in 1979, it's kind of a forerunner of Mathematica and Wolfram language. Its graphics were either on a printer, where they were done with plus signs and stars and x's and so on, or on one of these Tektronics kind of scope type devices. In fact, it's still the case in Wolfram language today that you can set it so that graphics will come out printed with characters. And occasionally when one's debugging things, or using some newly arrived kind of embedded processor, something like that, you can get these graphics coming out as stars and pluses and so on from modern wolfram language. In 2024 and so on. But any case, back to kind of the data visualization story. I think the thing that. Well, the thing in terms of data, there were these things called Calcomp plotters. They were pen plotters. That was kind of the ambient of the late 1970s, early 1980s. And around in the 1960s already there had started to be statistics programs. And so there were two big ones. One was called spss, Statistical Program for the Social Sciences. Another still exists called sas. Sas. And there were a few other more minor ones, but those were, I think, both developed on mainframe computers probably. Well, SPSS probably in the 60s to 70s and then SAS maybe 70s and beyond. And I think SPSS was bought years ago by IBM and absorbed into other things. Sas, I think, is not so much emphasizing making statistics software anymore, but that was a time when people. When sort of the statistics that people had developed. Okay, so a little bit more on statistics, because that was sort of where a bunch of this kind of data visualization came from. So I mentioned, you know, John Graunt and his bills of mortality from the 1600s. There was, you know, later developments of statistics. I think one notable development was from Gaussian. You know, notably there's the Gaussian distribution, the normal distribution. Gauss was interested in errors of measurement, and Gauss sort of did practical astronomy, discovered the asteroid Ceres. And one of the things of interest was kind of when you make all these measurements, what's the distribution of errors? And that led to the Gaussian distribution idea. But the thing that really made statistics take off was agricultural statistics. People in the 1920s, 1930s, trying to figure out kind of, if you are growing crops and you have this type of crop and that type of crop, and you put this pesticide on this crop and that crop, what will happen to it? And so they would divide off these fields into little plots and then they would say, what's the chance that this plant does well or doesn't do well with this type of pesticide on it or fertilizer on it or whatever else? I think maybe it's more fertilizers than pesticides at that time. Actually, amusingly enough, there was a place near where I lived when I was a kid in England. There's a little place, I think it's called Rothhamstead Experimental Station. And it was an agricultural research operation. And that then people wanted to figure out, okay, did this fertilizer do better than this with this type of corn plant or whatever else? And that was a big adventure in Trying to tease out effects from numerical data. And for that a bunch of methods got invented. Particularly a chap called Fisher invented a bunch of methods. This is when, well, the Gaussian distribution was widely used and things like the ANOVA method, analysis of variance. What is it called? The Fisher. Oh, gosh, what is the distribution called? It's the distribution of ratios F variance distribution, I think is what it is. And then the so called student T distribution invented by a chap called Gosset, that tells you things about the distribution of differences between means of Gaussian distributions. So on, all that stuff got invented in the 1930s. To my mind, it's sort of interesting. It got used at first in kind of agriculture, I think only quite a lot, later in medicine, where it's still very widely used, and in the social sciences also started to be used. I think this was a time when things like polling methodologies were getting more sophisticated and so on. And it became important to know what was the margin of error for this or that thing. All those kinds of questions. But that methodology of these kinds of statistics, the chi square distribution, the student T distribution, all these kinds of things, was interesting to me that in, for example, physics that was sort of unheard of even in the 1970s and the beginning of the 1980s when I was involved in doing particle physics research, experimental physicists would sometimes talk about the chi square distribution, sometimes talk about confidence intervals and so on. But that wasn't part of kind of the standard canon of kind of mathematical methodology that people thought about in physics. That was something that had really come from agriculture and gone through the social sciences and so on. A lot of those early statistics programs were developed for kind of social science research and things that went around that. And the visualizations were not terribly sophisticated. I mean, it was mostly just plotting things and maybe a bar chart or two, those kinds of things. One didn't really see much more sophisticated stuff in the 1970s, for example, I think then, well, let's see what happened next. I mean, in my own efforts, starting in 1986, when I started developing Mathematica Wolfram language, kind of one of the things that seemed like an obvious thing to do was to really kind of uptick the types of visualizations that one could get. There were tricky issues. One of the tricky issues was kind of what one might call automated aesthetics. That is, you're making this plot and it's got these various different parameters. It's got numbers on the axis that are six digits long. It's got things about where exactly do you plot? And so on. Even the very basic thing like what scale are you going to pick for the Y axis? You've got all these data points and there's some outlier data points that are really big and other ones that most of them are concentrated very close to zero or something like this. What automated choice should you make for the Y scale? And that was something we developed for version one and using kind of statistics type methods. But that was, at that time, that wasn't really a thing people thought about doing. It was like if you were. There were a variety of plotting programs and so on, but they tended to be, you tell them everything, you tell them what the scale should be, you tell them all those things and will go, you know, actually put, put the, you know, put the data on the, on the, on the screen or whatever. So this whole sort of discipline of automated aesthetics, I guess we kind of pioneered a bunch of that stuff. I think of, of trying to figure out things about what, you know, how you would make essentially a, an, an algorithm that would make aesthetic choices automatically. You know, there were folks like Edward Tufte who collected a lot of kind of classic visualization. And then there started to be like business visualization, spreadsheet type things. Those tended to use fairly simple kinds of plots. And I think they, they started doing some level of automated aesthetics. But that's roughly the history up until close to modern times. I mean, I think the challenge continues to be sort of when you're throwing a bunch of data, to what extent, how automatically can you render that in a way that is understandable to people? And there were different thresholds of what became possible to do. So. For example, even in SMP, back in 1980, I had some 3D surface rendering. And the big issue was hidden surface elimination. You've got a surface and it's like, what do you not see behind, so to speak. And that's fairly easy to work out. If you have a Z buffer, if you just have a divide space center into discrete bins and you just say on that line, going back from your eye, what piece of the surface do you hit first? But for example, one of the things we kind of pioneered in Mathematica in 1988 was Resolution Independent hidden surface elimination. That was a whole adventure using postscript, which was a resolution independent, still is a resolution independent way of specifying graphics. And it was a more complicated kind of computational geometry story to figure out. How do you make, how do you sort of precisely determine what little fragment of a surface will be visible or not rendering that fragment at sort of full resolution and so on. But so, you know, 3D visualization was a thing that sort of circa 1988 was becoming kind of comparatively easy to do. Although things like this resolution Independent story was still something that was on the frontier at that time. Things that weren't so easy to do were things like visualizing graphs and networks. You've got a whole bunch of points and they're connected together somehow. How are they? You know what, you know, what, what, how should you connect them? That was something where, when I was writing papers in the, in the early 1980s, I had to kind of, you know, lay things out with pieces of string and so on and eventually draw basically by hand these complicated networks. I should say that I know things that I wrote back in the end of the 1970s. I mean, I routinely used to have somewhat elaborate graphics that were sort of a mixture of computer and hand generated in papers I wrote. And they were definitely not a common thing at the time. I mean, I remember I was studying kind of production of particles in QCD in high energy collisions and so on. And I had all these pictures of quarks and gluons and such like that were kind of a visualization of what was going on. And that was sort of an unusual thing at the time. And. But back to, back to laying out graphs that really didn't become possible until. To do in the sort of routine way until after 2000 when I was doing them for my book New Kind of Science. Kind of an interesting story of a particular individual who really was outlandishly good at sort of untangling kind of nodes and graphs and just being able to immediately see this connects that, that connects to that. Later on she became a, an innovative knitwear designer, which I don't really know what was cause and what was effect there, but it's kind of interesting to see that there are people who have such very different levels of ability at something like, you know, can you untangle this graph? But eventually, probably beginning of the 2000s, we started to have good algorithms for doing that. And that's been something that's been very critical in a bunch of work that I've done in the last few years, which has made a lot of use of kind of complicated graphs and networks and so on. So anyway, a few thoughts there. Let's see. Oh, a question from George commenting on the fact that Nikola Tesla was building a machine for wireless transmission of electricity. It seems we're getting to a place where we can beam solar energy down to Earth from solar harvesting satellites. Curious what my take on the timeline for that is. I haven't followed that in great detail. I mean, I'm kind of like, you know, watch out if you're in the way of the microwave beam that's beaming lots of energy down from orbit. I don't really know what the effect of that is on the atmosphere and heating the atmosphere and all those kinds of things because kind of like, I'm not sure how realistic that is. And it's also the case that some of the things one wants to use power for, like computers, for example, if you really have lots of power available in orbit or in high altitude drones or something like this, you might as well just do the computing there. Making a kind of a funny kind of pun of technology development of the cloud, being in the cloud, so to speak, or above the clouds. So I don't have so much to say about that. Let's see. Diego asks, from my perspective, what is the importance of compression in computer science? Well, mostly here I'm going to talk about history, so we'll talk a bit about the history of compression. So compression is you take a message, it's written in English, you're going to transmit it in some digital way. How do you have the smallest number of bits that you have to transmit so that you can reconstruct either the exact message or at least a good approximation to that message? So the first place. Well, there were things done for cryptography, probably even in antiquity where people had code books where it would say, well, just like in modern times, there are things like 10, four as a kind of police scanner. I don't know whether they still use that, but I think that meant sort of okay. And you know, probably 1175 means, you know, there's a cat stuck in a tree or something like this. These kinds of code book ways of taking sort of common phrases and turning them into sort of short codes. That's kind of an early version of compression. And I think that was also done for kind of military coding, cryptography purposes of, you know, just this incomprehensible collection of numbers that were known to both sides. What they meant, I think the first place where I know that anything sort of systematic got done on that was probably Samuel Morse and Morse code, which I think was about 1830. And in Morse code, sort of one's using a telegraph machine, which can just send, just sending pulses of electricity down an electrical line across the country or wherever else it was laid and eventually across the Atlantic and so on. And the question Was how would you encode the, you know, I don't know, the enemy will arrive in two hours type thing or whatever it is. How do you take that piece of English, for example, and. And transmit it in this essentially digital form of bip, bip, bip, et cetera. And the sort of key thing in Morse code is there's short and long things. So there's either like. And then you keep going, or you can go, make a longer pulse. And so you have dots and dashes. And so then the question was, okay, we're going to take dots and dashes and we're going to encode the Alphabet with dots and dashes. How should you do that? So I think it was Samuel Morse was the one who figured out, well, E is the most common letter in English, T the next most common. So let's make E a single dot, T a single dash. And then as the letters get more obscure, you give longer encodings for those letters. So famously, SOS is. I think I should know that. But. So those are, you know, O is a much less common letter, so it gets three dashes, whereas E is the most common letter and it gets just a single dot. So it was clearly that was a way of compressing things by having the total number of dots and dashes was made smaller by the fact that common letters were encoded in shorter sequences than less common letters. So I'm not sure what happened for a long time, actually, with data compression, I think Morse code was. Was used for a long, long, long time. When it came to the first radio and things like this first television, there was no compression. That was because that was really an analog signal. I mean, in the case of radio, you were really transmitting the amplitude of the sound, vibrating at some number of thousands of times per second or whatever. That was what was being transmitted. That was the. The electromagnetic wave was of that intensity. And so there wasn't really a notion of compression for that. So then I think the next big thing in the history of compression is probably 1948. Actually, it was a bit earlier than 1948 and Claude Shannon's work on information theory. It's always been a big question in my mind the extent to which there was sort of direct connection between Alan Turing's work and Claude Shannon's work. I mean, to describe that during the war, Alan Turing, who had sort of done his work on Turing machines as models of computation, 1936, when the Second World War started, he got recruited to this effort at Bletchley park in England, halfway between Oxford and Cambridge were trying to recruit from both places, so they picked somewhere right in the middle between there. It was a very secret operation to break the German codes, primarily the Enigma codes. And Alan Turing's one of his big contributions there was essentially a statistical method because using the fact that you didn't. I think, essentially using the fact that in a language like English or German, there's lots of redundancy in the language. Like if you see a Q, you know, it's followed by U. So when you're trying to kind of figure out how did something get encoded, it's very important to know sort of what the source material was and what the regularities in the source material were. Because if you kind of knew. If you knew that the source material repeated, you know, like, for example, if you have a dispatch, and at the beginning of the dispatch it says kind of, you know, to the high command of whatever it is, and every dispatch says to the high command of whatever it is, then that's a thing that you can use to do cryptanalysis to figure out, oh, since we know what the source text for this is, we can figure out something about how the encryption works. And at that time, people had Enigma machines. They'd been made by a company in Sweden, I guess, and there were some Polish folk who had somehow gotten hold of Enigma machines. So the actual machine with its rotors and so on, existed. It was a question of how was the key set up on this machine to actually do the encoding. So Alan Turing invented this method to sort of reduce the amount of the number of possibilities that had to be tried for what keys might work. And that was really a story of sort of the statistics of text. That was something that was done as part of the Bletchley park effort for cryptanalysis, which was not declassified until. Well, until some parts of it 1970s, some parts of it 1980s. So during that period, it was all secret information. And meanwhile, in the us, Claude Shannon, who had worked on kind of connecting logic to electrical switching networks, that was his thesis at mit. He was involved working for the telephone company Bell Labs. What became Bell Labs, I guess it was. Was it already AT and T? I'm not sure. But the Bell system, anyway, the Bell telephone system, he was involved in making a speech encipherment system. And I think in the end that was a very elaborate, you know, Churchill to Roosevelt speech encryption system. I think it was called Delilah. And I think they installed the two ends of this thing with giant machines and so on. In both places, I think it was used Very, very rarely, maybe just once or twice. But in any case, that was something that Claude Shannon worked on at Bell Labs and it was then located in New York City. And at some point during the war, Alan Turing went to visit Claude Shannon. And I don't think one knows what was discussed in that visit, and I don't know to what extent the work that was done at Bletchley park was something that Alan Turing was kind of allowed to talk about with the Americans, so to speak. So it's unclear to me, at least what the transmission of information was there. But in any case, in 1948, after the end of the war, Shannon published this paper about what became information theory. And that paper is really about sort of what is the true amount of information as he thought about it in a message. And that was originally developed for thinking about cryptography and thinking about what you have to do for cryptanalysis. I don't think that's mentioned in Shannon's paper. And probably those connections were not publicly known for a long time afterwards. But what Shannon kind of did was to say, well, like Morse code, if there's a way of, you can ask the question, if you could do perfect Morse code, how short could a message be? In other words, if you could perfectly encode sort of things with blocks of symbols, how short could you make things be? And then he worked out this P log p formula for amount of information. Now it's a little bit of a. People get very confused about is it really the amount of information? Because I could have a whole, if I was going to, you know, I, I could just tell you PI, for example, and you would be able to say 3.1415925, whatever it is, I don't know beyond that. But you know, in principle an infinite number of digits. But I just told you PI. So that in a sense has an infinite amount of information in it. But on the other hand, it has a very finite amount of information because you just have to say it's the digits of PI. So what Shannon did was a very specific version of how you sort of compress information. That had to do with taking blocks of data and saying like in Morse code, you know, this block occurs more frequently. We'll encode it with a shorter block and so on. And that led to sort of block coding. I think the, I'm trying to remember what the dates of these things were. But the sort of famous example of so called Hamming codes by Richard Hamming, I'm thinking that must have been late 1950s that those were invented sort of as a sort of practical version of Shannon's kind of idea that you can compress data by having the right code words, sort of better, more sophisticated than Morse code for arbitrary kinds of data. And then there was a whole sort of bonanza of kinds of coding systems that were invented late 1950s, early 1960s, particularly this algebraic theory of coding got developed where people were able to relate kind of the production of code words and codes to properties of polynomials. I mean, one mathematical connection that's pretty easy to see is, well, you also want to have sort of all, all connected with error correcting codes as well, where you're transmitting the data and maybe a few bits didn't arrive correctly, but you still have enough that you can uniquely reconstruct kind of what the character that was being sent actually was. So, I mean, in a typical example of this, sort of sees the connection to other kinds of mathematics is if you imagine that you have these code words that represent sort of coordinates in some, the successive digits represent sort of values of a coordinate. You can think of sort of a whole bunch of spheres in high dimensional space. And I say spheres because as you go away from a given point, you're kind of changing some of the coordinates as you go in the different directions on the sphere. But if you say the sphere, you're saying that there can't be in any given direction. There are a limited number of coordinates that got changed. And so then there's a question of sort of packing spheres. And you say, how can you best pack these spheres in high dimensional space? And that kind of gives you an optimal way to make one of these error correcting codes and so on. But any case, the compression idea, there's this block coding idea. You take a block of data and you find a shorter block. You use the fact that some blocks are more frequent than others, and you encode the frequent blocks with shorter blocks and the less frequent blocks with longer blocks. That's how hamming codes work. That's how a whole bunch of kind of early compression methods work. Now there are places where compression got used, for example, in fax machines, if anybody's even ever heard of a fax machine anymore. Fax machines have been invented. Actually the time when telegraph machines were invented, even Gauss had a very early fax machine that worked through telegraph. You know, you're just sending, you know, is there a dot? Is there not a dot? And you're reconstructing that, that, that thing as an image. But fax machines use run length encoding Because a very common thing on a, you know, you're, you're making a page and most of it is white because there's no, there's no ink there and you just want ink in a few places. Run length encoding just says, you take the sequence of bits and whenever there's just 00, 0 a zillion times, you specify that by just saying 517 zeros rather than just giving the 0000517 times. So fax machines used that kind of encoding wherein fax machines code became common in the early 1980s when electronics became cheap enough that that made sense. That's how fax encoding worked. Then another sort of big idea in compression is dictionary compression. And that sort of came into practical use the Lempel Ziv algorithm and then the zip encoding. And the idea there is that you use, you're kind of numbering code words. So if you have a document that has lots of mentions of the word rhinoceros, eventually there will be an entry, a sort of a dictionary block that tells you, okay, every time we say it's thing 17 that means rhinoceros. And usually it's kind of a recursive thing. You first of all encode things with one sort of base dictionary and then you encode more things that come in as adaptively sort of recreating dictionaries as you go ahead. And that's a pretty efficient method of encoding. It's the one that's most commonly used generally for text encoding. And for English for example, it usually gets you about 50% compression for typical documents. That same approach, if you're using it on computer code, for example, can get you much larger compression because at least with low level languages there are fairly small number of keywords and so on that you have to sort of specify in the dictionary. Well then the other big game in town is image compression. Well, there's also sound compression. Okay, we'll talk about all of these. With image compression, the thing that really made the web, I suppose was the GIF compression scheme for transmitting images. GIF was originally invented by CompuServe, I think, which was an early kind of dial up computer bulletin board, et cetera, service. And the idea of a GIF is it's a way of taking the things you typically see in images. So run length encoding is the most basic example. But there are lots of other regularities in images and being able to just sort of encode the things that are common in images and encode those in a short way. There's sort of A whole family of these things. There's gif, there's tiff. TIFF was used a lot for encoding things for printers. Then there's jpeg. JPEG kind of is a method of compression that is lossy in the sense that, for example, it is encoding images so that when in images there's sort of a picture of an object, and an object has more or less uniform color except at its edges. And so it's a pretty good bet to say, for example, that well, lots of kinds of colors in a picture will not vary too much. So if you just say, well, what's the frequency that they vary? You just take low frequency variation using Fourier transforms and so on, and you look only at the low frequency variations and you ignore the high frequency ones. So some kinds of things, like a bunch of little cellular automaton images that I've made, compress very poorly with JPEGs because they have a lot of fine detail that is not what is in a typical photograph type image. So that's another kind of compression technique. PNGs are a slightly more modern version of that kind of thing. Then there are a bunch of other technologies with wavelets and so on. It's sort of the question is what are the things from which you can make an image, for example, most efficiently? When it comes to audio compression, some of what was done there was quite wild because some of it is really modeling things like the human vocal tract and saying, well, just how many parameters do you need to specify those sounds that humans make? And it's not so many. And so then you take the sound that was made and you try and sort of represent it by saying, well, what is the effective sequence of parameter values for the human vocal tract that will be necessary to make a sound like that? And you transmit only those parameters. And actually that's how MP3, for example, was originally developed. Although then it got used for music, which wasn't its original use case, I believe. And. But it worked well for that. And now there's MP4 and so on, which also there's a whole separate issue of compressing video where you have sort of differences between frames and not differences, but not so much difference between successive frames. So that lets you have other forms of compression and so on. But it's been a, you know, if you look at the sort of methods of compression, there was a period in the, I would say late 70s, early 80s, when a bunch of the kind of standard compression techniques like dictionary compression, like GIFs and the early JPEGs and so on, got developed and then the audio compression was a bit later than that. That's probably a decade later than that. Good audio compression. And that's kind of how it's developed. And occasionally people will say I've got this amazing new idea for doing compression. There's certainly kind of neural nets give you new kind of a potential ideas about how to do compression neural nets. One of the things they do a lot is to make so called autoencoders that are sort of general compression techniques. But in terms of sort of the vast mass of compression that happens and files and transmitting things and streaming services and so on, that's still these methods that were invented some time ago. Let's see. The question here from Johan. Do we know who designed written language? Are there still missing pieces in history to properly map out the history of written work? We certainly do not know the name of the person who invented written language. I mean written language was invented several different times. I mean it was invented Babylonians, then Phoenicians and so on. That was sort of one invention. The in. In. In America's different invention. In China a different invention. I think maybe India was still. No, India was a pretty late invention of written. Written of writing I think so, you know, writing was invented multiple times with multiple different systems. Whether it's the ideographic system, the alphabetic system, hieroglyphic system, whatever else. And one doesn't know sort of what the. Who was the first person who had that idea. Now, now there are alphabets where one knows something about their history. I mean, for example, you know, the Russian Alphabet, Cyrillic, you know, invented by Monk Cyril. I don't know the detailed story there. I know the Armenian Alphabet, Mashpot, somebody or other, I forgot the name, but that was a definite invention. I think 1400s maybe, but that was invented. And so, you know, the idea that there are specific alphabets that were invented by particular people, that is a, a thing. Obviously there are many languages. I mean There are like 7,000 languages in the world, probably decreasing progressively. Maybe there were 20,000 at one time. Some have gone extinct. But a lot of languages didn't have a written form. And so when people came in, you know, anthropologists, linguists, and so on came in in the particularly late 1800s. You know, a lot of the sort of colonial era work done, particularly British, French, maybe some folks who came and sort of provided alphabets, provided written versions of all sorts of languages that hadn't had a written form before. It's kind of confusing because it seems like this is a very Obscure language, very different in structure from languages we normally deal with, but it's written in English characters. And that's because that was introduced sometime in the 1800s. Let's see, there's a question here about. From Johan. No, from Alex. Sorry. What is the stage of development history around implementing cellular automata in hardware such as quantum dot cellular automata? What large scale hardware, accelerated simulations would be interesting next. All right, well, I can tell you a little bit about this history. So cellular automata things I've studied a lot, very simple programs where you just have the version that I studied, a line of cells, each is either black or white, and in a series of steps. The color of a cell on the line below, going down the page, for example, is determined by a rule in terms of the colors of the neighboring cells on the line. Before actually, two dimensional cellular automata was studied a lot before I started studying one dimensional cellular automata. They started being studied in the 1950s while there were many different reinventions of the same thing. But essentially you have a two dimensional grid and at every step you're saying I'm going to change the color of the cell based on the neighbors around that cell. And so how were those things used? Well, they were used for image processing in the 1950s. People made arrays of photocells that were able to detect light and then they would sort of put switching circuit connections that would figure out, would process these images, for example, removing edges, you know, identifying edges, things like that. That was all closely related to early attempts to do what we now think of as AI. Even then it was thought of a bit as AI. The term AI was invented in 1956 where and particularly things like perceptrons, which were kind of early neural nets that were based on just taking values of light levels in an array of cells and then combining them in various ways. Those perceptrons tended to have every cell connected to every other. Cellular automata are like what we would now call convolutional neural nets, where things are only locally connected to each other. But so there was a bunch of work that was done on hardware, two dimensional cellular automata. Even from the 1950s, the one dimensional ones, there was a branch that was kind of hidden from view because it was related to military cryptography and not quite one dimensional cellular automata, but very, very close to that was a bunch of things which still I don't know in detail what was done. It said that a bunch of Soviet cryptography made use of nonlinear feedback shift registers, which are kind of the same basic story as one dimensional cellular automata that would have been in the, well, probably 1950s. In terms of other hardware implementations, there are. Well, when people started getting interested in parallel computers, in particular the 1960s, the idea of having just this sort of array of cells and making them all do the same thing was a common kind of notion in these kind of parallel computers. People also wanted the individual nodes to do sort of sophisticated computation, but for sure they could do these simple kind of just look at your neighbors and change your state. Those early parallel computers weren't much used for cellular automata. They could have been, but they weren't. Even when the connection machine computer that I was involved with in the early 80s was being developed, that was a machine that was a 16 dimensional hypercube. And it was really very good for doing cellular automata. But kind of the hope was that it would sort of solve AI and be useful for lots of other purposes. That hope didn't really pan out, but it really was pretty good at doing cellular automata. But it was rarely used for that. Now at that time, in the early 1980s, I had kind of, I would say, raised the awareness of people about cellular automata and why they might be useful, for example, for making models of things in the natural world and so on. And so there were a whole bunch of people who were interested in making custom microprocessors based on cellular automata. And a bunch of designs got done. Very few actually got made. There was a group at MIT sort of this, they got started independent of me that were making kind of custom electronics versions of little cellular automaton machines. It was actually kind of, for me, kind of interesting, if frustrating that they had these things that would have sort of real time display of two dimensional cellular automata with all kinds of cool things happening on the screen. And it was frustrating because it's like all kinds of cool things are happening on the screen, but nobody's doing any science with this. They're just looking at the cool things that are happening screen. And so I spent a little bit of time sort of trying to extract science from what one could see kind of just running on the screen. Although I have to say I mostly ended up sort of using that as intuitive backup and then background and then going and implementing things on an ordinary computer to sort of, in a more controlled way see what was happening Now. Well, there's a long series of people building, wanting to build, thinking about building, proposing to build various kinds of hardware, cellular automaton devices. There are, and then quantum Cellular automata, which I sort of started. Well, I wasn't that excited about them, but they kind of got started in the early 1980s and that sort of to some extent segued into the idea of quantum dot cellular automata where you have these atoms or ions or whatever trapped in some particular site in some substance. And you can get sort of small scale interactions between local, local things, local, local atoms or whatever. And you can expect to do computation that way. And it's kind of one of the potential, though not so popular approaches to sort of making a quantum scale computer. Boy, I'm seeing lots of questions here from Liu. Curious about the history of eyeglasses. Why has the tech not seemed to have advanced much? Let's see what I know about this. Okay, so the idea, glass, transparent sort of thing you get from heating up sand, is a thing that was known to the Romans, but it wasn't something where you could get nice smooth glass and so on. It was a. The idea that you could make eyeglasses must have existed by the 1600s because there was sort of a big period of time when it was. Yeah, it definitely existed by then. There was. Lens grinding was kind of one of the sophisticated trades at the time. And for example, when the lenses weren't that good, but they were particularly relevant for folks who, you know, get old and lose accommodation in their eye lens and then you can't read close up things and that, you know, for many people who were kind of scholars, that was kind of a real problem. As they got older, they just couldn't read things. And so the invention of even very rudimentary eyeglasses, which must have been, I'm thinking 1400s maybe, I'm not sure. But then sort of lens grinding got better and better. You know, for example, Kepler, I think in the late 1500s had been, you know, was wearing eyeglasses, but yet he didn't have a telescope. And I think the reason for that was the lenses just weren't really good enough to be able to sort of combine them together and make a telescope and so on. And then when people in the Netherlands who presumably invented telescopes, this was around 1607, 1608, it's like, okay, you can take two lenses, they're well enough lens ground that you can actually see through both of them and you can make a telescope. And that's what Galileo found out about improved. And that's what led him originally he had the telescope for the merchants of Venice or whatever to be able to see, go up that big tower in Venice. Which is still there, and go see which ships are coming over the horizon to be able to trade on that information. But then Galileo had the idea, let's look up at the stars and discovered moons of Jupiter and all those other interesting things in 1608. And so that was sort of early. It was kind of new tool for science, new set of discoveries that could be made. Right around the same time the microscope was invented again just by combining pairs of lenses together. And so, you know, that was all a thing by the 1600s. I know Spinoza, famous philosopher, his day job was being a lens grinder. And I don't know, as I say, I think it was kind of a high end trade of the time. Now, in terms of sort of more detailed making of eyeglasses, I think my impression is the technology sort of steadily improved. And you know, you still, you see pictures from the early 1900s of people with very small glasses. And I don't know whether that was technological. I suspect that was technological in origin. The whole idea of all the sort of correction of a vision is something that not completely sure when that became understood, Let me see, there must have been, well, astigmatism. Who would have discovered that? I would say, I wonder when astigmatism was discovered, it must have been known in telescopes. I'm guessing that that was sometime in the 1800s, that that became a known thing. But the question of sort of grinding lenses that could have the cylindrical component to correct for astigmatism, at least approximately the spherical component and so on, that's I think, a much more recent thing. I mean, in the sort of mathematics of those things are these things called Zernike polynomials, which I vaguely think come from the 1930s, which are kind of the way of characterizing the sort of shape of a lens which determines how the wavefront of light will be distorted. And if you're. Most of the distortion comes from kind of features of the cornea. So for somebody like me who has mild astigmatism, that's because the surface of my cornea is not perfectly spherical. It has kind of a cylindrical component to it. But it's actually, if you look at it, of course I've looked at these things. You see, it's this complicated wiggly shape. It's not just a simple cylinder combined with a sphere. And that complicated wiggly shape can be represented by these higher order z polynomials. And the question then is, is there a form of lens that will back correct for that complicated wiggly shape? And the answer is not Necessarily, the question of what you can correct for is not. You know, the light can get distorted and the corrections will be. Well, you might not be able to untangle the tangling that happened. I don't know when. Multifocal glasses. Well, I mean, bifocals. Benjamin Franklin is famous for having invented bifocal glasses. And that was, I think, until. Oh, gosh, until 20 years ago, it was really common to have sort of bifocal glasses with an explicit line between a lens at the top that was for looking far away, a lens at the bottom that was for reading close up. And it's probably only in the last 20 years that the idea of multifocal glasses came in. And I think that's driven by materials development. Most glasses these days are plastic, and there's a question of can you make the refractive index of the plastic vary across a piece of material and so on? And those are the kinds of things you need to implement multifocal glasses in an effective way. But I don't think it's the case that there hasn't been. There's a bunch of technological innovation in kind of glasses, whether it's in the material they're made out of and how light they are, and how much they have ultraviolet coatings, how much they have coatings that cut down on extreme blue light that comes from computer screens and that people say isn't good for one, and all those kinds of things. I mean, I suppose the real question is when do we get kind of the type of thing that you can see in, like, the Apple Vision Pro or something where you have a purely virtualized image coming from camera on the outside to display on the inside, and that works. It's kind of interesting. You put on a Vision Pro and you can still read stuff, and all of that data is being completely digitally processed through the device. And so you can imagine all kinds of interesting things becoming possible there. All right, I think we should wrap up there, so thanks for joining me and see you another time. Bye for now.