Transcript of Episode 39 – John Koza on Bleeding Edges

The following is a rough transcript which has not been revised by The Jim Rutt Show or by John Koza. Please check with us before using any quotations from this transcript. Thank you.

Jim: Howdy. This is Jim Rutt, and this is The Jim Rutz Show. Listeners have asked us to provide pointers, some of the resources we talk about on the show. We now have links to books and articles referenced in recent podcasts that are available on our website. We also offer full transcripts. Go to jimruttshow.com. That’s timrutzshow.com. Today’s guest is John Koza, founder of National Popular Vote Inc..

John: Good morning, Jim. It’s great to be here.

Jim: Great to have you on, John. I don’t know how long we’ve known each other, but it’s been quite a while. Great to have you on The Jim Rutz Show. For those who don’t know John, he has one of the most incredible resumes of anybody I’ve known. In fact, maybe the most incredible. Back in 1973, he founded Scientific Games Inc., the inventor of the scratch-off lottery ticket, which was introduced in the 1970s. In fact, I quite remember the great hoopla around it when it was introduced in Maryland. In 1976, I was working at a car dealership, and I can recall all the other car salesmen other than me who have never found dumb ass gambling attractive were like utterly sucked into this stuff and they were constantly dashing across the street to the local liquor store to buy more scratch-off lottery tickets. That was a huge success, needless to say. If we go to any convenience store in most States, we see racks and racks and racks of these scratch-off lottery tickets. That was John’s invention.

Jim: Next, in the 1980s, John invented genetic programming, basically the application of genetic computation to the evolution of computer code. This was, again, another huge and seminal invention, which is in widespread use today. Two chip design software companies I helped launch use genetic program near the core of their products, and both of those companies were successful. Third, and I don’t want to say finally, because who knows what John will come up with next. In 2006, John launched National Popular Vote, an audacious move to replace the Electoral College and presidential elections without requiring amending of the constitution. So let’s now jump in with the first. John, why don’t you tell us a little bit about the scratch-off lottery ticket. How’d you come up with that idea?

John: Well, while I was a graduate student at the University of Michigan in the 1960s, I had a part-time job doing computer programming for a company that made commercial promotional games, the kind of games that were run at the time in gas stations and supermarkets to increase sales. One of the categories of games they had were called probability games, and these were game tickets where the customer would only rub off a subset of the number of rub off spots on the ticket, maybe three out of 12, and if the right symbols were under the three spots that the customer chose, they would win a prize. These were called probability games because every single ticket was a potential winner of perhaps $100.

John: So the tickets obviously had to have a secure covering, and that was a technology that existed at the time. But more importantly, if you were offering $100 in each ticket, savvy customers, and in particular savvy clerks in the gas stations and supermarkets might be able to find the winning spot, some tickets, by looking at little printing irregularities that existed on the tickets. So it was a problem of how to make these tickets secure so that you could offer a prize, perhaps $100 cash, on every ticket and do so safely without causing the supermarket or gas station to go bankrupt.

Jim: Okay. How did that then lead to the lottery tickets?

John: Well, we did successfully produce these probability games for commercial clients in the late ’60s and early ’70s, and we did so securely. We did it by printing roughly a half million different ticket combinations so that if you rubbed the first spot, even if there was a printing irregularity on the exterior or under the spot that a person might notice, it would be decorrelated with what was on other spots on the ticket. The way we did that is if there were 12 spots on the ticket, we might print four of the spots on one pass, and this is at a time when you were printing tickets on big sheets lithographically. You might have 10 different kinds of sheets each with say 100 tickets, so you’d have 1,000 different patterns. That would not be sufficient to provide security because if you had a given symbol under a given spot, 1,000 is simply not enough to decorrelate with what’s on the other spots.

John: So then we would take these big stacks of shapes and 10 stacks, we [inaudible 00:05:10] like the sheets one by one and then print 10 more, print four spots on each of the 1,000 tickets, so that now there were eight of the 12 spots and then we would again gather up the sheets one by one and do it a third time so that we would get roughly a half million different combinations of tickets. In particular, there would be no correlation between a pieced spot that you might uncover with one rub and where that desirable second and third symbol would be on the ticket.

John: So we were able to produce a secure game that offered a very substantial prize of $100 on each and every ticket, and do so securely. Now, what does that have to do with state lotteries? Well, the state lotteries were just getting started in the ’60s. There were just a handful of them, and they were all running rather uninteresting raffle type games where they would print tickets with, say a six digit number, and once a week they’d hold a drawing and if you match the number or part of the number, you’d win a prize. So there was no immediate gratification, there was no particular excitement or artisticness or attractiveness to the tickets, and sales were languishing in all the state lotteries.

John: So we had the idea, when I was working for this commercial game company, of saying, “Well, if the state lotteries could have an instant gratification, use this rub off technology… Although states would not want to run a building game, that would have been much too risky for a state government to undertake, we could take this super secure technology that was good enough to offer a substantial prize on every ticket and run a controlled game where the winners were predetermined and preprinted, and again, successfully prevent any clerk in a store from finding the winning tickets, plucking them out so that the consuming public wouldn’t be getting the winners. So we solved what was called the agent pickup problem for controlled games, which is, could the clerks who were selling the tickets in these hundreds of convenience stores around the state… that they couldn’t pick out the winning tickets.

Jim: Though I do remember it in Maryland, again, my home state, there was a big scandal, because it turned out there was a way to do it, which was that the winning tickets were slightly smaller than the non-winning tickets. I don’t know if they were printed separately before they were mixed in or what, and there was a gang of criminals that went around to all the convenience stores and all the liquor stores and paid the clerk some amount of money to go through the tickets and measure them and were able to pull out the winners. So there was at least one way somebody did hack that particular problem. Of course, they fixed that, but it’s an example of how smart adversarial agents were.

John: That was, of course, a commercial coming out of the state lottery. There was never an instance of a successful pick out in a state lottery.

Jim: No, that was the scratch-off Maryland State Lottery. They were able to hack it based on the sides of card. I do recall that quite vividly. I believe it was probably in the late ’70s, maybe early ’80s. But they fixed it. It was probably a problem where they were having them printed at two different printers and the size of stock was just slightly smaller on the winners. But anyway. So you develop this technology so that no one could get through the print, or at least not the cover thing. I imagine you could probably do it with an electron microscope or something, but not worthwhile to do it at clerk rip off level. How did that then turn into a company, and how did you convince the states to do this?

John: Well, because we did have a secure method of printing the tickets… I don’t know what you remember, but I can assure you none of the tickets produced by Scientific Games had ever been compromised in that fashion. Perhaps they ran a game with somebody else that I wasn’t aware of, but we had a method to securely print the tickets so that there would be no agent pick out. We first convinced the Massachusetts Lottery to run an instant lottery, which was a very big step at the time, because they were introducing a new type of game with this potential agent pickup problem. They did that in the spring of 1974, it was immediately very successful, triple the sales of the Massachusetts Lottery, and every subsequent lottery adopted it, and that was the beginning of our company, Scientific Games. Our business was selling instant lottery tickets to state lotteries.

Jim: Cool. Did you guys basically own that business or was it… Did you have competitors?

John: Well, when we started, we were actually the fifth company to get the contract from the Massachusetts Lottery to run their instant game. But each of the previous four companies that got the contract, which included some very well known bank note printing companies, each of them, when they printed the sample tickets for the lottery, the lottery was able to compromise the tickets and the contracts were canceled. So we were the fifth company to get the contract, and we successfully printed a secure ticket and got it on the street and produced the commercial success. From that, we went to all the other states at the time and offered the instant lottery. Of course, the state lotteries were quite anxious to start running the instant lottery.

Jim: Yeah. As you say, they’re big, big business for the states today. What do you think of the morality of things like that? I’ve read a fair amount that says that the use of these kinds, particularly the scratch-off lotteries, is inversely proportional to people’s incomes.

John: Well, I haven’t been involved in the lottery business for 30 years, but it was definitely not the case when I was there. In fact, we did extensive marketing studies. [inaudible 00:11:23] lottery, and they’re actually higher income profile than lotteries in general, that is the raffle type lottery or the lotto games that came along, mainly because the ticket was a more expensive ticket. It was a dollar ticket of the time, whereas other lottery tickets were 50 cents. So at least at the time when I was involved in the business, which was until the late ’80s, the demographics of the instant ticket was distinctly not lower income, it was distinctly middle income.

Jim: Okay. Well, that’s good. Well, that’s enough on the lottery. Again, an amazing insight in innovation and invention and then take to market with high quality. Let’s move next to genetic programming. As I understand it, you are a student of John Hollands, who’s one of my heroes. In fact, his book, Adaptation in Natural and Artificial Systems was the first book that I read in the area of evolutionary computation, and it drew me into working in that area and then eventually into the area of complex science. Maybe you could start a little bit with your work with John Holland and then how that led to genetic programming.

John: Well, again, while I was a student of John Holland’s in the late ’60s and early ’70s in the computer science department at the University of Michigan and he was my thesis advisor, he had invented the genetic algorithms in the mid ’60s, which is an evolutionary computation technique that is useful for solving a lot of engineering and optimization problems in an automated way. And then as matter of fact, as I was a graduate student working on my PhD thesis, that was at the same time that I was doing the part-time consulting for the game company that was producing the probability games.

John: In any case, I had gotten interested in the notion of automatically programming computers to solve problems. In 1963, I was in John Holland’s… one of his courses. I was an undergraduate at the time, and he had a course on adaptive systems. One of the papers we read was Friedberg’s paper on trying to have a computer automatically learn how to program itself to solve a problem, and that was through a mutational method. I was very taken by that idea at the time, and my eventual PhD thesis was somewhat related to that. It was not actually a thesis on genetic programming, it was attempted algorithm to infer grammar from data. But inferring grammar from data, it can be seen as a way of writing a computer program.

John: In any case, my thesis was sort of inspired from the field of genetic algorithms, even though it wasn’t in the field of genetic algorithms. It was inspired by this idea of trying to get computers to program themselves, even though it wasn’t exactly that. So somewhere around 1976, right when John Holland’s book came out and I was in the lottery business the time and quite busy running around selling tickets to state lotteries, I did buy a copy of his book right when it came out, and as you are influenced by it, so was I.

Jim: Very good. That’s interesting. For our audience, if you could go into the basics a little bit, just keep it very basic, we have a well educated audience, but not experts in either computer science or a genetic or evolutionary algorithms, how do genetic algorithms work? And then how was that idea extended to genetic programming?

John: Well, the idea of the genetic algorithm is vaguely patterned after the notion of DNA, that is, you’d have a linear string of symbols, with DNA it’s four symboled. In the classic, simplest form of the genetic algorithm, the string is just a binary string of ones and zeros. Those ones and zeros mean something in relation to some problem. For example, the first few symbols might together specify the size of an insect’s wing, and maybe the next few symbols might specify the body color, and maybe the next ones would be the type of claw the insect has, et cetera.

John: So the idea of the genetic algorithm was that there would be a string of ones and zeros that would represent something in the real world, something perhaps you were trying to optimize, and you would do the optimization in the following way. First of all, you would create a random population of these strings of ones and zeros, maybe 1,000 strings of length of, say 50, and totally at random. And then you would look at each of these random thousand individuals and say, “How good are these insects at doing whatever they’re supposed to be doing,” maybe foraging and getting fat on the food they eat. So you would assign a fitness to each of these thousand randomly chosen strings, fitness being how good are the things that these strings represent? It’s solving the problem at hand, which is, say getting fat. I’m happy.

John: That creates a population with fitness, which is somewhat similar to the Darwinian notion of natural selection. And then the notion of the genetic algorithm is you start with a random population of individuals and you modify that population and create a new generation. With a small part of that thousand, maybe 10%, you just do a random mutation where you flip one of the bits from zero to one or one to zero, and that would correspond to a random mutation, say of a DNA string, the same sort of random mutation where if it happens at the wrong place can make the difference between having sickle cell anemia because of having defective hemoglobin or not having sickle cell anemia or cystic fibrosis or not having cystic fibrosis.

John: So mutation is a way to slightly improve a given individual by making a very small, very localized change. That notion of breeding a population of binary strings had been floating around since the late ’40s. Actually, one of the first computer runs on the computer that they had at the Center for Advanced Studies at Princeton was a simple genetic algorithm run made in, I believe, the late ’40s. But what Holland added was the notion that the real engine of creation in nature and evolution is sexual reproduction, where a major part of the genome, this string of length 50 comes from one parent, the other part of the string comes from another parent. This is roughly analogous to the way sexual reproduction occurs in simple genomes. That would be the bulk of the transformation that creates the new generation.

John: So maybe 10% might be a mere copying based on reproduction and/or some mutation, but maybe 90% or 80% would be crossover, that is the sexual recombination where you take a major part of the genome of one individual and you mate it with a major part of the genome of a second individual, creating an offspring that has a major number of characteristics from the father and a major number of characteristics from the mother. So that takes us from the initial random generation to the first generation. And then the essence of the genetic algorithm is to go through many, many generations, maybe hundreds of generations or thousands, and constantly refine the population based on a bit of just straight reproduction, copying individual proportion based on their fitness, a tiny bit of mutation, again, slotting the individual’s proportion to their fitness, but then a major portion of the population being shuffled around in this very disruptive way called crossover to produce new individuals. It turns out you can solve a whole bunch of very interesting engineering optimization problems by using the genetic algorithm as well as a whole variety of other problems.

Jim: Yeah, it’s quite remarkable. I remember when I first started writing evolutionary computation, you’d go, “Holy moly, I do nothing at all other than set up this environment and evolution itself drives the solution.” As you say, many, many problems, it works. Sometimes it works very slowly. I would often describe genetic algorithms or my own particular variant, which was neural nets encoded as genetic algorithms… This was back in 2001. I’d say it’s a very general method, but a very weak one, in that it takes a very long time and a lot of computation to converge on a solution, at least for a difficult problem.

Jim: I’d also like to highlight for our audience the key factor, you mentioned it a couple of times, I’m going to mention it again, which is that these parents for the sexual reproduction of these genomes are selected in some fashion that’s proportionate to their fitness. Now, there’s lots of different algorithms by which one may choose to select. You might do rank order, you might do weighted strength, actually how strong the thing was relative to the alternative, you might do tournament selection. This is actually what I tended to use a lot, which is I’d grabbed four parents, have each of them play against each other and then the winner of each of those would reproduce. There are many, many different ways to tune that proportional selection for reproduction, and it turns out how you tune it tends to impact how quickly the genetic algorithm will converge towards a good answer.

Jim: Now, there is no right answer to all that from an important theory, which we’ve discussed before on this show, the no free lunch theorem. There’s no best way for all problems, which is one of the challenges of these techniques, is that the tunings, the meta settings for evolutionary algorithms, you can make some good guesses, but at some level, they’re trial and error to get those things set correctly. So, that’s the genetic algorithm. Could you explain to us how you extended the idea of genetic algorithm to the genetic program?

John: Well, so as I said, I got Holland’s book in 1976, but I was the founder and CEO of a company that I was running at the time selling lottery tickets to state governments. So I was quite busy, and I sort of followed Holland achievements with the genetic algorithm and all the other people who picked up the idea and refined it and introduced all the techniques, Jim, that you were just talking about, different variations and ways of doing it. But I was involved in the lottery business till ’87, between 1973 and ’87, and it was sort of an academic interest of mine to be watching what was going on in this new field of genetic algorithms.

John: So I left the lottery business in ’87, and at that point, I was involved a little in venture capital, but I got involved with some people at Stanford University in the computer science department and got to talking to them about genetic algorithms and said, “Well, wouldn’t it be nice if there was a course at Stanford University on genetic algorithms?” So they appointed me as a consulting faculty member, ultimately a consulting professor, and I started teaching a course at Stanford starting in ’88, which was just about the time when the genetic algorithm field was really starting to take off and started to have conferences and public proceedings. Lots of papers, as opposed to one or two papers a year, there would be a hundred papers a year in this growing field of genetic algorithms.

John: Of course, what motivated me was still this memory of Friedberg’s paper that I ran into in 1963, which was the notion that you could have a computer program itself to solve a problem. Now, Friedberg’s work was with binary strings, and you could quickly see that binary strings were not a particularly expressive way to deal with a problem like this. So I said, “Why couldn’t we apply the genetic algorithm to whole computer programs, and in particular whole computer programs in leaves, which was just simply a functional programming or tray like representation of [inaudible 00:24:50]?” So if you’re doing an addition, you have a function, addition, and two arguments and you add them together. If one of those arguments that you’re adding together was the product of say multiplying two other numbers, you would have this tree of operations.

John: So you can represent computer programs in functional programming languages as program trees. The interesting thing about program trees is that automatically suggests to you the idea of regarding a portion of a large program tree as part of a genome. If the program tree is just a little good at solving a problem and you’ve got another program tree over here that is just a little good at solving that same problem and you took a big chunk of one program tree from the father and another chunk of the program tree from the mother and you swap those sub trees, in effect, you would be doing something like the genetic algorithm does, one that takes a portion of a binary string, but you’d be taking a coherent chunk of a computer program.

John: That’s the notion behind genetic programming, namely you starting out with a population of 1,000 randomly created binary strings of length 50, but you start out with 1,000 computer programs, each of which is a program tree containing multiplication and addition and conditional operations and maybe some co-signs and exponentials and some other things, and you ask how good is this program at solving the problem that you have? How good is at doing a sorting algorithm? How good is it at creating a polynomial that matches a given sample of data? Maybe the stock market data. How good is this program at controlling the operation of a robot if the operations in the program are say, moving forward and turning and moving backwards and reaching out and so forth?

John: So what you get by moving into the world of computer programs instead of binary strings is a much greater richness of expressiveness, because you can have operations that are much more meaningful, addition, subtraction, move the robot forward, shuffle the things if you’re trying to develop a sorting algorithm and so forth, the individuals in the population would be much more expressive. Somehow it intuitively makes sense that if this program is somewhat good at solving the problem and this other program is somewhat good, maybe if you cross over, do the sexual recombination of part of one good program and part of another program and do it enough times, that you’ll get a offspring that’s better than either of the parents. It’s solving the problem.

John: That’s the essence of genetic programming, that you start with a random population of computer programs, you run the programs, see how good they are at solving the problems. Based on their fitness at solving the problems, you do a small amount of mutation, very small, one or 2%, a bit of reproduction, maybe 10%, and then you… with 90% of the population, you do these crossovers where you take a big chunk of one program and cross it over with a big chunk of another program and see if you get something better. It turns out you can solve just an enormous variety of problems with this genetic programming technique, whether they’re data matching programs, what we call symbolic regression, or control programs of controlling a robot to do something, or a controller and the sense of a PID controller, or whether it’s an electrical circuit. Is this a good filter circuit? Is this a good to amplifier? Is this a good temperature sensing circuit? Et cetera.

John: Turns out you can evolve computer programs that solve just a wide variety of problems in optical lens design, controller design, antenna design, data matching, robot control, starting algorithms. A number of people have worked on quantum computing algorithms that they evolved with genetic programming, and just the vast variety of problems.

Jim: I’m always amazed when I see the ever growing list of things that genetic programming can be applied to. What I remember from the relatively early days, I think I actually saw you present on this at Gecko, was that you had your genetic programming environments… had discovered some analog circuits that had actually been patented. Could you talk about that a little bit?

John: Well, yes. So one of my things that interested me of course was doing what I would call… The aim, of course, was to create some human competitive results with genetic programming. One of the categories we worked on a lot was analog circuit design. Now, there are all sorts of algorithms for digital circuit design, but analog circuit design is an art, as is antenna design and controller design. Frankly, most engineering design problems are matter art as well as mathematics and science. So we said, “Could we use this to automatically create analog electrical circuits?” That was one of the big projects we worked on.

John: Obviously, you could use it on digital circuits. That was quite straight forward. The idea there was… We went back through the patent literature from the early 1910s and twenties when telephones was the leading technology and things like filters and amplifiers and feedback were big issues, and we started saying, “Well, suppose we had a bunch of resistors and capacitors and inductors,” which is what they used in at that time, periods and telephones, “could you get an automated process that will just combine these common ingredients to produce circuits?”

John: Well, what’s a filter? Well, a low pass filter is like the woofer on your hi-fi, it passes the low frequency signals through freely and it suppresses the high frequency signal. Well, it turns out if you create a random population of circuits and ask, how good are these circuits at filtering a signal, you can easily compute the fitness and see how well does it pass, say frequencies from zero to 1,000, and how well does it suppress frequencies from 1,000 and above? So you can compute fitness and then you can start slightly mutating a few and reproducing a few, but more importantly, crossing over circuits to just do a little well at filtering with each other.

John: Lo and behold, we started evolving the Campbell filter, which is a classic ladder filter of inductors and capacitors that was patented I think in 1917, if I recall. And then of course, there were all sorts of improved filters, the Butterworth filters and the elliptic filters and the Cower filters and the M-derived filters and so forth. There’s a whole Johnson filters, there are a whole variety of filters that had particular additional characteristics. They were very good at creating a steep fall off or a this or that or more attenuation or what have you.

John: We were able to reproduce about a half dozen of the filter patents of the 1920s and thirties up to including the elliptic filter, which is, in a sense, one of the most complicated ones using genetic programming. And then we got involved with all kinds of other circuits and said, “Well, transistors were invented by the late ’40s and early ’50s. Why don’t we throw in transistors as well as resistors and capacitors?” Lo and behold, we started being able to evolve things that were patented in the ’50s, ’60s, and ’70s with transistors. That is significant invention, not just novelty toy circuits.

John: And then what you might have heard, Jim, at Gecko was after the new millennium rock, we found six patented circuits that had been patented by different universities or big companies, and we were able to recreate a circuit that performed as well or better than the patented circuit in six specific cases. In one of the specific cases, the genetically evolved circuit was essentially the patented circuit. In five of the six, the circuit was equivalent in performance to the patented circuit but quite different in the way it solved the problem, which is to say we were genetically evolving circuits that would avoid infringing the patented circuit. So not only were we reinventing things that were considered significant inventions by universities or companies at the time and worthy of patents, but we were duplicating patents and creating a method to essentially avoid infringing existing patents by creating equivalent circuits. Then we played the same game with optics.

John: Now, I had taken one electrical engineering course when I was a graduate student, so I’m not an electrical engineer, but I knew just a little about electrical engineering, knew absolutely nothing about optics. With the help of several programmers who also knew nothing about optics, we were able to duplicate six 21st century optical patents, that is, we were able to genetically evolve a lens system that equaled or exceeded the performance of six lens systems that had been patented by big companies in the optical field.

John: So we demonstrated that this process not only worked in a important area, electrical engineering, but also in another area where we could go into with no particular expertise of our own about lens system. So simply apply this domain-independent, well-established, well-defined mechanical method and duplicate patents. We duplicated some antennas, antenna patents, and controller patents. Other people, of course, who’ve worked on this have created all sorts of interesting other inventions of quantum computer circuit programs and many other things.

Jim: Again, just to underline for our audience, all these things were done untouched by human hands. It was not like a bunch of high priced computer guys sat down and wrote this stuff, these things were evolved, which is amazing. As I think I mentioned on the intro, I’ve actually been involved in two companies that use genetic programming. I was chairman of Analog Design Automation, which used genetic programming to improve human designed analog circuits. Our lead investor was Intel, we had all the big chip companies as our customers. We ended up selling the company for a nice price, if I may say so, to Synopsis.

Jim: And then we had another company called Solido Design Automation, which used genetic programming and other techniques to optimize manufacturability of computer chips. And then again, that company was very successful and was sold to Siemens not too long ago. So there’s just some real world, really hard problems that GP has been applied to. John, do you keep up on the state of the art in GP and do you know of what’s going on on the cutting edge these days?

John: Well, I’ve been staying closely involved with the Annual Human Competitive Competition, which is now in its, I think, 20th year. Every year, they run a competition where they give prizes for human competitive things that have been designed with any evolutionary computation technique, genetic algorithms, genetic programming or other techniques. The majority of the prizes have gone in the genetic programming category, but there’ve been a whole string of impressive results every year. If people go to www human-competitive.org, they can look at the progression of increasingly impressive things that have won prizes over the last 20 or so years.

Jim: Cool. Could you call out a couple that you were impressed by recently?

John: I personally it was very impressed with Lee Spector’s work on quantum computer circuits, and [inaudible 00:38:17] repeated winning a prize as both of them repeatedly won prizes in that competition in the area of game playing. But there’s been just a whole number of other ones. I don’t mean to slide anybody else by having mentioned those two.

Jim: Wow. Very good. What an amazing idea you’ve had, John, and how far it’s gone. So now let’s move on to the third area of innovation, and this is the national popular vote initiative. Could you start by telling us about it? Just kind of the high level, what the motivation was, and then I’ll ask you some more questions, we’ll go into some more detail.

John: Again, it goes back to my very busy career as a graduate student at the University of Michigan in the ’60s, very busy on frankly everything but my thesis for most of the years I was at [inaudible 00:39:09]. So I was had this significant part-time job of programming for the game company while I was working on my thesis. But of course graduate students then and now were very occupied with playing games as any previous graduate student well knows. In the ’60s, very complicated board games were very much the rage. The Avalon Hill war games and Avalon Hill have a number of other complicated games involving railroad, Switchyards, and there was diplomacy and risk and a whole bunch of games, Stratego.

John: But we got interested in creating our own game, so we created a game based on the Electoral College, which was topic of quite prominent debate in the ’60s. So we had this game that John Holland played with the graduate students. We had a computer program that would run simultaneously with playing the game, creating different statistical matrices so we could evaluate the state of the play at each stage, and we would pretend to be presidential candidates campaigning for president using the quirks and oddities of the Electoral College.

John: So this game, it was called Consensus, which we actually published commercially, lost the money on, and sold 3,000 copies, but this game was sort of always in the back of my mind how utterly ridiculous the Electoral College is as a method for choosing the president and the how subject it is to manipulation and producing odd results. So in the ’90s and 2000s, particularly the 2004 election, it had become increasingly clear that presidential campaigns had concentrated only on a handful of states. When Kennedy ran in ’60, he campaign in 35 States, and Nixon campaigned in all 50. When Kerry and Bush were campaigning in 2004, almost the entire campaign… Well, virtually, the entire campaign was in a dozen states, but essentially the whole campaign was in three states. That’s because under the current system, the candidate who gets the most votes inside each state gets all of the state’s electoral votes.

John: So if a state is 5% or 6% in one direction, it’s really out of reach during a presidential three month campaign. So if you’re running for president and you want to win, you do all your campaigning in the States that are within one or two or 3%, the so-called swing states or battleground States. This battleground that had been fairly large in the ’40s, ’50s and ’60s had shrunk by 2004 to just a handful of states, which is of course still the case today. The result is not only most of the people in the United States are politically irrelevant in the presidential election, in the general election, but you could also get candidates, elected president, who didn’t get the most votes in all 50 States, which of course occurred in 2000 and then more recently in 2016.

John: Just in general, obviously every vote was not equal, and most voters were just politically irrelevant. So it turned out I have lost a 25 cent bet when I was an undergraduate to a pre-law student in the dorm at Michigan. I had claimed that the winner-take-all rule was in the US constitution and the law students said, “No, no, no, it’s not in the constitution.” I did lose the bet, and I still… I remember that, because he was right. It turns out the way we elect the president isn’t really in the constitution. It’s in state laws, state by state, that have adopted these laws that award all their electoral votes to their candidate who gets the most votes inside the state.

John: I got to talking to a little lawyer friend of mine, he was a lawyer friend because when I was in the lottery business from the ’80s, he and I wrote lottery laws that we got different legislatures to pass, which we passed in several cases in California, Oregon, Arizona, through the initiative process. So he and I were quite aware of the process of lobbying state legislatures and passing laws by initiative, and we got to talking one day at lunch and said, “Why couldn’t you elect the president in the national popular vote, not by amending the constitution, but just by changing these state winner-take-all laws?”

John: We started researching this and quickly discovered that, well, not only was the pre-law student right back in the early ’60s when he swindled me out of 25 cents, but it turns out that the way to do it is with the States getting together with what are called interstate compacts. These are legally binding contracts among states do things. Usually, a state enters an interstate compact only because it gets something and gives up something to other states. In fact, an interstate compact is a way of resolving a prisoner’s dilemma, that is, no state would want to change its method of electing the president from winter-take-all to some other method unless it got something in return, in particular, didn’t reduce its own influence. So it turns out an interstate compact is a way to unravel a prisoner’s dilemma that might exist. So we combined the idea of the interstate compact with the idea that the legislatures had the exclusive power to decide on the president elected and created some legislation, which we introduced in a book in 2006, and we started going around to different states trying to get them to pass this law.

John: Maryland was the first state to enact it in 2007. So thank you Jim for being a Maryland person and being part of getting National Popular Vote started, and we went into this organization, National Popular Vote, started lobbying state legislatures, and one by one we got legislatures to start passing this, usually about one a year, until 2019 when we got four states in one year. Now we’re in a place where we have 15 states in the district of Columbia with 196 electoral votes that have passed this interstate compact. It will go into effect when states having half the electoral votes, which is 270 out of 538, pass the same law. We’re working to get the remaining states, which would be… We need states with 74 more electoral votes to get to 207.

Jim: Wow. What an amazingly audacious thing, all starting from a game. It’s interesting, I was another one of those people that loved Avalon Hill games [inaudible 00:46:28] I started playing Tactics II when I was nine at a friend’s house that I actually bought used a copy of Stalingrad when I was 10 and became a hell of a good Stalingrad player and slayed many of the rest then even more humorously, I later made a small investment in the Avalon Hill company, which was publicly traded, and I added to it and added to it. The stock never did well, the company was horribly unprofitable, was run very badly by a couple of clowns who had bought it.

Jim: Well, I got to meet Curt Schilling, because it turned out that he and I were the two largest non-insider shareholders, and we conspired together to try to buy the company one time but it didn’t work out. But we ended making some money when the company was acquired by Hasbro, I think it was, or one of the big game companies. But anyway, a little interesting history on games. But now let’s get back on the popular vote. Could you explain a little bit about what a state compact is? You kind of touched on it. Then maybe more specifically, what’s in this proposed state compact?

John: Well, yes. So let me first connect the and the state compacts with the lotteries, because the reason we thought of this is that in the mid ’80s, lotto games were getting started with state lotteries, and lotto games offered a very large prize as you know. Some of the smaller states, Maine, New Hampshire and Vermont, because they had so few people, they couldn’t run a lotto game the way to New York or New Jersey could with a big multimillion dollar prize. So they got together and passed an interstate compact creating the Tri-State Lotto Commission, and that’s how I became aware of interstate compact.

John: As it turned out, my company, Scientific Games, won the first contract to run up a lotto game for the Tri-State Lotto Commission. That was our first contract for a lotto game, our specialty being instant games, but we were branching out into lotto games. So I became aware of the idea of an interstate compact and also became aware of the fact that Congress usually doesn’t have to consent to states entering into interstate compacts. Now, many times Congress does have to consent, but we have looked into it as a legal matter because when we were in the process of bidding on the contract, we were scratching our heads wondering whether this Tri-State Lotto Commission was a legal entity and actually had any legal authority to run this game or give us a contract.

John: It turns out that if a topic does not threaten federal supremacy, the states can enter into a compact without congressional consent, and gambling was considered something that did not threaten federal supremacy because Congress basically took the position that as far as they were concerned gambling is illegal. Similarly, because the Constitution specifically gives the state legislatures the power to decide how to award their electoral votes, this is another area where an interstate compact would not require congressional consent. Not that we were against having congressional consent, but obviously it’s another step, and we were pleased to discover that at least based on existing judicial precedents, a compact that involves some exclusive state power and doesn’t require congressional consent.

John: So to answer your question, Jim, how does the compact work, it says basically, Delaware, which one of the states that passed our bill in 2019, will give its three electoral votes to the presidential candidate who gets the most popular votes in all 50 states in the District of Columbia, that is Delaware is throwing its three electors into a pot of electors that’s going to go to the candidate who wins the national popular vote. Now, of course, Delaware wouldn’t want to do that alone, that’s the prisoner’s dilemma aspect. It would only want to do that if it knew it was going to change the system to create a national popular vote.

John: So the second operative clause of our compact is, the compact doesn’t take effect till states having a majority of the electoral votes enact the very same law. So when we get to states with 270 electoral votes, there will be a group of states that have passed the same law saying they’re going to give all of their electors to the candidate who gets the most votes in all 50 states. Then that law goes into effect, and you have a national popular vote for president because whoever gets the most votes in all 50 states is going to get at least 270 electoral votes, which is enough to become president. So you have a method, by changing state law, of changing the way the president is elected in all 50 states.

Jim: Very, very, very clever. We did learn a little bit about the prerogatives of the state to do what they want with their electoral votes during that Florida snafu in 2000 where the Florida state legislature was threatening to just flat out appoint electors for George Bush if either the U.S The Supreme Court or the Florida Supreme Court ruled the result the other way, and constitutional lawyers said that they had a pretty good argument to make there. A few questions for you. You list your website does 16 jurisdictions, 15 states and the District of Columbia that have enacted your multi-state compact. As I’m looking down the list, all of them are either mostly Democratic or Democratic leaning states or jurisdictions. Any reason why that seems to be the case?

John: Well, when we started in ’06, George W. Bush was still president, and Republicans generally didn’t support this heavily. We would get a few Republican votes. Then in 2008, as he was leaving office, we started getting about a third of the Republicans in state legislatures and roll calls to vote for our bill because the Bush 2000 election was receding from memory. It correctly occurred to people that this had nothing to do with the 2000 election, but everything to do with future elections. As a matter of public policy, it is not good that people get into the White House who don’t have a strong popular base of support, and it’s certainly not good that three out of four states that are ignored politically in the presidential election.

John: So we went through a period then, when Obama was president, when the Democratic states got less and less interested in passing this bill. First of all, we had already gotten several Democratic states. Secondly, there was a sort of arrogant, triumphalist mentality that had set in on the Democratic side. The Democrats had a permanent lock on the White House. And why change the system since it had elected Obama? Now, curiously, 1989, the phrase Electoral College lock was invented by a Republican, Bigsby, who said that the Republicans had a lock on the Electoral College because Reagan and Bush had just been successful and Nixon had won twice before that and this many states had voted Republican five out of six times.

John: Well, the same completely nonsensical, arrogant belief settled in on the Democratic side, it was called the blue wall theory, that because states with 242 electoral votes had voted six times in a row, that Democrats would always win the White House. Well, that blue wall of theory went out the window just as quickly as the Republican Electoral College lock went out the window since the next election arrived. So we went through a period when we started getting a lot of Republican support, passed the Oklahoma [inaudible 00:54:42], which Republican controlled the Arizona House, two thirds of the Republicans sponsored the bill and two thirds of both parties voted for it, and 47 seven out of 56 Georgia senate members sponsored the bill in 16. The only reason it became law in New York was that the Republican controlled senate passed it.

John: So we went through a period where Republican support was increasing. And then of course we had the 2016 election, and then suddenly instead of like a square dance, everybody changed party lines and all of a sudden the Republicans no longer thought that it was important that every state matter be politically relevant and that every vote would be equal, and the candidate with the most votes would. All of a sudden, the Democrats said, “Oh my goodness, the Electoral College is permanently biased against us.” So we’ve gone through this cycle of partisanship, and we’re plodding along, and you have to just explain the issue, one by one, to the 7,400 people who actually have a vote in this matter, which are the 7,400 state legislators.

Jim: Interesting. Of course, what I always find interesting… how short term people’s reactions are. But the one you didn’t mention it was the close call in 2004 where Bush had like three million popular vote margin. But if 200,000 votes had switched in Ohio, John Kerry would have won the Electoral College with a significant shortfall on the popular vote, again, on the Democratic side. So people forget all about 2004, I guess because it didn’t happen. But it was a close run thing. I’m with you, that people who try to lock in short term advantage are not thinking properly. But I just had an idea I’m going to throw out to you. I think I know how you can get this thing passed by both Democrats and Republicans. Are you interested?

John: Well, of course.

Jim: Okay. Suppose you amend it, you’ve got to get back everybody to amend it, and say that this will go into effect 20 years after the 270.

John: Well, as a practical matter, that wouldn’t get you any votes. It’s an interesting idea, but it doesn’t change the way people look at it at the moment. The 2004 election wasn’t the only near miss. Carter almost lost the election even though he was ahead by 3 million votes. People forget Nixon, in ’68, just squeaked by, with a very tiny number of votes switching in a couple of states would have made Humphrey the president even though Nixon did win the national popular vote in ’68. Similarly with Truman Wilson’s re-election in 1916 was a matter of a switch of 3,000 votes in California, even though Wilson was extremely far ahead in the national popular vote. So the system we have is very quirky, that’s why the game we had in the ’60s was so interesting, because it had so many quirky and frankly ridiculous results that you can get out of this winner-take-all system. But [inaudible 00:57:56] was hardly unique.

Jim: Yup. That’s a very good point. I don’t know if you remember this, John, but you sent me a draft of your book when it was in process. That was 2003 or 2004, and I read it and we had a phone call. I maybe mailed you a few comments. I was going back through my memory, and I still have those same three concerns that I had then. So let me run them by you and get your reaction. I’m sure these are not the first time you’ve heard these, because you’ve been out there being shot at by people for many, many years. So I’m sure you’ve got good answers. But at the time, these were my cautions, at least, which is, as you point out, there’s been many, many close elections at the presidential level. In fact, I call, after we had our conversation, I wrote a little simulator to see how often very close elections in the aggregate popular vote would occur, even assuming a fairly large variance at the state level, and the answer was a lot.

Jim: So what happens if we do have a really, really, really close one, so close that if it were to have a state level, we could demand a recount? For instance, in 1960, it was .1% of the popular vote, in 1968, .7, 2000, .5. Most states have at least some form of re canvassing at the 1% threshold, and often at .5%, there is the ability to demand a recount. How in the world could we do a national recount of a popular vote that was under half a percent, let’s say, particularly when not all the states are involved in interstate compact?

John: Well, the reality is, you can’t do a recount now in presidential elections. You recall there was no recount in 2000, even though the difference at the state level was 537 votes, and 537 votes determined who was president of the United States. There was no recount in Pennsylvania or Michigan in 2016 simply because same as in 2000, the candidate who was initially had went into court and found a technicality and ran the clock out. The key point is there’s a clock on this recount. Because of the 20th Amendment which moved the inauguration from March 4th to January 20th, there’s only seven weeks between the election and the inauguration, and there’s actually only a few weeks in a practical sense, because it takes a couple of weeks to create the first count.

John: So the other practical consideration is most state laws don’t actually work to let you have a recount. Take Ohio, which is quintessential battleground state, it turns out there’s a 15 day waiting period in their recount law. Well, between November 4th and December 12th, you just don’t have 15 days to wait to start the court proceedings after you’ve spent two or three weeks with a count. So the first problem is effectively, there are no recounts practically possible right now in presidential elections. Secondly, they would be much less frequent in the national popular vote. The threshold is not a tenth of 1%. The number of votes changed in a recount is about 300 at the state level, and that’s based on actual statistics of the 28 recounts over the last 20 years. The chance of a recount is one in two hundred.

John: So if we ran 200 presidential elections, yes, once in eight 800 years, you would be close enough to warrant a recount. These recounts that occur in some states when there’s a 1% gap are public relations recounts. They’re designed to instill public confidence. But 1% of votes do not change in any recount, at least any recount that I’m aware of. Certainly not in the last 20 years where the average change in a recount is three hundred votes, and where six out of seven recounts simply reaffirm the previous results.

John: But to answer your ultimate question, how would you run a recount, you you need a national recount law. We wrote one and put one on our book, which is more needed under the current system than would be needed under national popular vote because the current system creates 50 opportunities for a recount every four years whereas we would create only one opportunity for a recount in four years. But we do need a right to recount that’s unconditional and can’t be brushed aside with delaying tactics and self-serving or running out the clock, which is what occurs now. So recount is a messy area. All I would say is things would be better under national popular vote because there would be less probability of having a recount in the first place.

Jim: That seems reasonable. But if we did have one early without the national recount law, we’d have a serious problem.

John: Well, of course we would. But the reality is you have a serious problem now because if there was a close vote nationally… Remember, 12 of the 50 states are closely divided battleground states. There would also be probably 12 states that are close, and you have recounts there.

Jim: Yep. That’s a good point.

John: Look, if you have to have a national recount, you’d have one. Almost all the votes are counted in the first 12 hours. That’s how we get election result by election night or Wednesday morning. So in a parallel process, it’s not like 137 million ballots are dumped on the desk of one guy in Washington and he’s told he has to recount them, votes would be recounted the same way they are counted, which is at the local level, in parallel, by teams in every locality and every state. So I wouldn’t get too wrapped up over that. There are more lotto tickets sold on a given week than there are ballots, and the lotto tickets are issued with the same sort of accuracy as credit card transactions by the hundreds of millions. There’s no big deal conducting a recount because it would be conducted in parallel.

Jim: Okay. That’s a tolerable answer. Let’s hope it doesn’t happen. Next one, I think this one was the one I was most concerned at the time, was, what happens if it’s very close and that one of the compact states that its own popular vote went strongly the other way and it decides to renege? It’s at least strongly arguable that the US Constitution gives state legislatures absolute discretion on how to allocate electoral votes. [inaudible 01:05:01] the state legislature meets between the Tuesday election date and they’re meeting the Electoral College and a legislative vote says, “Nope, we’re going to renege, and here’s our vote, all nice and legal under our state constitution. We renege, and we’re appointing the electors for the other guy.” And that then breaks the compact.

John: Oh that’s a total non-problem. First of all, the state legislature’s power to decide how to allocate their electoral votes is substantial, but it’s limited by other provisions of the Constitution. Notably, the fact that interstate compacts are contracts, and the Constitution prohibits a state from impairing an obligation of contract. So it’s a really airtight constitutional law that no state can get out of an interstate compact except in the manner that the compact itself allows. Our compact says if you try to leave between July 20th and January 20th of a presidential year, your departure occurs after the inauguration of the new president.

John: There’s been no case where the Supreme Court or any court has ever allowed any state to wiggle out of any interstate compact ever. So I wouldn’t worry about that. Also, there is a federal law on top of it which requires that presidential electors be selected based on laws in place prior to election day. So there’s also a statutory reason why that can’t happen. In a practical sense, it can’t happen because most state legislatures are not in session in that short period. Even if they were, most state constitutions only let a new law take effect 60 or 90 or 100 days after it gets passed. So it wouldn’t even matter if there were no impediments clause, if there were no federal statute. Even if all the legislatures were in session, most states could not pass a law that would go into effect quickly enough to renege, even if it was possible.

Jim: Although, of course, Florida actually was going down that road to do so in the 2000 election.

John: Well, actually, they weren’t. What Florida was doing was they were passing a resolution saying that if the courts vacated the results of the vote by the people on November 3rd or whenever it was, then the legislature was reaffirming the choice of the Republican slate of presidential electors, but only if a court had done that. So if you actually look under the hood of what was going on in Florida in 2000, they weren’t coming in and saying, “We have the power after election day to override our own voters.” State does not have that power because of the federal statute I mentioned.

Jim: Right. That’s good. I just learned something new there. That sounds pretty damn solid. Now we’re getting out of the kind of hard machinery aspects and more to, is it good. Could a national popular vote result in more polarization? Suppose today the Electoral College system requires people to compete in some of the more closely balanced states to win, you can’t go too far afield and win in Ohio or Wisconsin or Florida or Pennsylvania. If we went to a national popular vote system like John is suggesting, might not the result be for the campaigns to try to really run up the margins in the most polarized states, the California’s, the New York’s, the Texases, et cetera, rather than having a campaign that’s aiming more towards the centrists states?

John: Well, of course the way candidates actually campaign inside the battleground state is simply the same way they would campaign nationally. So when you’re running for president inside Ohio, which is a closely divided battleground state, of course you’ll try to run up the vote in the parts of Ohio where your party is strongest. That goes without saying. So that process already exists. You’ll milk the areas where your party is strong, but remember both parties are doing that simultaneously, and you go after the votes in the middle in Ohio, which are available to your party.

John: Now, a related question is, would candidates campaign only in the big cities, for example, which is one we hear very often? So we’ve looked at how candidates actually campaign, and the way they actually campaign now inside a battleground state… Remember, inside a battleground state, every vote is equal, candidate with the most votes wins. That’s exactly the formula of the national popular vote, every vote is equal, candidate with the most votes wins. Inside Ohio, 22% of the state is outside the standard metropolitan statistical areas, the so-called rural areas, and they get 25% of the visits. So for all practical purposes, they get visits in proportion to their population.

John: The big cities, the Cleveland, Cincinnaties, Columbus, individually and as a group, they get exactly the percentage that their population is, and the metro areas get the same percentage, and the mid-sized cities, of which they’re seven in Ohio, the Youngstowns and Daytons et cetera, that it also attracts population. Why is that true? Because if you’re running a presidential campaign in a battleground state, you’ve get the most knowledgeable professional advisers in the world telling you that if every vote is equal, every vote is equal, and it would be insanity to campaign otherwise. We’ve gone through and mapped the appearances of candidates in Florida and Ohio and of the dozen or so battleground states, and the candidates do not hover around Miami or Philadelphia or Cleveland or Denver or Des Moines or Las Vegas, they campaign all over the state because that is the way you have to campaign if you’re going to win. So a lot of this rhetoric about big cities controlling elections or polarization is really just that. When you actually look at candidates campaign, they are very rational. And if they’re rational, they have to follow the voters.

Jim: Great answer. That sound like you guys have done the work and you have the data that shows that people do act rationally and apply their level of effort to where the voters are, no doubt, seasoned by their estimate of how many voters they have in an area. They’re not going to spend a lot of resources in an area that’s 90 percent for the other guy. That sounds good. Sounds like you thought these up. What are some other objections that are out there that you have answers for? Just give me two more objections that you regularly hear that you feel like you’ve got good answers for.

John: What we think we have good answers to are 131 that we cover in our book. We actually cataloged all of the things that different people have said, and we think we’ve got a good answer for all of them. But one of the ones that comes up quite often is… I call it federalism. It’s somehow an uneasy feeling that somehow this is going to upset the federal system. You really have to look at it though. We do have a good system, we divide power between states and the national government. That not only allows for innovation at the state level of experimentation, but it also prevents one party from ever getting a foothold.

John: For example, suppose the Constitution said that the state legislatures didn’t control how the president was elected but Congress did. That’d a fairly natural way of writing a constitution. In fact, that is the way it’s done in many countries, because there is no federal system with states with meaningful amounts of power. Then you’d have an election system where the party in power would write the election rules, and anybody in power can think of ways of changing the election rules to keep themselves in power. So you have a dispersal of power.

John: Well, there’s nothing about how you elect presidential electors, whether you elect them by congressional district or state level or national, that in any way increases or decreases the amount of power that state government has or the national government has. So an argument we often hear but it reflects sort of a natural concern, a legitimate concern but for which there’s a good argument is, this is not something that’s changing the structural amount of power that different levels of government have, it’s a way of changing the way you’re counting votes. We get a lot of that, and I think that’s not really very convincing.

Jim: Okay. You said you had one other one.

John: Well, we have 131 other ones.

Jim: All right. Let me tell you, people, actually, I just clicked on that, nationalpopularvote.com/answering-myths. Oh my God, do they have them all I answered. As you heard John, he can slay these objections. Wow. He blew me out of the water. So if you think you have an objection to the National Popular Vote and interstate compact, go to there, and we’ll put that link up on our Web site for the episode and go see what he says. I suspect you might be well spanked. These guys have obviously done their work.

John: Whenever anybody proposes change, it’s perfectly [inaudible 01:15:09] natural to try to think of every possible objection or thing that can go wrong, and it’s perfectly normal that there be 131 objections. If you’re proposing a change, you need to have a good answer for all or virtually all, we think we have good answers for all of the objection, and you have to make the case. The burden of proof is on the people who want to change things.

Jim: I love it. That’s what I’d call a good faith way to argue. You basically let people bring forth the objections and you knock them down one after the other. I wish more of our political process worked that way. Well, John, I want to thank you for a most interesting interview. You have had just the most amazing set of professional experiences, inventing a significant cultural phenomenon, the scratch-off lottery ticket, invented one of the more important intellectual contributions in computer science and genetic programming, and now doing strong work in changing how we elect our president in a very clever way, and one that seems sound without amending the Constitution. I got to say, you’re an inspiration.

John: Well, thank you, Jim, for inviting me to be here today.

Jim: Yeah. It’s been wonderful. I’ve learned a thing or two. With that, we’ll wrap her up.

Production Services and audio editing by Jared Janes Consulting, music by Tom Mueller at modernspacemusic.com.