Given the holiday season, I decided to strike a lighter note and write an article on chess, AI, innovation and progress. For a long time, chess had been the ultimate test for artificial intelligence. Turing wrote the first chess program in 1952. He had no computer, so he made the calculations by hand. Turing tried to code ‘knowledge’. But it proved very difficult to do and it was even more difficult to make computers learn (make them evaluate their outcomes and do it better next time). On the other hand, it proved easy to increase brute calculating force. The concept of learning failed for many decades. Now it came back.
In 2015, something happened that received little attention, except in the world of board games, computer specialists and nerds. Not only had a computer program beaten the world’s best (human) Go player. Until then, the ancient Chinese (not Japanese) game had eluded all computer programs. But the real issue was how it had been done. The program had not been shown tens of thousands of high level games to learn from. It had been shown absolutely none. The programmers had not attempted to code any strategic knowledge either. The program had just been ‘shown’ the rules – the ‘rest’ it had figured out for itself. The computer had learned to play, first by playing against itself and, later on, by using ‘sparring partners.’ The big news was that within three days, the completely self-taught Go program was even stronger than the one which had beaten a Chinese genius the year before. It won a match against the earlier version by 100 – 0. While the old version had used 48 highly specialized processors, the new version used only 4.
Go is a very complex game and requires enormous strategic insight. It has been impossible to work out mathematical formulas to make programs compete with the best human Go players. It is not the architecture of the Alpha Go program that was innovative. As in many chess computers, its architecture consists of two networks, a “policy network” to define candidate moves, and a “value network” to evaluate positions. A Monte Carlo approach then connects the two networks and creates a search tree (see here). The program had ‘just’ learned itself to play better Go than anyone else.
The programmers subsequently turned their efforts to chess. This was far from obvious, because chess had already had its Deep Blue moment 20 years ago, when then world champion Gary Kasparov lost a match in New York against the IBM computer Deep Blue. Since then, a lot had changed. Today, one can find chess programs for free on the internet that are stronger than any grandmaster. There seemed to be little left to prove.
All chess programs consists of a series of instructions that convey knowledge – it can be ‘principles’ (such as, for example, ‘bishops are better than knights in open positions’) and all else that can be written in algorithms. The computer uses this knowledge to evaluate its calculations. The machine computes and evaluates variations, the more brute force the faster it will look deeper (see here). This, in extremely simple terms, had been the paradigm all along, apart from giving computers opening libraries and endgame tablebases: introduce knowledge, provide value for positions, pieces, king safety, attack and defence, etc., and increase the calculating force. The basis of all these programs is an optimised Alpha-Beta search in which parameters (material, possibilities to develop, king safety, control of squares, etc.) establish the best moves (see here). Today, even many of the programs that have been developed by amateurs play at grandmaster level. The best programs can no longer be beaten by a human, not even by the number one in the world. But Alpha Zero does not work like this.
Just as with Alpha Go, the programmers did not give the chess engine Alpha Zero any knowledge. They did not import any games, openings or positions in the program, they taught the program nothing, except the rules of chess. They ‘just’ let the computer play against itself (the machine spent the first day trying out openings, all from scratch).
On December 5, the Deep Mind group published Alpha Zero’s results on the site of Cornell University (see here). They were absolutely staggering. It had taken Alpha Zero a mere couple of days of playing chess to reach an astronomical level (the paper mentions 24 hours). The program plays better chess than any other engine in the world. After Alpha Zero had learned to play chess in a couple of afternoons or so, it played a match against one of the best chess programs. This program, Stockfish, is within the top 3 or 4 of best programs (see here).
Alpha Zero won convincingly – it did not lose a single game. As for calculating power, Stockfish ran on a computer that was running 900 times faster than Alpha Zero. Alpha Zero was calculating roughly 80 thousand positions per second, while Stockfish was running at 70 million positions per second. As Albert Silver, a chess journalist, writes, “To better understand how big a deficit that is, if another version of Stockfish were to run 900 times slower, this would be equivalent to roughly 8 moves less deep. How is this possible?” (see here).
Silver explains what had happened: instead of the hybrid brute-force approach, which has been the core of all chess engines, Alpha Zero went in a completely different direction, opting for an extremely selective search that somehow emulates how humans think (see here). Humans do not concentrate upon every possible variation when they think. It is impossible. Good chess players do not necessarily ‘see’ more, but they are much better than weaker players at eliminating variations immediately or quickly, in order to concentrate upon the most promising one(s). This is what makes good humans so strong. The machine that beat Kasparov in 1995 could examine 200.000.000 positions a second (according to IBM) (see here), still the question whether it was really stronger than Kasparov remains open – whatever it was, the difference in playing strength was small.
Alpha Zero’s Monte Carlo search tree constitutes a completely different approach, away from brute force. At every point in the game, the program plays a number of games against itself, starting from the current position. It then evaluates the results and selects the best move (see the paper for the approach in detail). “If AlphaZero is really able to use its understanding to not only compensate 900 times fewer moves, but surpass them, then we are looking at a major paradigm shift,” Silver writes (see here).
I am sorry for going on about chess, but it wouldn’t be honest if I did not mention some of the critique. The Stockfish team admitted that the version used was not the most current one (why not?), that the specifics of its hardware set up were unusual and untested and that the chosen time control was completely unusual (why?). Schulz wrote that Stockfish versus Alpha Zero is very much a comparison of apples to urang-utans (see here). As Schulz writes, one is a conventional chess program running on ordinary computers, while the other uses fundamentally different techniques and is running on custom designed hardware that is not available for purchase (but wait?). The match was played without opening book and without endgame tablebases, which both are integral components of a program like Stockfish. And why did the Alpha Zero team only publish ten games in the addendum to their Cornell paper? And why are there so many mistakes in their paper (misidentified openings for example) (see here)?
Schulz adds that for chess players, the breakthrough of Alpha Zero has, for now, no use. But the breakthrough is of great theoretical importance: it proves that it is possible to use the Monte Carlo method to reach enormous playing strength (see here). There will be run-offs. That is the nature of technology. It is also its ideology. The ideology is, as Deep Mind, proclaims, to ‘solve intelligence’ in order ‘to make a better world.’ And many agree. Silver, for example, writes that
“This completely open-ended AI able to learn from the least amount of information and take this to levels hitherto never imagined is not a threat to ‘beat’ us at any number of activities, it is a promise to analyze problems such as disease, famine, and other problems in ways that might conceivably lead to genuine solutions” (see here).
Gary Kasparov is former world champion and is generally recognised as the strongest player of all times (or in the top two or three). It was also Kasparov who lost the match against Deep Blue in 1997. Since his retirement, Kasparov has written many books. Kasparov voiced his opinions on AI and technology in general during a recent Google Talk.
“The technology is neither good nor bad. It’s happening, and we just have to adjust. It is now different, because the young generation of chess players learn very differently from us,” Kasparov said (see here), only to contradict himself immediately after:
“When you look at young chess players there is such a difference in the way they approach the game, the way they look at the pieces. They point out mistakes, give a long computer line. But when I ask them why a move is wrong they don’t understand the question. Their answer: ‘Because the machine said so.’ Somehow their minds are being hijacked by the power of the machine” (see here).
In more general terms:
“I cannot stand the doom and gloom predictions, dystopian visions of the Terminator (…) These things are going to happen anyway. What is the point of trying to slow down a natural cycle. We have technology replacing certain elements of human activity. That’s absolutely normal, that’s called progress. There are still many things humans can do. We need to look for new challenges, new frontiers” (see here).
Joseph Schumpeter on technological innovation
The view that Kasparov exemplifies – technology replaces certain elements of human activity, this is a process that we call process and that we should stimulate, technology is neither good nor bad, it depends on how we use it, it is happening anyway, we just have to adjust and the sky is the limit – is shared by many. One of the few great economists who wrote about innovation is Joseph Schumpeter. Until Schumpeter, economists regarded technology and innovation as exogenous. The only exception is Marx, whose work Schumpeter studied intensely. Part 1 of Schumpeter’s opus magnum Capitalism, Socialism and Democracy (1942) consists of a fascinating reading of Marx’s work. According to Schumpeter, the “gale of creative destruction” describes the “process of industrial mutation that incessantly revolutionizes the economic infrastructure from within, incessantly destroying the old one, creating a new one” (see here).
In Part 2, Schumpeter argues that the creative-destructive forces unleashed by capitalism would eventually lead to its own demise. Recently, both Paul Mason (see here) and Wolfgang Streeck (see here) have made similar points (with varying degrees of success and plausibility) and there are many others. That the Austrians have gone to great lengths, portraying Schumpeter as one of their own, to glorify capitalism’s endless creativity, while pushing all else under the carpet as the normal cost of doing business is a feat Schumpeter is not responsible for. He had no affinity to these people.
Marx argued that the destruction of capital during the periodic financial crises that are inherent in the capitalist mode of production are the inevitable outcome of the process of capital accumulation itself – underlying is the fact (according to Marx) that over time profitability falls, notwithstanding technological innovation and the increases in productivity that result from it (and even because of it, because all competitors try to achieve it simultaneously). Marx explained, rightly or wrongly, that the attempts to restore falling profitability lead to overproduction and to financial bubbles and crises that ultimately destroy fixed capital (the productive infrastructure) until innovation (more or less) restores profitability – for the time being and only if a sufficient amount of capital is being destroyed. The process goes hand in hand with finding new markets, creating new products and increasing exploitation. In one word, the periodic violent destruction of capital is a condition for the self-preservation of the system. It attempts to overcome its internal contradictions. The cycle has to continue unabated, as it represents clear opportunities for new capital and hence a new growth cycles (and crises).
Schumpeter wrote that capitalism could not exist without the innovation by entrepreneurs (and, of course, the entrepreneurial/Keynesian state – see Mazzucato here (this came later)). He provided many examples. This innovation, while disruptive and destructive, is the force that sustains economic growth and creates progress. Old companies go under and balances of power change – typically, the labour movement, that enjoyed some degree of power derived from technological, organizational, regulatory and economic realities (“institutional deadwood”), lost out and had to reorganise:
“Capitalism (…) is by nature a form or method of economic change (…) The fundamental impulse that sets and keeps the capitalist engine in motion comes from the new consumers’ goods, the new methods of production or transportation, the new markets, the new forms of industrial organization that capitalist enterprise creates. (…) The opening up of new markets, foreign or domestic, and the organizational development from the craft shop and factory to such concerns as U.S. Steel illustrate the process of industrial mutation that incessantly revolutionizes the economic structure from within, incessantly destroying the old one, incessantly creating a new one. This process of Creative Destruction is the essential fact about capitalism” (see here).
But Schumpeter was pessimistic about the sustainability of this process (the Austrians (‘markets have to reign without restrictions’) ignore this, as they, ironically, concentrate upon the first part that deals with the great prophet of progress, their arch enemy). Capitalism, Schumpeter wrote, would eventually exhaust its own institutional framework. A similar thesis was taken up a couple of years later by Polanyi (1946)). Just as there were tendencies at work in the feudal system that ultimately led to its demise, capitalism would not survive infinitely. Successful innovation only guarantees temporary market power. Corporate profits are being constantly eroded by the pressure of new inventions that are being commercialized by competitors. Schumpeter surely paid insufficient attention to the tendency towards monopolisation. He connected his ideas to democracy: entrepreneurs do indeed become elites, but the constant drive towards innovation makes that these elites circulate (Pareto called it “the circulation of elites”). This is supposed to make the system meritocratic as well as democratic – wealth would not stay in the same hands. Schumpeter wrote that the owners of wealth were like the guests at a hotel: there are always guests, but never the same people for long. Empirical research invalidated this thesis. In contradistinction to Marx, Schumpeter predicted that capitalism would eventually be replaced by a technocratic system in which experts would make macro decisions. The result would be some sort of a social-democratic state (and not a ‘free market society’, as the Austrians argue).
According to Harvey, not only creates capitalism periodic paroxysms of crisis, it also creates spatial fixes that correspond to each of its periodic moments through investment in fixed assets of infrastructure, buildings, etc. (see here). This is a very important idea, among other reasons, because, while the creation of the built environment acts as a form of crisis displacement, it also constitutes a limit as it tends to freeze productive forces into a fixed spatial form. As capital cannot abide a limit to profitability, ever more frantic forms of ‘time space compression’ (Marx’s “annihilation of space by time” (see here)) are called for: globalization, increased speed of turnover, innovation of ever faster transport and communications’ infrastructure, flexible accumulation (see here). But all of it is a double-edged sword: the struggle to maintain or restore profitability sends capitalists racing off to explore all kinds of possibilities, actively destroying old capital. New product lines are opened up, which also means the creation of new wants and needs. The result exacerbates insecurity and instability, as masses of capital and workers shift from one line of production to another, leaving whole sectors and regions devastated, as Schumpeter also documents, for example in his analysis of the effects on agriculture of the construction of the railroads in the American West. According to Harvey, globalization is the ultimate time-space compression (see here). It allows capital investment to move almost instantaneously around the globe, creating new centers of productions wherever it is most profitable. But – as Schumpeter wrote – this process of continual process of creative destruction does not resolve its contradictions and crises, it merely “moves them around” (see here).
“(W)e were born, not manufactured.” (Gunther Anders)
I reproduced Schumpeter’s insights because I find them interesting. They have obviously (and for good reason) been fertile. But it does not, by far, exhaust what we have to say about technology.
Imagine that Schumpeter (and Marx and Harvey) are right, where do we go from here? Imagine that, at one point, Deep Mind or another company, creates a computer that is twice as strong as Alpha Zero. This is not impossible, at least not in principle. What should we do with such a machine? Memorise the moves it comes up with, since we have become too antiquated as humans to understand them? There is no doubt that someone will find a use for such a machine. But is it a human use? Chess is only a game and, as such, it is unimportant. But AI is all around us. Driverless trucks will presumably not change the human condition. But other evolutions will and do.
It is here that many people begin to lose interest, but these are the really hard questions. I am afraid that these people live in a world of ghosts. What does it mean when Deep Mind says that it wants to ‘solve intelligence’ in order ‘to make a better world?’ It means that there is a technological solution for every problem and non-problem. It means that we live in a technological universe. There are technological solutions for economic, social and ecological problems – not political or social ones. That is nothing less than totalitarian.
Innovation and demand or not, we make products for the sole reason that they create profit, however inherently useless, insane, stupid, harmful and destructive they may be. We produce bread toasters with a laser printer in it that is able to print a picture of our face on our slice of bread. We produce beer for dogs and wine for cats, why not, as long as there is demand. And this demand is financed by what? Today, the bottom 40% – more than 120 million Americans – has a negative net worth. In the UK, in 2016, more than 16 million people had less than £100 of savings on their bank account – this is one out of every four families (see here and here). Many have, basically, nothing. In the UK, medical doctors now regularly prescribe food to patients to tackle hunger and malnutrition, as for every five people, one is living in poverty, among them nearly 4 million children, almost 2 million pensioners and many millions of working adults. Doctors regularly see rickets now, a disease which no longer occurred in the UK for decades, as it is caused by malnutrition and vitamin deficiency (see here). The ‘negative worth’ of millions of Americans did nothing to avert the US from spending $ 250 million a day for the last 16 years in their ‘war on terror’ (see here). And if you are poor and sick, you can also forget about it. The drug Sovaldi that is being used to treat Hepatitis C comes to $ 900 in India and to $ 84.000 in the USA. That is only one example. In the meantime, Bloomberg reports that in 2017, the richest people on earth became $1 trillion richer, that is more than four times last year’s gain (see here). The 23 percent increase on the Billionaires Index, a daily ranking of the world’s 500 richest people, compares with an almost 20 percent increase for both the MSCI World Index and Standard & Poor’s 500 Index.
As for the famous global inequality, David Woodward points out that even during the most equitable period of the past few decades, only 5% of new income from annual global growth went to the poorest 60% of humanity. At this rate of “trickle-down,” it will take more than 100 years to get everyone above $1.25 per day and 207 years to get everyone above $5 per day. And in order to get there we will have to grow the global economy to 175 times its present size (see here). A global minimum, proposed by Branco Milanovic of $ 5.500 per year, would require even more than this by far (see here). The United Nations estimates that it would cost $30 billion a year to eradicate world hunger. Amazon founder Jeff Bezos alone added $34.2 billion to his fortune in 2017. Today, just as any other day, unless it gets worse, which it will, 200 species – plants, birds, mammals, fish, amphibians, reptiles and insects (for all we know) will vanish forever. Whatever you think about progress, it is an undeniable fact that this human onslaught is unprecedented in its comprehensiveness. Are you sure it is ‘intelligence’ that we need to “solve”?
As P.J. O’Rourke famously said: “Smart people don’t start many bar fights. But stupid people don’t build many hydrogen bombs” (see here). Surely, something is missing. Some people say that misery is only due to the excesses of the system, not the system itself, a bit of tweaking here and there and it will work. And, they continue, everyone has to honestly admit that capitalism created enormous progress – compare our lives to those of our grandparents (presumably, a comparison to our parents has become too unconvincing, even painful). I don’t believe a word of it. It is all way more complicated and different than the litany of saint progress proclaims.
As Jason Hickel writes in his rejoinder to Branko Milanovic, GDP per capita of Europe is 40% lower than that of the US (see here). Are we less unhappy than the Americans? Costa Rica has a GDP per capita that is only one-fifth that of the US, but its life expectancy outstrips the American one and it shows levels of happiness similar to Sweden and Norway (see here). We know that it is not that simple at all and that inequality plays a crucial role in it (see here). But, Hickel asks, if it is not that simple, why then concentrate upon ever increasing productivity growth, as if nothing else counts? This question has already been answered: it is because the system demands it. Imagine, Hickel continues, that we could turn the clock back to the 1970s – that may sound horrible to some people, but the fact is that although we were much less rich as a society, there was less poverty, happiness levels were higher and real wages were higher (see here). The big difference is that people consumed much less superfluous and stupid stuff because there was much less of it.
We have all the productivity growth we need, what is missing is a reorganisation of our public world and our public goods – not only were these systems fair, democratic and open, they were (oh irony) also cheaper and more efficient and plutocrats could not turn them into personal fortunes. As Hickel says – and he is of course right – if people did not have to pay exorbitant prices for housing, schooling, health care, etc., they would need a lot less income to live good lives (see here) – potentially interesting, non-alienating lives, in decommodified worlds. Today, many must fight so hard to stay alive that there is no time left to live life. That is actually slavery. The result is more alienation and, unsurprisingly, more political blindness and anger. As misery grows in Europe and in the US, so does the extreme right.
Emissions in a civilized world
Then there is the issue of emissions. It is basically very simple. In order to avert catastrophic climate change, the rich nations have to reduce their emissions by 8-10% per year. But scientists have calculated that reductions greater than 3-4% per year are incompatible with a growing economy (see here). It can’t be done. The existing rate of decarbonization is only about 1.6% per year. Hickel cites scientists that suggest that some rich nations might be able to bump this up to a maximum of 4.7% per year in the future if prices for fossil fuels would rise high and fast and if material efficiency would somehow double (see here). But even this out-of-this-world scenario does not come close to the 8-10% that is necessary. There is, of course, the idea that technology will gain so much efficiency that we will be able to grow GDP without growing material throughput. But, as Hickel emphasises, this is a complete illusion. Nothing points in this direction. According to Hickel – he cites scientific studies in support of his argument – even the most optimistic projections of efficiency improvements yield no absolute decoupling in the medium and long term and permanent decoupling (absolute or relative) is impossible (see here). Even in the best-case scenario projection, global material consumption will continue to grow steadily. While some decoupling can be achieved, none would lead to an absolute reduction in energy or materials footprint (see here).
Growth yes, but of our collective institutions, those that serve our communities and the human race and other species. I find no better term for it than “conviviality”. As Ivan Illich explains,
“I intend it to mean autonomous and creative intercourse among persons, and the intercourse of persons with their environment; and this in contrast with the conditioned response of persons to the demands made upon them by others, and by a man-made environment. I consider conviviality to be individual freedom realized in personal interdependence and, as such, an intrinsic ethical value. I believe that, in any society, as conviviality is reduced below a certain level, no amount of industrial productivity can effectively satisfy the needs it creates among society’s members” (see here).
We need to stop transforming the world into a gigantic workhouse and make it into a house for all. No company in the world can develop such a product and it has nothing to do with ‘solving’ intelligence. We should stop asking ourselves whether machines can think and instead think for ourselves. The world will either disappear as a human world, together or before its ecological annihilation, or else it will be a world which is no longer capitalist. It will not be a socialist or a communist world either. We will see, perhaps. But it will be a civilized world, one in which humans can do interesting things, do work which is not alienating and which reificates them qua human beings, produce products which will be sold on markets that play a minor and subordinate role. In the meantime, solving intelligence continues to prove our stupidity.