The Bear’s Lair: The AI revival of Thomas Mun
Thomas Mun (1571-1641) wrote the mercantilist classic “England’s Treasure by Forraign Trade” arguing that the country should run a trade surplus and thereby build up “treasure.” It was derided by classical and Keynesian economists alike. However, almost 400 years after its writing, it may be coming back into fashion in the world of Artificial Intelligence, with the “treasure” concerned being what one might call “Robo-IQ.”
Mun’s economic approach was universally despised by the classical economists from Adam Smith onwards, who pointed out that not every country could run a trade surplus simultaneously and that the bounties and restrictions necessary to attempt it caused trade blockages that reduced overall wealth. Today, economists are disdainful about paying attention to a country’s balance of payments at all, describing President Trump as an economic illiterate for doing so.
However, when looked at in light of the economic conditions prevailing when he wrote it in the 1620s (it was only published after his death in 1664) Mun’s masterpiece looks prescient. Early modern states like James I’s Britain faced a permanent shortfall of cash. Inflation due to the influx of gold and silver from Latin America had wiped out the value of their traditional levies such as feudal dues, while they had nothing like the bureaucratic efficiency needed to impose an income tax, a product only of the 1790s.
In Mun’s time there were no real capital markets, partly because each debt obligation was incurred by the King individually, so was difficult to recover after his death. At the same time, warfare was becoming more expensive. Hence “treasure” was indeed the principal need of the 1620s monarch; it was far more important than trade as such. A century earlier, before the 16th Century inflation, trade had been prioritized by Henry VII, but economic conditions back then had been easier and monarchical liquidity much greater.
British rulers followed Mun’s prescriptions to a large extent during the 17th Century, with beneficial results. They developed colonies in North America and the Caribbean, producing addictive products, tobacco and sugar, which could be subjected to a heavy excise before being sold not only in Britain but also in Continental Europe to buyers without such colonies. They also developed East Indian trade links for tea and coffee, which had similar advantages. Add to that the invention of “Dutch finance” – the Bank of England was founded in 1694 – and by the early 18th Century Mun’s policies were no longer necessary, since tax revenue was generally sufficient, and the now institutionalized state had the ability to borrow vast sums.
In the eighteenth and nineteenth centuries, Britain could profitably move towards free trade, as it did first under Walpole and later under Pitt, Liverpool and Peel. But without Mun’s policies for 100 years, Britain would have remained impoverished and relatively backward.
In the field of artificial intelligence, we are at present in a very similar position with respect to AI as Mun’s King James I was with respect to money. The supply of AI is extremely limited, partly because the supply of people who can design AI is extremely limited, and we are probably quite some time from AI systems that can design and build further such systems. A few of the world’s largest companies, all of them U.S. or Chinese, notably Google, Apple, Amazon, Facebook, Baidu, Tencent, and Alibaba, have an oligopolistic control over AI, and there is a severe shortage of AI outside that oligopoly. You can align this with James I’s Britain, desperately short of specie, and facing one immensely rich supplier, Spain, which controlled the New World mines for it and through its mining activities drove up prices and drove down tax yields.
In such an environment, Mun’s prescription for success may again be applicable. Two lengthy articles have recently appeared discussing this: one, by the tech investor Ian Hogarth, outlining areas in which “AI nationalism” is taking hold, the other, by the great Henry Kissinger in the Atlantic, suggesting that the rise of AI will bring the end of the Enlightenment. Both articles work on the presumption that AI will be an overwhelmingly important technology once it is fully deployed; my usual skepticism causes me to question that, just faintly, but not to attempt to contradict it.
Hogarth’s leading example of a transaction that would have been blocked on mercantilist principles was the acquisition by Google of the British AI company DeepMind Technologies for $500 million. DeepMind, which developed the successful Go program AlphaGo, which learned from human Go players and its successor AlphaZero, which mastered Go solely by self teaching, was Britain’s leading AI company, and its absorption into the Google behemoth has left Britain without a significant independent AI presence.
Hogarth’s solution to the increasing importance and scarcity of AI is to make AI a public good, with generous amounts of government-funded research to boost the capabilities of countries like Britain that have only modest ones. That seems to me like a recipe for failure; it is also not the solution Mun would have chosen, knowing as he did the incapacity and corruption of early 17th Century governments. Kissinger also proposes a Presidential Commission of eminent thinkers to work out the philosophical problems caused by AI – again this would appear a recipe for massive new regulation and very little progress.
Mun’s solution would be to recognize the extreme scarcity of the critical AI resource, which one might call “Robo-IQ.” Robo-IQ is not the mere computer power available, nor the volume or quality of software that has been written for it, but a combination of the two, recognizing the artificial intelligence capability of a specific country. Increasing Robo-IQ is very difficult, because there are so few human specialists capable of creating it. Only in the very long term will critical masses of Robo-IQ become self-perpetuating; with the machines themselves being capable of creating better machines, with more Robo-IQ and higher artificial intelligence capabilities.
Mun’s policy for artificial intelligence is thus clear. Each independent country must seek to maximize its stock of Robo-IQ, doing deals that increase it, and blocking deals that allow it to fall into foreign control. The Google takeover of DeepMind would thus not have been permitted, because it reduces the independent British stock of Robo-IQ, which any good Mun-trained mercantilist knows to be undesirable. Policies would also be implemented to encourage creators of Robo-IQ to settle in Britain and deploy their skills there; British universities with active departments that trained creators of Robo-IQ would be encouraged to offer scholarships to the best foreign students, thereby increasing the country’s ability to develop Robo-IQ in the future.
For the United States and China, which today dominate the global stock of Robo-IQ, Mun-mercantilist policies would be somewhat different. Apart from putting up barriers to looting Robo-IQ by the other major Robo-IQ power, these countries should put in place policies that prevent Robo-IQ from being squandered and encourage its development by as many independent sources as possible.
They should thus regard the global oligopolists of Robo-IQ with deep suspicion, knowing that their corporate incentive is to buy up and suppress smaller developers of Robo-IQ and to monopolize the Robo-IQ creative talent. The oligopolists’ attempts to impose constraints on Robo-IQ development should also be resisted. Their false morality and politics will lead to blockages of important advances much like they do in genetic engineering; and mis-appropriation of funds to extreme leftist causes, such as Google’s creation of Jigsaw to combat “toxic” ideas. These constraints will slow the development of new Robo-IQ and risk losing the current national stock of it to foreign competitors.
The Venture Capitalists who have funded the oligopolists are also wedded to the idea that data is king when it comes to AI. This is the only conceivable reason Uber is worth so much — it has ride data from which it can extrapolate self-driving AI. In the minds of its VC stockholders Google is similarly highly valued because it has all the data from its constantly patrolling Street View cars.
Big Data almost certainly has diminishing returns which we’re now hitting. Data regulations across the world (the EU and China) are making it hard to gather and hold on to large amounts of data. This will further benefit the oligopolies who can get past the initial bureaucratic hurdles; however, it will also make Big Data a bigger liability.
Governments have a natural predisposition towards data projects: they create large behemoth crony bureaucracies that can even be used to help win elections. However, in the hunt for Robo-IQ the most efficient players will win, not the ones with the most data to throw at a problem. Thus, data should not be given much weight in the Robo-IQ search. After all, our brain does not work by throwing massive amounts of data at it and sorting it all; it can ignore most data and pick out which bits not to ignore. That ability, not crude information collection, is the key to accumulating Robo-IQ “treasure.”
The natural impulse of governments faced with the scarce strategic resource of Robo-IQ will either be to attempt to control it themselves, creating a state-Google that monopolizes Robo-IQ and kills innovation altogether, or to develop Robo-IQ in cahoots with the current oligopolists, imposing regulations as demanded by the oligopolists and preventing smaller competitors from flourishing.
Thomas Mun would tell you that this is precisely the wrong approach to maximizing the stock of Robo-IQ “treasure.” After all, while James I’s state chartered the initial colonizing companies in the New World, it did not go into the colonizing business itself. The development of tobacco and sugar plantations was entirely carried out on a for-profit basis by the private sector, normally the small and medium scale private sector. The same policies should apply to the maximization of Robo-IQ: breaking up the oligopolies rather than encouraging them while nurturing the resources needed to develop new Robo-IQ producers and encouraging the small-scale private sector to do its job.
Eventually, Robo-IQ will become plentiful, probably around the time AI machines are able to develop future AI for themselves rather than relying on human developers. Once that happens, Adam Smith’s maxims will once again apply, and the correct policy will become completely free trade with as few barriers and as little regulation as possible, and international deals being freely transacted. But until then, the mercantilism of Thomas Mun will be appropriate, focused on developing the nation’s stock of Robo-IQ. Applying those principles through state control and oligopolies will however work as badly as it did in the 17th Century.
(The Bear’s Lair is a weekly column that is intended to appear each Monday, an appropriately gloomy day of the week. Its rationale is that the proportion of “sell” recommendations put out by Wall Street houses remains far below that of “buy” recommendations. Accordingly, investors have an excess of positive information and very little negative information. The column thus takes the ursine view of life and the market, in the hope that it may be usefully different from what investors see elsewhere.)