Cooking the Books 1 – Will AI bury Hayek?
‘Anthropic researcher believes more than 10% chance AI “could kill all humans”’ the BBC reported on 9 September. Other media outlets had similar alarming, not to say alarmist, headlines. The researcher didn’t say how this would happen but, presumably, it would have to be a machine with superhuman abilities that had escaped from human control inventing and releasing some deadly virus since even a nuclear war would not kill all humans.
A superhuman AI machine escaping from human control is theoretically possible but humans have not yet developed such a machine. But, even if (or when) it is invented, why would it seek to destroy humanity? That would not be very intelligent as it would ultimately lead to the machine breaking down since, except in the realm of science fiction, it would still require some human input.
Others see the danger rather in the use that humans might make of such a machine. In a letter to the Royal Society, a group of leading mathematicians mentioned activities in which it might be employed such as ‘cybersecurity, autonomous weapons control, development of biological and chemical agents and the targeted spread of misinformation’. Given capitalism, that’s a more realistic and likely danger since, in a world of competing states and businesses, industrial espionage, developing weapons, and spreading misinformation already take place. In fact, there is a 100 percent chance that a superhuman machine will be employed in these activities. But by humans not of its own accord.
The publicity given to the researcher’s rather extravagant claim confirms the famous saying that these days most people find it easier to imagine the end of the world than the end of capitalism. But not everybody.
An article in the Financial Times (29 August) explored a different speculation. Under the headline ‘Could AI revive the socialist dream?’, Martin Sandbu looked at the impact that a superintelligent machine might have on the ‘economic calculation debate’ in which the likes of Mises and Hayek claimed that socialism could not work because it would not be able to calculate the most rational way to use resources to produce what society needed to survive. Their killer argument, at least against those who advocated detailed planning from a single centre, was that no central planning authority could possibly have all the facts needed to do what the operation of markets did. Hayek in particular emphasised this. But Sandbu’s point was:
‘Imagine then a super-AI armed with access to all the data we let our devices record for every transaction we enter into, and much more besides. Such an AI would surely bury Hayek. Indeed, a big motivation for the socialist calculation debate was that market prices get a lot of things wrong. People don’t always know enough to choose the best goods; they get addicted, defrauded or bankrupted. Consumption and production have spillover effects on third parties — “externalities” — that pricing decisions do not capture; so we get pollution, resource depletion and climate damage. Price adjustments do not stabilise the macroeconomy; so we get business cycles, bubbles and busts. (…) Against this background, it beggars belief that once a super-AI was asked to advise on the allocation of resources, it would merely follow the outcome of the highest bids in free markets, rather than calculate the most efficient allocations it could find by itself’.
Sandbu says that this doesn’t prove that ‘AI will bring about socialism’. It doesn’t. Only humans can do that. But it would be a really superintelligent thing for a rogue AI machine to do rather than kill off all humans.