The American financial authorities are marking Artificial Intelligence(AI) as a possible risk to the financial system. This seems correct -- new ways of doing banking often do seem to come with the risk that someone will misuse the new ways. For there's really only one major problem in the financial world; one core difficulty. Solving that does indeed open up new opportunities, but new ways of thinking you've solved it, without actually having done so, lead to people going bust in spectacular fashion.
This also leads on to an interesting point about AI. Only if it is actually biased, in the ways the activists say it must not be, will it be useful in trying to solve that central banking and financial market question.
That major, central, problem was identified by Adam Smith. It is that the difficulty is in finding those people worth lending money to, investing in. New ways of shovelling the money out the door in loans or equities aren't difficult -- new people worth sending the cash to are the difficulty. As Smith pointed out, the person willing to pay a high interest rate is precisely the person unlikely to pay the money back. For his plans and ambitions might not be wholly grounded in reality.
So, if the use of AI is going to be about how loans get made, the nuts and bolts of the process, then there's a substantial risk attached. For people might well think they've got some new method which really works until they find out that they've been sending all the money to people who really shouldn't be getting it. And then go bust with that resounding noise of a financial institution imploding. This did, after all, happen with mortgages only 20 years ago. New types of mortgage were devised in the American market, many of them issued – but it turned out that the shortage was not of types of mortgage, but of people who should have a mortgage.
But let us assume, for the sake of argument, that AI will be used to uncover those who are indeed worth lending to but which current systems don't identify. Also think of what the activists are insistent about how AI should work: AI should not include any of the biases of our current society. It must be race, gender, sexuality, and so on blind that is. But more than that, the insistence is also that any difference in outcome across race, gender or sexuality must be a result of that bias in society.
That's something that isn't true. Women are higher-risk borrowers in some ways -- that likelihood of taking time off work, and thus income, to give birth for example. They're also lower-risk in other ways -- fewer women try to launch ludicrously ambitious schemes than men.
Those differences between groups, where they're a result of what economists call taste discrimination (essentially, just human bias), AI should not include them. But when they're the result of rational discrimination then AI must include them. But the current activist insistence seems to be that any difference across groups can only be as a result of bias, of taste discrimination. There's no room left in that world of equitable outcomes for there to actually be rational discrimination -- something which I insist simply is not true.
AI must include the bias of society. That part of it that is rational at least. For if we start lending money, or investing in startups and equities, without taking account of the things we do in fact know and that are coded into society as a result of that knowledge then we're going to be doing a lot of lending to, investing in, people who shouldn't be getting the money. Which will lead to the banks and investors all going bust again.
AI does offer opportunities. It also offers potential problems -- just like any and every technology ever adopted. But an effective AI in the financial sector will have to violate those activist demands of a non-biased AI.
The simpler way to put this is that we want to use AI to manage reality. Therefore the AI must actually describe reality for it to be useful. Creating AIs that only describe the world as some wish it would be doesn't work. Thus AI must be as rationally biased as society is.
Tim Worstall is a senior fellow at the Adam Smith Institute in London.
