One of the many interesting aspects of the current global financial crisis is the role that AI and advanced computer software has played in it. It’s worth asking a few questions about exactly what this role has been — and what it suggests for the future.
There’s not much “strong AI” used in the finance industry today, but there are plenty of “narrow AI” software techniques in place, alongside other advanced mathematical methods — recognizing subtle nonlinear patterns in financial data, and relating financial data with other information about the economy, news, weather and so forth. And, certainly, this software played a role in the crisis that unfolded this fall.
But before we blame the machines, it’s worth digging a little deeper. The root of the problem really came down to the ways people chose to use this software to serve their own ends. And it’s arguable that, if we’d had strong AI instead of narrow AI involved, a crisis like this would never have come about.
Getting into the nitty-gritty, perhaps the central factor that paved the road to the recent financial disaster was the use of massive leverage — a technique allowing investors to gamble much more money than they actually have. If you bet $10 on some financial instrument with 30x leverage, you’re effectively betting $300. This means that if the instrument you’re betting on goes up by 10%, you make $300 from your $10 investment. But if the instrument goes down by 10%, you lose $300, which may be more money than you have. Leveraged investment obviously has great potential for both risk and reward — so its effective usage hinges on the accurate assessment of risk.
And risk assessment is one place where AI and other advanced software comes into play. The central aspect to consider, in understanding the strengths and limitations of advanced techniques for risk assessment, is context. The known mathematical and AI techniques for estimating the risk of complex financial instruments (like credit default swaps, and various other exotic derivatives) all depend on certain contextual assumptions. They apply well in some contexts, and not others. At this stage, some human intelligence is required to figure out whether the assumptions of a given mathematical technique really apply in a certain real-world situation. So, if one is confronted with a real-world situation where it’s unclear whether the assumptions of a certain mathematical technique really apply, it’s a human decision whether to apply the technique or not.
As a single example, Iceland’s financial situation was mathematically assessed to be stable, based on the assumption that (to simplify a little bit) a large number of depositors wouldn’t decide to simultaneously withdraw a lot of their money. This assumption had never been violated in past situations that were judged as relevant. Oops.
And of course, these technical considerations synergize with human psychology in various ways. Suppose you work for a financial institution and you’re asked to assess the risk of some complex financial instrument like a credit-default swap. Suppose the tools that you have at your disposal don’t let you make a confident assessment? What are you going to do? Admit it, or just try your best, and hope the world doesn’t fall too far short of the assumptions of the techniques you apply? Bear in mind that the reward structure for high-level financial employees tends to give big bonuses for decisions that lead to short-term gains. So if you play a role in a big win you may get rich … but if you play a role in a big loss, the worst that happens is you’ll need to find another job. But consider a more interesting question: What happens in another decade or two, when the AI field has progressed a bit and we have yet more intelligent software that is able to automatically assess whether the assumptions of a certain mathematical technique are applicable in a certain context?
it’s arguable that, if we’d had strong ai instead of narrow ai involved, a crisis like this would never have come about.
It may sound cavalier to say so (and the reader should understand that, as the owner of a small business, I’ve certainly not escaped the impact of the recent credit crunch), but my feeling as an AI expert is that these sorts of problems we’ve seen recently are merely hiccups on the path to super-efficient financial markets based on advanced AI. (On the other hand, it’s hard to say exactly how long it will take for AI to achieve the needed understanding of context, to avoid this sort of "minor glitch.")
Once the transhuman future fully kicks in, it’s quite possible that money will become obsolete. Or maybe, as in Charles Stross’s novel Accelerando, the next form of intelligence will emerge from complex AI-powered financial instruments. But well before these radical possibilities come about, I’d suggest that a big difference will be made by the deployment of more strong-AI-ish financial analysis systems with broad understanding of context. Whether further glitches occur along the path to this future is another question, of course… and the best piece of advice I can offer in this regard is that the movers and the shakers of the financial world should pay attention to the limitations — as well as the power — of current techniques, while also keeping an eye on the development of more advanced AI methods capable of overcoming these limitations.
Ben Goertzel is the CEO of AI companies Novamente and Biomind; a math PhD; writer; philosopher; musician; and all-around futurist maniac.