why our next economic crisis could be caused by artificial intelligence

The only worse thing than basing all of human existence around the highs and lows of arbitrary numbers is putting computers in charge of these numbers.

Our world is often at the mercy of stock markets. When the markets run hot, the rich get richer, and when they tumble, the not rich lose their jobs, so despite the disparity of returns from said markets, we have every incentive to keep them trending up and to the right. We also know that, in very fundamental terms, the trick is to buy low and sell high no matter how complex the transaction or valuation processes are under the hood so we tend not to pay much attention to how markets today really work on a day to day basis, often overwhelmed with impenetrable financial jargon when we actually try. But the SEC knows all these ins and outs, and its experts are very worried.

The regulator’s main concern is that while we keep on adopting AI models for almost everything as our world has becomes so vast, complex, and interconnected, humans can’t keep track of it all. But even as we try to figure out how AI could improve entire economies, we don’t know how accurate these models will be over the long term. You see, the power of AI lies in keeping track of potentially billions of market parameters, from critical, to useful, to frivolous but potentially instructive. That much information will overwhelm any human, yes, but computers can use it to find predictive patterns, trends, warning signs, and investment opportunities in the right context.

why big finance embraced machine learning

But the problem is that we can’t check if those patterns really hold while also betting money on or against them without suffering a financial blowout if we read the outputs wrong, or the model fails when its training doesn’t match real market dynamics. Keep in mind that valuations change all the time, rising and falling prices create changes in the political and regulatory landscapes, and market trends can force people to make choices that seem weird to models that mostly care about the spreads and trends of statistical indicators and assumes that humans follow game theory, when in fact they do not, no matter what economists favored by pundits bellow.

Even worse, these models can fail in subtle ways, their predictions growing less and less accurate, building up to a sudden crash after what looks like nothing more than slightly sluggish performance as the real world data finally falls enough out of sync with the models’ training sets and feedback loops that their accuracy plummets low enough to enable that fatal set of trades and advice resulting in the bottom suddenly falling out. Having seen what happens if funds are convinced that their models are perfect and eliminate almost all risk in the subprime boom, the SEC is not convinced that the financial industry learned its lesson and won’t repeat their follies with AI.

Their nightmare scenario is that after a decades long era of basically free money with near zero interest rates, investors are too used to outsize returns and growth. Armed with AI models, funds could make the pitch that they’re now able to adjust to markets in real time thanks to a magic technological black box, therefore they can still deliver massive returns. Investors plow trillions into the funds powered by machine learning algorithms, make a killing for a while, then, with one “black swan event” like another pandemic, war, or a destabilizing spasm of political violence, the model sputters out, repeating the Great Recession as the dust starts to settle.

why markets need trust, not blind faith

Without a very detailed map of what happens during “these unprecedented times” as we’ve been calling the past three years, there’s a very high likelihood that the models won’t understand what certain market shifts really mean and tell investors to make the wrong bets. And what happens next is what’s bothering the SEC. Will the banks jump in to override the models before the trades are automatically made, or reverse them for a few weeks of ultimately harmless chaos? Or would they simply assume the AI is correct and let it all ride — off a cliff into another recession? Do all these trillion dollar models come with safeguards? Recent history says we should be very skeptical.

Ultimately, it’s those safeguards that are sorely needed. Markets are still flush in more money than ever before, managing debts that exceed $307 trillion — compared to the global monetary supply of maybe about $83 trillion — in an economy many orders of magnitude more complicated than it was even 30 years ago, and getting even weirder and more complicated thanks to the meteoric, world-rattling rise in automation. And now, on top all that, as the age of easy money seems to be over, to keep engaging in computer-enabled speculation as if nothing has changes without a proper Plan B and safeguards would be criminally irresponsible.