how artificial intelligence can help fix economies
Perhaps one of the most interesting and alarming things about the COVID-19 pandemic is just how big of a spotlight it cast on the flaws in our global economy. No small part of the blame lies with supply-side economics, a toxic dogma based on nothing more than a napkin doodle by an economist few took seriously which shellacked the world with four major recessions in 40 years. But if we tear down the current economic order, what do we put in its place? With no shortage of alternatives which look far better on paper and even have positive track records on small scales, how do we know which would work best given the scale, dynamism, and complexity of modern global trade without expensive and prolonged trial and error?
According to Salesforce, yes, the tech company, the answer may be artificial intelligence. Now, it’s important to note that the idea of using AI to command economic policy isn’t exactly new and some utopians went as far as to suggest that some form of computer-mediated communism is the only way out of our current politically and socially unstable predicament. It’s also crucial to point out that a massive neural network with billions of inputs, millions of outputs, and thousands of layers may not understand all the intricacies of the economy it’s trying to model and come up with nonsensical and ineffective solutions just as often as good ones. Thankfully, Salesforce is not proposing either of those approaches.
Their much more elegant and practical solution is to use multiple interacting AI models to come up with different ideas for policies we could try based on the scale and complexity of economic scenarios it’s simulating. Instead of starting with pre-existing assumptions about how people will respond to a certain action, assumptions which have often been proven wrong, incomplete, or guaranteed to end badly, we can start with a blank slate, simulate the economic interactions we want to better understand, and see what policies the AI will determine to be the winning ones. That ability to challenge our existing worldviews while carrying out thousands of experiments to refine our ideas can be a very powerful tool in reshaping our economic priorities.
Of course, there are limitations here too. Salesforce’s model was very simple, had few agents, and its interactions were trivial compared to the real world. However, it’s enough to show that the approach can be scaled up and yields enough interesting insights for economists to want to use it before recommending new policies to get us out of a recession, cool down a potential bubble, or overhaul tax policies. With our current simplistic, outdated models easily corrupted by political dogmas failing to make the world a more stable place, never mind better, it would make far more sense to dial up an economy at the scale you want to study and see what the math has to say before committing to a new policy idea.
As the models will grow larger and larger to simulate entire nations, regions, and even the world itself, there’s a high likelihood that the AI will recommend ever more bizarre policies with lower confidence intervals, but that’s actually okay. The end goal isn’t to create a utopia based on the oracular proscriptions of a magic black box, but to find patterns and connections that escape our existing modeling techniques, see how closely they match what we see in the real world, then fine tune new proposals in enough trial runs across enough scale models to offer genuinely new solutions likely to make a real difference when implemented, minimizing both human bias and cognitive dissonance in the process. And that ability would be invaluable in the long term.