Algorithmic Policymaking: The Way Out of COVID-19?
As “The Great Lockdown” runs its course; economic collapse has ensued. This is certainly a crisis like no other, and developing and developed countries alike face concurrent crises: a health crisis, a financial crisis, and a collapse in commodity prices. The sheer magnitude of collapse is a once-in-a-lifetime experience. Consequently, there is considerable uncertainty about what the economic environment will look like once the pandemic has subdued. Policymakers simply haven’t seen anything like this in living memory.
But there is hope. In late April, researchers from Harvard University and Salesforce unveiled the “AI Economist.” In their own words, the AI Economist is
“a new line of research that studies how to improve economic design using AI with the goal of optimizing productivity and social equality for everyone. This new AI framework is designed to simulate millions of years of economies – in parallel – to help economists, governments, and others design tax policies that optimize social outcomes in the real world.”
The AI Economist is data-driven: it uses reinforcement-learning (RL) to model millions of interactions between different economic actors. So-called “rational agent programs” are trained on historical data, and then given an objective to achieve. Google’s folly with AlphaGo in 2016 proved to be wildly successful, as the computer system clinched a 4-1 victory over a South Korean Go grandmaster. Clearly, there is considerable potential for the technology.
Algorithmic economics is nothing new. Amazon has been using it for years to set prices by modeling auction behaviour. But the AI Economist proves to be the finest mallet in the policymaker’s toolbox during this pandemic – as it allows policymakers to test proposals in a dynamic environment. The team hopes that the AI Economist could simulate the impact of universal basic income, environmental regulations, and fiscal stimulus measures. Such simulations could prove to be invaluable during a downturn, where policy changes can be particularly sensitive and rash.
Salesforce has admitted that the model isn’t “realistic enough at this point.” It does not model many real-world factors - such as inherited wealth, voluntary work, and the ability to set up a company. It is also difficult to model irrationality. Richard Socher, Salesforce’s chief scientist, admits that “AI agents are hyper-rational. But people are not always super-rational. Jamaica is happier than Germany even if the objective measures would suggest otherwise.” After all, economics is not easily predictable nor can it be approached from a scientific perspective. Its interactions are massively complex.
Tread carefully, say some. Historical data sets do not reflect marginalized members of society and can succumb to partisan political abuse. But it is irrefutably a step in the right direction. Whilst the AI Economist might not be ready to lead us out of the current downturn; it certainly keeps us hopeful for the future.