Economic inequality is a universal concern as it has adverse impact on economic prospects, health & social welfare. The onus of addressing this issue falls on governments, who are inherently tasked with minimizing inequality to the greatest extent achievable, if not eradicating it completely. Taxes are often the primary tools that are used to address this issue.
Taxes come with an inherent trade-off between equality and productivity. A higher tax rate can mitigate economic disparities but might also discourage workforce participation Conversely, lower tax rates can boost productivity but exacerbate economic inequality. An optimal tax policy must optimize the balance between equality & productivity.
The Salesforce research team devised a two-level reinforcement learning* framework to test The AI Economist.
In the first level or the “inner loop”, workers or ‘agents’ emulate citizens' behaviour. They engage in labour, receive income, and pay taxes. They adapt their behaviour overtime to maximise utility given a fixed tax policy.
In the second level or the “outer loop”, The AI Economist analyses inequality, wealth distribution, and other market dynamics to adjust the tax policy to maximise our social objectives. The primary aim for The AI Economist here is to maximise agent productivity and equality.
The interaction of the inner loop and the outer loop creates a dynamic environment for all agents. In other words: because the post-tax income for the same type and amount of labour can change over time, agent decisions that were optimal in the past, might not be optimal in the present.
The experiment simulated and contrasted the AI Economist's recommendations against outcomes derived from the 2018 US Federal Tax Rates, the SAEZ Model for taxation, and the Free Market model. The AI Economist demonstrated a 16% improvement in the trade-off between equality and productivity compared to the SAEZ Model, and it outperformed adaptations of the US Federal income tax and the free market by an even wider margin.
Furthermore, the AI Economist showcased effectiveness in simulations involving human participants. It achieved competitive equilibrium between equality and productivity, surpassing baseline models. This suggests the potential of this approach to enhance real-world social outcomes.
The AI Economist's purpose is to facilitate an unbiased examination of policy impact on real economies, surpassing the complexity levels that conventional economic research struggles to address. The research team envisions a wealth of exciting research directions at the intersection of machine learning and economics, offering significant opportunities for machine learning to generate positive societal influence.
The AI Economist opens the doors to a new era of collaborative exploration at the nexus of machine learning and economics. As this technology continues to evolve, it holds the potential to revolutionize how we comprehend and address complex societal challenges.
* Reinforcement learning is an area of machine learning concerned with how intelligent agents ought to take actions in an environment to maximize the notion of cumulative reward.
References:
https://blog.salesforceairesearch.com/the-ai-economist/
Comments