top of page

Position Statement on the Scope of Salesforce’s AI Economist in Tax Regimes

Submitted by Siddharth Jain, Research Member.

  • Recently Salesforce rolled out the ‘AI Economist’, a novel line of research that leverages reinforcement learning to support policymakers and regulators anticipate the outcomes of various tax schemes. The objective of AI Economist is to help leaders build strategies that achieve clear social targets such as improving the middle class.

  • The new AI system uses a set of AI agents to model how actual individuals would respond to various taxes. By gathering and selling resources and commodities, and building houses, each AI agent generates income. Through changing their travel, selling, and constructing behaviour, they learn to optimize their utility (happiness). When the simulation is going, the AI Economist is trying to maximize taxes and subsidies to achieve different objectives. Therefore, what the system does is that it simulates centuries of years of income under different slabs and patterns of income in a short period of time, and then produce an algorithmic prediction of how the people would react to different types of taxes.

  • Classic tax theory emphasizes on individuals who receive money while doing work, receiving revenue from wages while incurring labour costs. Individuals are believed to vary in their degree of expertise, meaning that low-skilled employees are less successful and earn less income with the same amount of labour as highly skilled workers. This adds to inequalities, and the challenge for policymakers is that whilst wage growth can be desired to increase prosperity, higher taxes will decrease the amount individuals want to operate, which can have an especially strong effect on highly qualified employees.

  • The study is focused on the assumption that an "optimal" equilibrium between efficiency and equity can be achieved by the right tax policy. Although there are undoubtedly measures that can improve both growth and equity, inevitably a free system produces trade-offs between the two.

  • Taxes and subsidies are critical tools used by policymakers in rising inequalities and redistributing income. However, with a broad variety of societal priorities, such as the trade-off between equity and profitability, the framers have not yet worked out how to enforce effective tax policies. Economic theory cannot fully model the complexities of the real world, and precautionary real-world tax experimentation is nearly impossible. Through AI Economist, Salesforce would be using reinforcement learning to tax analysis to provide simulation and data-driven approach for determining optimal taxes for a defined socio-economic target.

  • The planner integrates a social welfare feature that considers the trade-off between income equality and productivity, where "equality" is specified as a complement to an index on wealth distribution. When it does all of this, the agents/AI population learns to cheat the feature and tax schedule to reduce their effective tax rate, partially by leveraging loopholes such as rotating high- and low-income tax cycles.

  • This fiscal tug-of-war entails the AI Economist and agents — each self-improving in their skills as long as a sense of equilibrium is achieved.


  • In the experiments conducted, The AI Economist increases the equity and growth balance by at least 16 % relative to the modification of the Saez tax system, US Federal income levels, and the free market economy. The results also showed that the redistribution enhanced the quality across all policies at the cost of productivity.

  • The AI Economist learns to set different tax schedules: higher top tax levels and lower-middle-income levels which produce higher net low-income subsidies. As the agents were trying to find and work on the loopholes, the system was simultaneously trying to find retaliation to these loopholes and has been successful in doing so to a considerable extent.

Indian Perspective

  • While the developers are positive about the outputs of AI Economist in varied dynamics, the system would be put to the real test in the Indian economy. India is still a developing country with the world’s second-largest population. To add more to this, the income gap in India is also one of the largest in the world, therefore, making the work of the AI much more difficult and complex.

  • If AI Economist is used for the work tax framing in India, it would be interesting to see how the people react to it because the population is still trying to get hold of the GST (new regime introduced on 2017).

  • Therefore, it would not be wrong to say that a more dynamic and adaptive response is required from the AI, especially in the third-world economies like India since these are unstable economies as compared to countries like the USA, Germany, Japan, etc. Also, a lot of other factors have to be kept in mind such as geography, size of the working population needs and wants of the population, aspirations from the government of the population, etc.


  • AI Economist can prove to be a tool that offers economists and administrations unparalleled simulation capabilities to improve their analysis and conclusions. To help them run tests on the AI Economy, we want to work with more economists and governments to see how this relates to the actual world.

  • Although the developers acknowledge this is an early effort and expect over time to build in more nuanced inputs, they compare it to early work into genomes. It didn't yield tangible results right away, but with time we've seen the growth of resources like CRISPR and they believe this methodology will have a positive effect on tax policy because it expands on its initial work.

  • It is beyond doubt that the integration of artificial intelligence and policymaking would be of great help and would resolve a lot of issues at hand, the developers also need to address the problem of legal liability. Whether a legal liability can be created against the AI Economist or not, and if ‘yes’, to what extent? Also, if such liability would be uniform globally or different sets of law would be needed in every country.

  • Concludingly, the advent of artificial intelligence-based policy formation would surely bring in some problems of its kind, but the merits of the AI outweigh the demerits of it, hence making the use systems like the AI Economist inevitable in the future.

For further reference:

For more queries, do not hesitate to mail us at


The Indian Society of Artificial Intelligence and Law is a technology law think tank founded by Abhivardhan in 2018. Our mission as a non-profit industry body for the analytics & AI industry in India is to promote responsible development of artificial intelligence and its standardisation in India.


Since 2022, the research operations of the Society have been subsumed under VLiGTA® by Indic Pacific Legal Research.

ISAIL has supported two independent journals, namely - the Indic Journal of International Law and the Indian Journal of Artificial Intelligence and Law. It also supports an independent media and podcast initiative - The Bharat Pacific.

bottom of page