A Thought Experiment on the Application of Neuro-Evolution Machine Learning in International Taxation Policymaking

Abstract

The article proposes a novel method to apply Neuro-Evolution Machine Learning in International Taxation policymaking by generating a simulation of the international taxation framework. The simulation can be conducted in similar way to a "computer game" where multinational enterprises and jurisdictions are interacting in a flexible programmed environment designed to simulate the real world international taxation framework. The neural networks of the simulated jurisdictions and multinational enterprises are self-trained through Neuro-evolution machine learning with a predetermined goal to maximize their fitness formula (such as net tax for jurisdictions and profits after tax for multinational enterprises) and without human intervention.

The proposed method can be adapted in evaluating different tax policies via examining sets of policies and activities of simulated jurisdictions and multinational enterprises when the simulation reaches an "equilibrium" (i.e. the most effective policies and tax planning are adopted by the artificial intelligence). Examining shifts in the equilibrium is valuable in evaluating the impact of different tax policies and predicting tax planning response of multinationals to the new policies. The system can be also adjusted to determine which combination of policies can achieve various social goals by changing fitness formulas variables and examining which policies the artificial intelligence will adopt.

Keywords: International Taxation, Tax Policy, BEPS, Base Erosion and Profit Shifting, tax, taxation, machine learning, neuro-evolution, neuroevolution, Artificial Intelligence

JEL Classification: K34