Introduction

Have you ever wondered how Artificial Intelligence (AI) could be used to improve tax laws? While it might seem like a futuristic concept, framing tax optimization as an adversarial game between taxpayers and tax authorities offers a unique perspective.

The New Zealand Context

New Zealand’s General Anti-Avoidance Rule (GAAR) relies on judges interpreting the law, creating a system where precedent sets future rulings. This lack of clear, unambiguous rules makes it challenging to achieve fairness and efficiency in tax collection.

The Challenge of Incomplete Laws:

Our current tax laws are incomplete, representing an approximation of an ideal system that rewards contribution, protects individuals, and ensures fairness. Mistakes and loopholes exist, allowing some taxpayers (with help from accountants and lawyers) to exploit these weaknesses.

Tax Optimization as an Adversarial Game

Imagine a game between the Inland Revenue Department (IRD) and taxpayers. The IRD writes and passes tax legislation, taxpayers organize their finances to minimize tax burden (legally, of course!), and the IRD observes the aggregate tax loss. They can then choose to consult the “oracle” (judges) at a high cost to determine if a taxpayer’s strategy is legal.

The Difficulties

This game presents several challenges:

  • Computation Asymmetry: Taxpayers only need one strategy - minimize tax legally. The IRD needs to find and close all potential loopholes.
  • Memory Asymmetry: Taxpayers can easily track changes in tax law, potentially collaborating to find new strategies. The IRD struggles to model and track the financial maneuvers of countless taxpayers.

The Learning Curve:

We’re essentially trying to approximate the ideal tax system with limited information. As the IRD learns and improves its tax laws, taxpayers adapt their strategies. The key is to find a balance between learning efficient tax collection and minimizing loopholes exploitable by taxpayers.

Open Questions

  • Can we design un-exploitable tax laws (a truly complete set of rules)?
  • Is designing a secure system inherently harder than finding ways to exploit it?
  • How can we prove a tax strategy is truly un-exploitable?
  • Could we measure “gameability” to assess loopholes’ ease of exploitation?
  • How can we analyze historical data to see how quickly the IRD learns compared to taxpayer adaptation?

Conclusion

Exploring tax optimization through an AI and game theory lens offers new perspectives on creating a fairer and more efficient tax system. By understanding the challenges and open questions, we can move towards a future where technology helps us achieve a more balanced approach to tax collection.