Embedding AI and blockchain into next-generation tax policy design

If there is one regulation we can partially convert into code, it’s tax.

Author: Ignacio Longarte, Expert in Digital Economy Taxation

I feel that we need to change the way we structurally design the tax law production process in order to adapt it to -driven regulation that benefits from the evolution of technology and business in a digitalized global economy. Redesigning how we conceptualize some tax law segments to make them AI processing-compatible will be a landmark first step that can lead to many other possibilities. Though this is easier said than done, it’s possible to make significant progress in the international taxation field. We must seriously consider doing this as soon as possible and completely out of the current international “battlefield circuit” of digital economy taxation, in which tax sovereignty and taxation rights for the new era are being discussed in a heated basis and with clear and officially recognized discrepancies between different country’s interest groups and economic blocs. We must then advance and work on the structural approach and the logic of the system in parallel.

Machine-readable tax policy can improve public services dramatically, contribute to leveling the international playing field and foster digital businesses real-time iterations while generating efficiency savings for taxpayers—but first, we need a strategic visionary architecture with a comprehensible logic to inform the system.

The capacity of a country to autonomously write up and design its tax code system is a critical foundation for each sovereignty, and there is no need to break with this entirely to redefine the boundaries of this concept for a data-driven, technology–centric global economy and make it a reality. Nonetheless, some sacrifice in national autonomy is absolutely necessary if we want to magnify rather than obstruct the economic interdependencies among countries and blocs in the digitalized economy.

Much has been written about the need to regulate the fields where AI will make a significant impact. However, the idea behind this article is different: it explores how we can rethink the way we design regulations to make them suitable for post-coming-into-force automatic execution and building applications for them that contribute to social progress. And “tax regs” would be a perfect lab.

The idea would be to develop a structural AI-driven regulatory design that can be used by supranational organizations, governments and local tax administrations as a founding base to foster international commerce and provide certainty in some parts of it.

For doing so, we need to develop teams that can build up a comprehensive understanding of the production/reduction of tax policy problems/objectives within the logic of computation, at least for some segments/pieces of every new tax regulation where that “logic-reduction” would be perfectly possible for an expert in the field.

This article explores how we can rethink the way we design regulations to make them suitable for post-coming-into-force automatic execution and building applications that contribute to social progress.

Defining a new taxonomy for each tax regulation would probably be necessary. It would be difficult and impractical to carry out this exercise without connecting the approach to related fields where regulations use part of the same data. In any case, a strategy and roadmap must be defined.

In the last 10 years, we have seen an increased, vibrant focus on applying AI to legal systems, mainly oriented towards developing knowledge-based systems and smart information retrieval tools.

Most AI-based research is focused on enabling machines to perform tasks that would otherwise require human intelligence. In the legal field, past authors proposed semantically driven web services that could allow end users to negotiate and mediate in consumer disputes, family commerce, etc[1].

But let’s take a step back and elevate ourselves to gain a broader perspective. Only then we will realize that we need to produce some segments of the digital economy regulations to be AI-ready, integrating machine-dominated logic that can help us focus on other parts of the regulations or international economic agreements.

We could assert that all regulations have two types of articles in their body text:

  • Objective regulatory segments: the articles within any law that establish objective requirements or set off triggering elements that place you within the bounds of that regulation, and enumerate the consequences if you fall within it and fail to comply with it. In these segments, thresholds, numbers, percentages, formulas and measurable data are behind the key concepts and are more easily processed using NLP techniques.
  • Subjective regulatory segments: They are subjective in nature, or at least require a level of interpretation and a value judgment.

The objective regulatory segments are hard-fact driven and have the support of data available for machines to review their application with precision. These segments, whose text or requirements can be more easily reduced to be comprehended by a machine— which is connected to the right data sources with the right timing or sequential logic—can be used to establish co-relations with data from taxation and other disciplines as well.

Tax policy regulations also have subjective segments, just like any other regulation, and the more we move to the supranational level, the more they have. This is a discipline with probably one of the highest proportions of objective segments, which are typically measured in numbers, attributes related to numbers, parameters or formulas, or data that can be gathered regularly from a company’s or digital platform’s enterprise reporting systems. Redesigning the way we produce these segments to increase the quality of their objectivity for later machine automatic potential processing will produce very positive synergies.

Thinking about artificial intelligence in tax administration, if we start to produce some common tax law segments in “containers” like modern software development techniques suggest, we’ll ensure interoperability and scalability.

Artificial Intelligence tax preparation: A digital culture-driven approach to redefining tax regulation production

Tax regulatory framework needs to lead by example. Although it contains some segments/sections/parts that can be interpreted as any other regulation, there are many other parts of corporate tax, personal tax, VAT regs or transfer pricing whose logic can be reduced to a machine-driven language in order to enable immediate application, operation and review-audit of the regulation’s compliance or adherence level.

In comparison with other regulations, it will have a relatively bigger proportion that can be “software-defined,” precisely because part of what it covers is a structured type of data with strong numeric and categorization base. So clustering information in this field for the purpose of training the machine and defining an AI strategy should be easier to apply than in other regulatory fields (such as penal/criminal law for instance).

The structural definition itself will require a separate article that will follow later, but will need to be initially based on supervised learning from an AI-technical outlook.

There are many benefits that can be achieved in the international tax scheme by applying this type of framework, but the most evident is moving to automatic returns in personal income tax and in the more complex field of corporate tax returns for big MNEs or automatic custom clearance for products at borders.

Here are some indicative examples of other use cases:

  • Unification of elements to facilitate having just one value for all taxes (corporate tax, VAT, customs, as well as accounting).
  • Machine learning to process and test an enhanced objective definition of tax substance with supervised data taken from an MNE’s ERPs and other data sources, or to determine the main areas where value is produced in an MNE.
  • Tax and accounting asset management and control chains.

Artificial Intelligence in tax administration: Simplify first before securing controllable tax regulations as code through blockchain-based technology

It is no secret that software will rule the world. Telecom networks are moving to software-defined networks and data centers are moving to a hyper-converged data center infrastructure, and the same is expected for the regulatory production process. We need to move towards software and AI-defined regulations and execute tax policy directives more than in other fields— for what it represents internationally in terms of gross GDP impact on any country in the world, and because it fits the purpose well (as indicated earlier).

Once we do the necessary preliminary work to greatly simplify international tax principles, we’ll need to reduce some segments of tax regulations to a machine-learning mode so every tax regulatory body will have a “legislation as code” portion. When this challenging—but possible—task is complete, we’ll need to shift our attention to the issue of algorithmic governance, [1] although this is clearly further out:

“Algorithms can be used to monitor and to control iterative cycles of information within, and between, database flows. Algorithmic governance means governance by algorithms, in addition to the already existing governance of algorithms”.[2]

This recent concept requires a socially-driven research approach. Concern that humans will lose the capacity to control self-trained algorithms is reasonable, but distributed ledger technology that informs blockchain comes into play at this point in the tax arena.

Many organizations, like the New York University’s AI Now Institute, are indeed warning about the potential “bias” introduced by some algorithms and they recommend that every public agency that employs automated decision systems should carry out what they call “Algorithmic Impact Assessments”, in order to evaluate their potential bias level and allow public stakeholders and external auditors to examine it.

Due to the nature of tax regs, if we are smart in the way we structure the reg code, we can greatly increase the algorithms’ objectivity.

In addition, in the tax field, we can use some of the theoretical attributes of underlying blockchain technology, like the immutable transaction record and the capacity to in-brick logic into the system, to our benefit.

The issue of the algorithm’s solvency and bias is a separate topic in itself, but perhaps we need to seek out the solution in the consensus process ruling the smart contract that governs the execution of specific tax regulation segments, where voluntary taxpayers and tax administrations joined the regulation execution network on an automated basis—that’s why it’s absolutely necessary to simplify international tax principles and regulations first.

By introducing blockchain into the tax policy production process equation, other questions around best public-private collaboration strategies arise. As in other fields, we can envision several initial blockchain nimble use cases to begin testing, but it’s paramount to define the right strategy and priorities beforehand.

I don’t think we can support the whole new tax-coded regs through blockchain; we should instead begin by working on the critical parts that DLT tech contributes.

Most of the disruption in this distributed ledger arena comes from the startup level of the ecosystem, but the business model around a technology that could be used/ruled by a public body and by companies or individuals will likely need more than a public-private partnership code of conduct. We must first define open-collaboration initiatives and grant them funding, with full stakeholder involvement.

I know an initiative that is working in developing a block-chain based ecosystem to remove the cost, burden and complexity of the withholding tax refund process for the Asset Management industry in some transactions. This is a good example of an application development for pieces of tax regs affecting international funds movements that could have previously been converted into code to be applied by any fund, vehicle or industry player on an automated basis benefiting all participants in the value chain.

Deploying Artificial Intelligence to transform the tax world: Conclusion

There are two main emerging digital elements that led to the rise of the digital economy between 2005 and 2018, and that generated heated discussions around the BEPS Action 1, which the EU participates in with firm determination. They are AI, which is becoming a reality, and blockchain, which is now working in our favor as the basis a new way of producing international tax regulations supporting a highly principled and more nimble tax policy design.

The fundamental supranational issue here is that the enormous economic interdependencies generated between countries/regions through the digitalized economy collide with many countries taking unilateral paths under their tax sovereignty. That’s why we need to define a minimum set of system logic rules we can use to automate the execution of certain parts of the international economy tax system on a controlled basis, through some tax regulation segments being transformed into software code and matching the dynamism of the digital economy.

For this to happen, all countries will need to surrender a degree of autonomy in favor of a higher objective. Otherwise, the system will collapse at a point.

Our new role should be to embed the core logic and control the parameters that go into the international tax system affecting the digital economy from the foundation of the process. And to do so, we need a master plan soon.

Ignacio Longarte is an associate professor at IE Law School and member of our Hub editorial committee. He is an expert in Digital Economy Taxation, Digital Business Models and integrating regulatory, tax, IP and data governance elements across the international value chain to create new business models — his focus lies in innovating through data-driven business models. He has profound expertise in transfer pricing, digital value chain, international taxation, intangibles and permanent establishment risk mgmt. He is currently Chairman of DET3, a European based “Digital Economy Taxation Think Tank”, and a strategic advisor of listed global multinationals developing core digital elements in different sectors. Longarte is co-author of the first Transfer Pricing book in Spanish (2009-2011): Business Restructuring and Permanent Establishment chapters, is member of the Board/Advisory Board in European based Digital companies and a start-up investor. 

Note: The views expressed by the author of this paper are completely personal and do not represent the position of any affiliated institution.

[1] Marta Poblet-Pompeu Casanovas and others: ”Artificial Inteligence Approaches to the complexity of legal systems (Springer, 2009)”.
[2] See: “Algorithms, Performativity and Governability”; Lucas D. Introna, Lancaster University.
  See also www.governingalgorithms.org