This article, titled “Why ‘Artificial Intelligence’ Should Not Be Regulated,” by Daniel Braun from the Department of Hightech-Business and Entrepreneurship at the University of Twente, presents a critical commentary on the global efforts to regulate Artificial Intelligence (AI). Published in Digit. Gov. Res. Pract. in June 2025, the article argues against the use of “Artificial Intelligence” as a regulatory category.
The central argument is that while there is a potential for harm from AI systems, “Artificial Intelligence” is a vaguely defined label that is neither suitable nor necessary for comprehensive regulation of technological risks. Instead, Braun advocates that lawmakers should focus on regulating high-risk applications of software, regardless of whether they are labeled as AI.
Key points and insights from the article include:
- Ambiguity of AI Definitions:
- The term “Artificial Intelligence” has been problematic since its coining 68 years ago at the “Dartmouth Summer Research Project on Artificial Intelligence,” as the word “intelligence” itself is ill-defined and difficult to measure in humans.
- Lawmakers lack a universally accepted definition of AI, making regulation challenging.
- In academic literature, there isn’t a single accepted definition either; Russell and Norvig’s standard reference identifies four dimensions: systems that act human, think like humans, act rationally, or think rationally.
- A common trend, both in scholarly literature and legislation, is to use “AI” interchangeably with “Machine Learning” (ML), even though ML is only a subset of AI. This often leads to the exclusion of important branches like expert systems, which are knowledge-based systems applying symbolic representations of human knowledge.
- Some definitions even suggest that as an AI approach improves and achieves “human-level” performance, it would cease to be considered AI.
- Critique of the EU AI Act:
- The article particularly focuses on the European Union’s (EU) AI Act, which is seen as potentially influential for future global AI legislation.
- Despite its stated goal to take a risk-based approach and regulate harmful applications rather than technology, the AI Act is criticized for failing to achieve this.
- The definition of AI within the EU AI Act evolved significantly through different legislative drafts from the European Commission, Council, and Parliament, eventually adopting an updated definition from the OECD in its final version.
- The final definition describes an “AI system” as a machine-based system operating with varying levels of autonomy, potentially exhibiting adaptiveness, and inferring how to generate outputs that influence environments.
- Braun highlights the ambiguity of the term “infers” in the AI Act’s definition and points out a significant contradiction: the recital of the AI Act explicitly aims to exclude “simpler traditional software systems or programming approaches” and systems “based on the rules defined solely by natural persons” while simultaneously including “logic-based and knowledge-based approaches that infer from encoded knowledge or symbolic representation”. This inclusion of logic-based systems, such as expert systems, which are typically based on human-defined rules, creates an oxymoron.
- The AI Act’s introduction of a separate category for “general-purpose AI models” (like GPT models), regulated independently of their application risk and based on computational power (FLOPs), is seen as a deviation from its core risk-based principles and an attempt to respond to public debates about generative AI. This metric is criticized for being quickly outdated and not reliably indicative of a model’s capabilities or risk.
- Argument for Application-Based Regulation:
- The article uses the example of hiring systems (System A: ML-based, System B: expert system, System C: simple rule-based system) to illustrate that the potential for harm, such as discrimination due to biases in grading schemes, is independent of the underlying technology (AI or non-AI software).
- The author asserts that the harm a system can cause is not primarily determined by the technology it uses. Therefore, regulatory guidelines should demand transparency, explainability, and proper documentation for any software system making significant decisions, regardless of its AI classification.
- The commentary concludes that lawmakers should abandon the pursuit of defining “Artificial Intelligence” for regulation and instead focus on addressing the risks posed by different applications of software. It suggests transforming the AI Act into a “Software Systems Act” that regulates all software systems impacting people’s lives.
- Call for Interdisciplinary Engagement: The author emphasizes the importance of computer scientists and AI researchers taking a more active role in policy debates, as current AI regulations, including the AI Act, are deeply technical and make assumptions that are debated within the AI community.
This article serves as an important contribution to the ongoing debate about AI regulation, offering a technical perspective that challenges the fundamental premise of regulating “AI” as a distinct category.
Reference: Braun, D. (2025). Why “Artificial Intelligence” Should Not Be Regulated. Digit. Gov. Res. Pract., 6(2), 30:1–30:16. https://doi.org/10.1145/3696010
