7 February 2025 — The guidelines published by the European Commission on the definition of an artificial intelligence system fail to provide clarity.
The Guidelines published by the European Commission are supposed to help developers, users, affected people and enforcers in interpreting the definition.
The EU’s AI regulation ‘AI Act’[1] defines an AI system as:
“a machine-based system that is designed to operate with varying levels of autonomy and that may exhibit adaptiveness after deployment, and that, for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that can influence physical or virtual environments;”[2]
We point out three issues in how the guidelines interpret this definition.
Logistic regression is within the scope of the definition
Paragraph 42 elaborates that “Systems for improving mathematical optimization”[3] are out of scope. In this context, linear or logistic regression are provided as example methods. Outside this context, these methods would be within the scope of the law.
This becomes clear when reading Paragraph 45 that makes the distinction between “optimising the functioning of the systems” and “adjustments of their decision making models”[4]. The latter remains within the scope of the AI Act, which covers applications, including high-risk areas such as evaluating “the eligibility of natural persons for essential public assistance benefits and services”[5] that involve decision making about people. The use of logistic regression in such applications would involve making “adjustments of their decision making models”.
Thus, logistic regression is within scope of the AI Act.
Contradiction between a Recital and the guidelines
In its attempt to distinguish between AI systems and “simpler traditional software systems or programming approaches”, the guidelines contribute to more confusion.
Recital 12 of the AI Act states, “A key characteristic of AI systems is their capability to infer” and the “capacity of an AI system to infer transcends basic data processing by enabling learning, reasoning or modelling.”[6] Any one of the characteristics suffice to transcend “basic data processing”.
However, the guidelines contradict the Recital: “while those models [optimisation methods] have the capacity to infer, they do not transcend ‘basic data processing’.”[7]
Strange reasoning
The reason provided is bizarre: “An indication that a system does not transcend basic data processing could be that it has been used in consolidated manner for many years.”[8] How long a system has been used should be irrelevant here.
This is not the only case where strange explanations are used to place AI techniques out of scope. Another excuse is performance (emphasis added):
“All machine-based systems whose performance can be achieved via a basic statistical learning rule, while technically may be classified as relying on machine learning approaches fall outside the scope of the AI system definition, due to its performance.”[9]
These guidelines fail to provide clarity and instead add to the confusion.
Fortunately, these guidelines are not legally binding. We hope that the regulators and bodies responsible for fundamental rights will use sound reasoning in interpreting the definition of AI systems.
See ICCLs related work on artificial intelligence.
Notes
[1] Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence and amending Regulations (EC) No 300/2008, (EU) No 167/2013, (EU) No 168/2013, (EU) 2018/858, (EU) 2018/1139 and (EU) 2019/2144 and Directives 2014/90/EU, (EU) 2016/797 and (EU) 2020/1828 (Artificial Intelligence Act). URL: https://eur-lex.europa.eu/eli/reg/2024/1689/oj
[2] Artificial Intelligence Act, Article 3 (1).
[3] It is unclear why the US English spelling “optimization” is used. The guidelines place “Systems for improving mathematical optimization” out of scope despite recognising in Paragraph 24 of the guidelines that explicit objectives of an AI system “may be specified as the optimisation of some cost function, a probability, or a cumulative reward”.
[4] Guidelines, Paragraph 45.
[5] Artificial Intelligence Act, Annex III, 5 (a).
[6] Artificial Intelligence Act, Recital 12.
[7] Guidelines, Paragraph 42.
[8] Ibid.
[9] Guidelines, Paragraph 49.