AI System Definition Guidelines: A Critical Review
The recently published guidelines by the European Commission regarding the definition of an artificial intelligence (AI) system have been met with criticism for their lack of clarity. These guidelines were intended to assist developers, users, and enforcers in understanding the definition, yet they appear to add confusion rather than resolve it.
Understanding the AI Act
The EU’s AI regulation, known as the AI Act, 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.”
This definition is crucial as it sets the stage for what is considered an AI system within the scope of regulation.
Key Issues Identified
Upon reviewing the guidelines, three significant issues emerge regarding their interpretation of the AI system definition:
1. Inclusion of Logistic Regression
The guidelines state that “Systems for improving mathematical optimization” are out of scope. However, it is mentioned that methods like linear or logistic regression fall under this category. This is problematic because, in different contexts, these methods can be included within the AI Act’s scope. The distinction made in Paragraph 45 between “optimising the functioning of the systems” and “adjustments of their decision-making models” indicates that the latter remains governed by the AI Act. Thus, applications utilizing logistic regression for crucial decision-making processes would indeed fall within the law’s purview.
2. Contradiction with AI Act Recitals
The guidelines attempt to differentiate AI systems from traditional software systems, yet they contradict the AI Act. Recital 12 of the AI Act emphasizes that a key characteristic of AI systems is their capability to infer, which transcends basic data processing. However, the guidelines assert that certain optimization methods, despite having the capacity to infer, do not surpass “basic data processing”.
3. Questionable Reasoning
One justification provided in the guidelines states that a system’s long-term usage could indicate it does not transcend basic data processing. This reasoning seems flawed, as the duration of a system’s use should not determine its classification as an AI system. Further, the guidelines suggest that “All machine-based systems whose performance can be achieved via a basic statistical learning rule” fall outside the AI system definition due to their performance. Such explanations only contribute to the prevailing confusion surrounding the guidelines.
Conclusion
In summary, the European Commission’s guidelines on AI system definitions are criticized for failing to provide the clarity they aimed for. Instead, they introduce ambiguity and confusion about what constitutes an AI system under the AI Act. Fortunately, these guidelines are not legally binding, and it is hoped that regulators will apply sound reasoning in their interpretation of AI systems moving forward.
