On November 8th, 2023, in the midst of the stalled inter-institutional negotiations between the Council of the EU and the European Parliament (EP) on the regulation of foundation models (such as Generative AI) in relation to the future AI law, the Organisation for Economic Co-operation and Development (OECD) announced that it had updated its definition of AI systems.
It should be stressed that the OECD’s previous definition had been closely followed by European co-legislators in relation to the EU AI Act proposal. In its negotiating position of 14 June 2023 in relation to the AI Act, the EP proposed amending recital 6 of the AI Act in order to “closely” align the notion of an AI system in the AI Act with “the work of international organisations working on artificial intelligence” (Amendment 18), such as the OECD.
The European Parliament’s definition of an AI system is as follows: “a machine-based system that is designed to operate with varying levels of autonomy and that can, for explicit or implicit objectives, generate outputs such as predictions, recommendations, or decisions, that influence physical or virtual environments.” (Amendment 165)
The OECD’s previous definition of an AI system: “machine-based system that can, for a given set of human-defined objectives, makes predictions, recommendations, or decisions influencing real or virtual environments. AI systems are designed to operate with varying levels of autonomy”.
The EP’s proposed definition largely overlaps with the OECD’s previous definition. However, the Recommendation of the OECD regarding updating the definition of AI systems enlarges the notion (the additions are set out in bold and the wording that is to be removed has been
“An AI system is a machine-based system that
can, for a given set of human-defined explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as makes predictions, content, recommendations, or decisions that can influence ing physical real or virtual environments. Different AI systems are designed to operate with vary ing in their levels of autonomy and adaptiveness after deployment”.
According to the reasoning of the Professor of Computer Science at the University of California (Berkeley), Stuart Russell, who is behind the new definition of AI systems, it was deemed necessary:
a) to reflect the scientific consensus on the description of an AI system’s objectives;
b) to stress the important role of input;
c) to consider also the generative AI systems;
d) to clarify and align with other international processes and;
e) to integrate their adaptiveness.
More specifically, Professor Stuart Russell states that adding the word “content” to the updated definition stems from the need to apply the OECD Principles also to generative AI systems:
“The addition of the word “content” clarifies that the Recommendation applies to generative AI systems, which produce “content” (technically, a sub-set of “predictions, recommendations, or decisions”) such as text, video, or images.”
It should be noted that although the OECD’s definition has evolved, the alignment with the EP’s proposed definition is still apparent, except for the following words: “infer”, “input”, “content” and “adaptiveness”. The OECD is working on an Explanatory Memorandum to complement the definition and intends to provide further technical details. Meanwhile, a fifth trialogue on the AI Act is planned for the 6th of December 2023.