OECD updates AI definition ‘to inform AI Act’ – EURACTIV.com

The Council of the Organization for Economic Co-operation and Development (OECD) adopted on Wednesday (8 November) a new definition of artificial intelligence, which will be included in the new EU regulation on artificial intelligence.

The OECD was originally established to administer the Marshall Plan, the US stimulus package to finance the reconstruction of Europe devastated by World War II.

Since then, the organization has remained an international forum for economic cooperation with 38 member countries, and is often considered a club of rich countries. In this context, in 2019, the OECD proposed an an influential set of principles for reliable AI policies, which included an early definition of AI.

Wednesday’s decision officially updated that definition and is likely to be included in the EU’s upcoming AI regulation. The definition is a key aspect of the upcoming law as it defines its very scope.

“An artificial intelligence system is a machine-based system that, for explicit or implicit goals, infers, from the input data it receives, how to generate results such as predictions, content, recommendations or decisions that (may) affect physical or virtual environments. Different AI systems differ in their levels of autonomy and adaptability after deployment,” the new definition reads.

This definition was discussed in mid-October by the OECD’s Digital Economy Policy Committee and Working Group on Artificial Intelligence Governance. According to a presentation given at this joint session, the timetable was adjusted “to inform the EU AI Act”.

Compliance with the Law on Artificial Intelligence

The AI ​​Act is a legislative proposal to regulate AI based on its potential to cause harm. EU institutions are working to finalize the provisions of the world’s first comprehensive law on artificial intelligence by the end of the year.

In March, MEPs working on the dossier agreed to follow the OECD’s definition of artificial intelligence to maintain semantic alignment with international partners. The idea of ​​reaching a common ‘taxonomy’or classification system, for key concepts related to this emerging technology is also an important part of the work of the EU-US Trade and Technology Council.

However, EU legislators have faced the uncertainty that the OECD itself is on the way to updating its definition based on technological and market developments. That’s how the parliamentarians formulated the definition in an attempt to anticipate the future changes of the OECD – which they succeeded quite accurately.

As the Law on Artificial Intelligence entered the last stage of the legislative process, the so-called trilogues in which the European Commission, the Council and the Parliament agree on the final provisions, policymakers decided to put the debate on the definition on ice until the OECD makes its decision. mind.

The rationale for the change

According to a joint presentation seen by Euractiv, the reasons for updating the definition relate to the international harmonization of AI definitions, reflect developments over the past five years, improve technical accuracy and clarity, and make it “more future-proof”.

One of the main changes was to remove the reference to the fact that goals must be defined by humans in order to capture cases where an AI system can learn new goals.

According to a draft explanatory memorandum shared with the presentation, “design goals can be supplemented by user instructions when the system is in operation,” as is the case with foundation models.

Moreover, the memorandum notes that there could often be a mismatch between the explicit goal and the result, or unintended consequences.

The term ‘infer how to generate outputs’ is also introduced to explain when an AI model receives inputs from the environment and arrives at a corresponding output through one or more algorithms.

The type of output that AI can produce is also extended to content such as text, videos or images, as is the case with generative AI models such as ChatGPT and Stable Diffusion.

Finally, the reference to adaptability reflects that some AI systems may evolve beyond the design and implementation phase, particularly those based on machine learning techniques.

Next step

With the new OECD definition of artificial intelligence now official, it is expected to be included in the EU’s proposed AI law. However, EU policymakers already received the revised definition in mid-October, and no internal text reflecting the change has been circulated to date.

(Edited by Nathalie Weatherald)

Read more with EURACTIV

Source link

Leave a Comment