Defining Artificial Intelligence

As everybody familiar with the field is aware, a generally agreed-upon definition of AI does NOT exist.

The AI Regulation drafted by the European Commission only defines AI either by attempting a scantily sketched taxonomy in Annex I (machine learning, statistical approaches, formal approaches, and that is all as of now) or through a teleological approach (which are the objectives of an AI system?).

However, while it is undeniably true that artificial intelligence in general has been a moving target since the Fifties, it is also true that some kind of all-encompassing definition is much needed, lest such a yawning gap should adversely impact governance of research, development, and innovation.

This publication by the JRC tries to fill in exactly THIS gap. The authors try and tackle the hard problem of tracing the boundaries of a definition by reviewing AI literature through natural language processing methods (which are themselves applications of AI). Together with a qualitative analysis of 55 key documents from policy, research, and industry, they manage to summarize «features of the concept of artificial intelligence as reflected in the relevant literature».

A reader’s companion to the forthcoming developments of the EU regulation.