Publikation
Towards a Technical Definition of Artificial Intelligence and Machine Learning Software Systems
Lars Nolle
In: Artificial Intelligence XLII. SGAI International Conference on Artificial Intelligence (AI-2025), 45th SGAI International Conference on Artificial Intelligence, December 16-18, Cambridge, United Kingdom, Pages 334-340, Lecture Notes in Computer Science, ISBN 978-3-032-11401-3, Springer Nature Switzerland, Cham, 12/2025.
Zusammenfassung
The widespread adoption of artificial intelligence (AI) has generated excitement but also concerns about related risks, leading to new regulatory efforts such as the EU AI Act. Defining AI for legal purposes remains difficult, as common behavioural definitions rely on comparing systems to human intelligence, which is itself not clearly defined. Listing specific AI methods is also ineffective due to rapid innovation. To address this, this work surveys the “AI zoo”, i.e. the broad range of techniques found in AI, from symbolic logic to machine learning (ML), and proposes a taxonomy based on core abilities like search, optimisation, and learning. It argues that the fundamental element uniting AI systems is heuristic search in large spaces. Thus, it is proposed to define an AI system as a software that comprises of search heuristics to find solutions in vast search spaces. A machine learning system is defined as software that uses AI to adapt a generic model to a specific task. The findings should be taken as a starting point for discussion amongst the academic community in order to come up with better technical definitions for AI and ML software systems.