Defining intelligence for AI research

Characterize intelligence with a precise, operational definition that can guide and evaluate research aimed at creating machine systems exhibiting genuine intelligence, including work in natural language processing and large language models.

Background

In reviewing the origins of artificial intelligence at the 1956 Dartmouth conference and the aim of creating "genuine intelligence," the article situates modern NLP and LLM developments within longstanding debates about what constitutes intelligence. This foundational uncertainty affects how researchers set goals, assess progress, and justify methods across different AI paradigms.

The authors explicitly acknowledge that the field still lacks a settled definition of intelligence, underscoring a core conceptual gap that complicates both scientific evaluation and broader legal and ethical considerations of AI capabilities.