Can A.I. Generate New Ideas?

Artificial intelligence start-up Harmonic announced that its AI system, Aristotle, in collaboration with OpenAI’s GPT-5.2 Pro, solved a long-standing “Erdos problem,” sparking debate on AI’s role in academic research.
While some laud AI’s ability to assist in solving academic problems, critics argue that the AI’s solutions often rely on existing, human-generated research rather than generating novel insights.
Terence Tao, a renowned mathematician, likened the AI’s approach to that of a memorizing student without a deep understanding of the underlying concepts.
The debate highlights persistent questions about AI’s true capabilities in the field of research: whether it can generate new, human-unthinkable ideas or merely replicate known methods.
AI systems like GPT-5 can analyze vast amounts of information, suggesting hypotheses or experiments that experts haven’t considered, thereby streamlining research efforts.
Dr. Derya Unutmaz describes AI as a powerful tool that effectively assists in filtering useful information, although human expertise remains crucial for validation and direction.
While AI shows potential in academic and scientific fields, the consensus remains that human researchers are still key to integrating and refining AI-generated proposals.