A Quest for AI Knowledge

This paper examines how artificial intelligence (AI) tools that excel at interpolating within known domains reshape scientists’ research strategies. Extending the Carnehl and Schneider (2025) framework, we show that AI’s impact on research novelty is non-monotonic: scientists ignore limited AI, constrain their ambition to match moderate AI capabilities, and pursue more novel research only with sufficiently advanced AI. This pattern emerges from a fundamental complementarity—scientists strategically “work to the AI” to maximise the value of AI-enabled perfect decisions. We identify critical capability thresholds where private research incentives align with social optimality, particularly when AI operational range exceeds the social planner’s optimal research distance without AI. Under these conditions, scientific knowledge evolves through uniform “stepping stone” expansions, with discoveries positioned precisely at the AI’s capability boundary. Our findings reveal that AI neither uniformly promotes nor inhibits research ambition but fundamentally restructures how scientists balance novelty and reliability, with implications for science policy in an AI-enhanced research landscape.