BU linguist warns AI reliance could erode scientists' own creative edge

On November 11, Najoung Kim, a language scientist at Boston University, laid out her concerns about AI dependence affecting research ideation in a multi-post thread, sparking discussion in the AI for Science community.

Confirmed

• She does not strongly doubt the proposition that AI could produce breakthrough-sparking ideas in arbitrary research fields; her concern lies with the consequences, not the capability itself
• As of today, pure AI ideation using public commercial models without substantial human input yields unimpressive results; but she can reasonably foresee the approach becoming effective enough in the near future to consistently produce ideas above roughly the human median
• She further notes that "no worse than the median" sounds fine, but median-level research output has limited real-world impact: a field's collective progress comes from exploring the entire population of ideas and filtering out the few truly influential ones
• Her core worry is that in the future, humans may collectively atrophy in their ability to generate creative ideas due to reliance on AI

Unconfirmed

• On mitigation, she suggests methods like "expanding sampling volume" might alleviate the homogenization of AI-generated ideas, but she is unsure whether the compute costs of such fixes are affordable for researchers outside commercial AI providers; moreover, providers could make these capabilities inaccessible at any time. These points were raised as open questions, not conclusions

Why it matters

• The thread separates "whether AI can generate good ideas" from "whether humans should retain the ability to generate good ideas," pointing out that even with continually improving AI performance, human capability loss and idea homogenization remain independent risks
• Compute costs and access restrictions mean that if fixes rely on large-scale sampling, independent science's capacity for idea generation could become further concentrated in the hands of a few commercial AI vendors

2026-11-11 ~ 2026-11-11 · 6 related posts

• [source] Najoung Kim: not doubting AI's creative ideas, but fearing human deskilling — najoungkim · 2026-11-11
• Najoung Kim: AI-only ideation isn't great yet, but may soon beat the median human idea — najoungkim · 2026-11-11
• Najoung Kim: an 'above median' research idea still isn't impactful — najoungkim · 2026-11-11
• Najoung Kim warns AI dependence in ideation may cause 'convergence to above median' research — najoungkim · 2026-11-11
• [source] Najoung Kim: fixes for AI ideation homogenization may be priced out of academics' reach — najoungkim · 2026-11-11
• Linguist Najoung Kim warns commercial AI providers could monopolize independent scientific progress — najoungkim · 2026-11-11