AI in Research: Is It Quietly Deskilling Scientists?
- Lille My
- Jun 24
- 4 min read
Welcome to this week's scan of how AI is reshaping research — curated from Nature, Science and the leading business and social-science journals.
The deskilling question: is AI eroding the abilities it was meant to amplify?
For two years the dominant story about AI in research has been one of acceleration: more papers, faster analyses, lower barriers. This week Nature turned the lens around and asked what we might be losing. In a feature bluntly titled "Is AI ruining our skills? Early results are in — and they're not good", the magazine surveys emerging evidence that heavy reliance on AI assistants can erode the very competencies — critical reading, problem-solving, careful writing — that the tools are supposed to support. The worry is not that AI produces bad output, but that outsourcing the cognitive work atrophies the human judgement needed to catch when it does.
That individual-level concern has a collective twin. In "Will AI spark a scientific renaissance — or a diffuse monoculture?", Nature weighs whether broad AI adoption broadens discovery or quietly funnels the field toward the same crowded, high-reward questions — a convergence that can look like productivity while shrinking intellectual diversity. The two anxieties reinforce each other: if researchers lean on the same models to decide what to study and how to study it, both individual skill and collective pluralism can thin out at once. Notably, some fields are already responding — Nature reports that mathematicians are developing explicit rules for AI use, and argues other disciplines should follow rather than wait for norms to emerge by accident.
Why it matters: None of this is an argument against using AI — the productivity gains are real and, for most researchers, unavoidable. It is an argument for using it deliberately. Treat AI as leverage for tasks you already understand well enough to audit, keep doing the hard cognitive work that builds and maintains expertise, and — at the level of your lab or field — protect the odd, unfashionable questions the models will never nudge you toward. The researchers who thrive will be the ones who let AI amplify their judgement rather than replace it.
More from this week
An autonomous medical AI agent that acts inside the health record
Researchers introduce MIRA (Medical Intelligence for Reasoning and Action), an autonomous AI agent that operates within a sandboxed electronic health record with governed access to patient data and the ability to initiate permitted EHR actions under defined safety constraints. Rather than only answering questions in a chat box, it is designed to manage patient cases — a concrete step toward 'physician copilots' that work inside clinical workflows. The authors frame it as a test of whether such systems can reach physician-level performance while staying within safety guardrails.
How AI-generated concretizations shape human creativity
This Information Systems Research study examines 'agency configurations' in human–AI ideation — the fine-grained ways creative work is split between person and model. It tests how representational differences in AI-generated inputs, specifically textual versus visual concretizations of an idea, influence the creativity of the resulting ideas and the cognitive effort the human ideator expends to refine them. The finding that the form of AI assistance, not just its presence, shapes outcomes is directly relevant to anyone designing AI tools for innovation or brainstorming.
Measuring what makes an idea creative — with embeddings
An idea is more than the sum of its words or ingredients; what makes it original is how the elements combine. This Journal of Marketing paper uses representation-learning methods — word embeddings — to quantify the internal coherence among the components of a creative idea and links that coherence to how the idea is evaluated and whether it succeeds. Using a large-scale dataset of over 57,000 online recipes as a testbed, it offers marketers and researchers a scalable, data-driven way to measure creativity that previously relied on human judges.
Does making AI evaluation transparent actually help?
As organizations fold AI into traditionally human evaluation processes like recruitment and performance review, the complexity of the algorithms often makes them opaque, limiting users' understanding and potentially inducing stress and behavioral change. Transparency is widely advocated as the fix. This Information Systems Research study examines user behavior and downstream effects under algorithmic evaluation and finds the benefits of transparency are not automatic — a useful caution for any organization rolling out AI-assisted assessment.
The 'AI spine': an org design for scaling GenAI value
MIT Sloan Management Review profiles a new kind of internal AI organization some companies call the 'AI spine' — a cross-functional structure that rapidly develops generative-AI use cases with users' help. It enables sharing of ideas and expertise across business units, sparking ideas about where GenAI can improve processes organization-wide, while disciplined project governance keeps resources focused where a positive impact is most likely. A practical template for research-adjacent organizations struggling to move past pilots.
That's this week. Forward it to a colleague who's wondering whether AI is sharpening or dulling their craft — and explore AI tools for your own research at gaiforresearch.com.
Comments