What can the latest AI do for science ?
The latest development is the emergence of systems that can orchestrate parts of the scientific workflow:
• literature synthesis
• hypothesis generation
• recursive critique
• integration across fragmented knowledge domains
This may speed up incremental discovery, especially in fields already rich in structured data such as computational biology, chemistry, and drug repurposing.
👉 Yet, true scientific autonomy remains out of reach for AI
Current systems remain heavily dependent on:
- existing literature distributions
- prevailing conceptual frameworks
- human-defined objectives
That creates a risk of epistemic homogenization: many researchers using similar AI systems may converge toward similar hypotheses and fashionable paradigms.
The most difficult aspects of science remain weakly automated:
- recognizing when assumptions are wrong
- identifying meaningful anomalies
- reframing questions
- exercising experimental judgment
⬛ AI can help science as long as we recognize both its capabilities and limitations.