Do we actually need to write research papers at all?
If we are going to lean on the creativity and the cross-domain thinking of human researchers, then surely that should be where the human time goes.
More issues
BioSingularity. The substrate determines the ceiling
Biology becoming computationally tractable depends less on model size than on the data substrate underneath. The ceiling is infrastructure.
Ground the model, or it invents the evidence
arXiv now bans authors for unchecked LLM output. The real fix is trusted, curated, subject-specific models rather than more policy.
The new era of going “Fast and Far” in research
A Nobel Prize in Chemistry went, in effect, to software. The old trade-off between small fast teams and large slow ones has stopped holding.
What parts of the academic knowledge creation & dissemination pipeline can we automate with AI?
Four working prototypes testing which parts of the research pipeline, from writing to review to publishing, can actually be automated.
OpenAccess.ai: How Cheap Can We Make Academic Publishing?
Article processing charges run to thousands per paper. OpenAccess.ai tests how much of the editorial stack can genuinely be automated.
Preprints.ai: How Much of Peer Review Can We Automate?
A large share of peer review is mechanical checking. Preprints.ai tests how much of it can be scored automatically, at preprint scale.
OpenScience.ai: Using Open Data to Generate Research That Doesn't Yet Exist
What claims are already latent in ClinVar, GTEx, STRING and Open Targets that nobody has written up yet? OpenScience.ai explores that gap.
FAIRdata.ai: Making Open Data FAIR-er
Most repository data is technically FAIR and practically unusable. FAIRdata.ai is an experiment in closing that gap automatically.
Moving from "share data because you should" to "share data because you'll get something back."
Allen AI's AutoDiscovery generated its own hypotheses from an open dataset. That return on sharing is exactly why I founded Figshare.