The Eleven Types of AI Every Researcher Now Works With

Ask a researcher in 2020 whether they used AI and most would have said no. Today the honest answer is four or five kinds — often without calling them AI at all. They are profoundly different technologies, and their failure modes differ too. Treating “AI in research” as one thing is how researchers get burned.

11 categories

01 Literature discovery & search Proven Finds, ranks and answers questions about published work - from keyword ranking to agents that read papers and cite an answer. Risk Confident citations to papers that do not say what is claimed. 0 tools 02 3D prediction: structures & materials Proven Predicts protein, nucleic-acid, complex and materials structure from sequence alone. The category with a Nobel Prize. Risk A confidence score is not experimental validation. 0 tools 03 Virtual organisms & simulation Frontier Models living systems - knock out a gene in silico and get a predicted phenotype before anyone touches a pipette. Risk Predicted phenotypes are hypotheses, not results. 0 tools 04 Synthesis & knowledge management Emerging Digests the papers you already have: grounded Q&A over your library, extraction, screening and lab memory. Risk Summaries flatten the caveats the original authors were careful about. 0 tools 05 Data analysis, quant + qual Emerging Turns spreadsheets, images and transcripts into analysis - generated code on one side, semantic coding on the other. Risk Generated code runs without error and still answers the wrong question. 0 tools 06 Writing & editing Proven Drafts, tightens and translates scientific prose, from abstracts and figure legends to cover letters. Risk Fluent text that overstates what the data supports. 0 tools 07 Peer review & integrity Emerging Screens submissions for image manipulation, plagiarism, statistical error and undisclosed generated text. Risk A false accusation does as much damage as missed misconduct. 0 tools 08 Grant writing & screening Emerging Drafts proposals, matches funding calls to a lab’s actual work, and screens applications at the funder end. Risk Proposals that read well and promise what the lab cannot deliver. 0 tools 09 Lab automation & self-driving labs Emerging Robots, schedulers and closed-loop systems that design, run and interpret experiments with less human intervention. Risk A closed loop optimising the wrong objective, very efficiently. 0 tools 10 AI lab managers: LIMS, ELN & ops Emerging Sample tracking, inventory, protocols and compliance records, with AI both reading and writing the record. Risk Automated records nobody checks are an audit failure in waiting. 0 tools 11 AI co-scientists Frontier Agents that propose hypotheses, plan experiments and iterate on results across the whole research cycle. Risk Demos outrun deployment. Accountability stays human. 0 tools