Presentation: How is AI changing the way we do empirical science?

AI in Everyday Research — how is AI changing empirical science?Opening talk, Bilendi Academics Netzwerktreffen 2026, Trier, 20 November 2026 · Prof. Dr. Michael Bošnjak, Trier University Artificial intelligence is entering empirical research not as one tool but at every step. This talk organises the current evidence along a single research cycle of eight phases — inform, plan, preregister, conduct, document, analyze, archive, publish — and walks through it three times: for meta-analytic syntheses, for experimental studies and for (online) survey studies. For each pass, peer-reviewed studies from 2024–2026 show what already works, how well, and where it fails: LLM-assisted screening and risk-of-bias assessment in systematic reviews, hypothesis generation and the AI-written paper that passed workshop review, automated checks of registrations against papers, LLM pretesting of questionnaires, and the coding of open survey answers. Performance figures are reported with their evidence class (journal article, preprint, report, company statement) and with the limitations named by the authors. The closing part looks ahead to hybrid research teams — humans in the loop, AI agents in the workflow — with an illustrative five-layer design for a survey study, five rules for responsible use, and what this could mean for a panel provider such as Bilendi. The talk is an impulse for the afternoon, not an inventory: a snapshot of what is possible today, with the methodological caveats that a methodologist would insist on. Supplement: Study summariesA 46-page companion document provides a one-page summary of each of the 44 studies, reports and guidelines cited in the talk: research question, design and data, how AI was used, key findings, caveats and a takeaway for research practice, plus the full reference. Each summary carries an evidence badge (peer-reviewed article, preprint/working paper, report, company statement) and the slide number(s) on which the study appears; a contents table lists all studies by research-cycle phase. Figures were checked against the publications or their official abstracts (October–November 2026) and incorporate the corrections of an accuracy audit.