Predicting Unseen Scientific Interactions: Conditional Flow Models Hallucinate Non-Additive Effects
Year: 2026
Type: preprint
Abstract: Conditional generative models are increasingly used for scientific prediction beyond the direct training support. Yet scientific questions often concern whether one intervention changes the effect of another. Such interactions can be of primary interest while explaining little of the total variation, allowing models to appear accurate overall without recovering the relevant scientific effect. It remains unclear how these models learn interactions and when these successfully extrapolate to unseen conditions. We therefore study interaction recovery in flow-matching models using a physiological time-series simulator, drug dose-response data, combinatorial cell perturbations, and molecular conformer generations, where we separate recovery from hallucinated interaction structure. In the controlled setting, individual effects are learned much earlier than interactions and models hallucinate interactions where none exist. Across settings, we find qualitatively different generalization problems, with successful prediction depending on whether condition structure continues from observed to held-out targets. When it does, interactions are recoverable. If not, models can generate plausible responses while missing the interaction. We furthermore find that hallucinations increase as targets become less identifiable, and that successful interaction recovery in output space can fail under nonlinear readouts. Finally, we show that aggregate fidelity metrics can conceal whether models learn and extrapolate interactions.
Source: arXiv (Cornell University)
Authors: Gijsen, Sam, Seiler, Moritz, Ritter, Kerstin
Institution: University of Tübingen
Language: English
Subfield: Statistical and Nonlinear Physics
Field: Physics and Astronomy
Domain: Physical Sciences
Open Access status: green
Awards: G2074461085, SFB 1404: FONDA – Grundlagen von Workflows für die Analyse großer naturwissenschaftlicher Daten, FOR 5187: Personalisierte Psychotherapie für Patient*innen mit fehlendem Behandlungserfolg: Mechanismen, prädiktive Marker und klinische Anwendung, G3986097662, FOR 5363: KIFOR Integration von Deep Learning und Statistik zum Verständnis strukturierter biomedizinischer Daten, TRR 265: Verlust und Wiedererlangung der Kontrolle über den Drogenkonsum: Von Trajektorien über Mechanismen bis hin zu Interventionen, EXC 2064: Maschinelles Lernen: Neue Perspektiven für die Wissenschaft, G8142800921
OpenAlex ID: W7222361793