Replicating science with AI teammates
Inherent, founded by DeepMind alumni, has released Faraday, described as an AI teammate whose core capability is replicating published scientific papers. This is more than a novelty; if robust, it could accelerate literature review, replication studies, and hypothesis testing across disciplines. The prospect raises questions about the integrity of automated replication, especially around provenance, data sources, and evaluation criteria. The implications for peer review are profound: will journals and conferences adopt standardized replication checks powered by AI agents? And what about the risk of misinterpretation if an agent misreads a methodological nuance? The tech community is likely to see a spectrum of use cases—from automated extraction of experimental setups and datasets to generating annotated summaries and even drafting replication plans for independent labs. On the flip side, replication by AI creates an opacity problem: how do researchers audit a machine-generated replication, ensure adherence to preregistered methods, and verify that the conclusions are faithful to the original work? The Faraday project could accelerate scientific progress while forcing the ecosystem to rethink norms around authorship, reproducibility, and accountability. Investors and policy makers will also watch with interest as AI-driven scientific workflows become a normal part of the research lifecycle. In short, Faraday may represent a watershed moment for AI-assisted science, provided the community can establish robust standards and transparent evaluation frameworks.