Science without scientists? In defence of human understanding
Year: 2026
Type: preprint
Abstract: Science has two grand goals, producing results and increasing human understanding, which have usually advanced together. Accelerating science by automating around human limits threatens to decouple them, as AI research output can scale with compute whereas humans can only read and digest new research at the pace biology allows, and some AI-generated results may be too opaque for anyone to follow. I argue that this could create a comprehension lag, a growing gap between what is discovered and what humans understand. This lag may begin inside individual labs, as scientists become "commissioners" directing their own teams of AI agents, but could extend to entire fields, as the frontier moves faster than people can follow. I discuss three costs of a comprehension lag: a loss of control over AI-accelerated research, a loss of capacity as aspects of training and apprenticeship are automated, and a loss of meaning that scientists find in their work. I then propose five measures to mitigate the comprehension lag, including new research directions for the cognitive and behavioural sciences, and argue that human understanding deserves protecting, both to keep oversight of AI-driven research and as an end in itself.
Source: PsyArXiv (OSF Preprints)
Author: Jonathan Prunty
Institution: University of Kent
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225898343
