The UK AI Security Institute, alongside the government's new Future of Work Unit, ran a randomised controlled trial with 500 participants testing AI augmentation on four O*NET-derived workplace tasks. On average, AI access produced a 19% improvement in quality scores, a 25% reduction in time taken, and a 61% improvement in points per minute, though results were heterogeneous: three tasks showed significant gains while one (open-ended strategic planning) showed none. The study also flagged small sample size as a constraint, calling the findings preliminary. For financial institutions, this is one of the few government-run RCTs generating experimental data on AI productivity at the task level, providing a potential anchor for economy-wide exposure estimates that could inform workforce and automation risk modelling.
340 other tracked developments touching the same claims.