CAMBRIDGE, Mass., October 7 — Researchers at the MIT Media Lab have pulled back the curtain on a new framework designed to measure what artificial intelligence actually does to the people who use it, not just what it outputs. The platform, called ImpactBench, launched Tuesday under the lab’s Advancing Humans with AI program alongside partners at UC Berkeley Haas, the USC Marshall Neely Center for Ethical Leadership and Decision Making, and the Psychology of Technology Institute.
A New Kind of Label
The system puts leading frontier models through their paces across 800 metrics and 26 benchmarks that span physical, psychological, and social dimensions. Results are then distilled into AI Nutrition Labels — a format deliberately modeled on the panels found on food packaging — so that anyone can see at a glance how a given system scores on avoiding harm and nurturing wellbeing. The labels flag negative behaviors such as factual hallucination, sycophancy, and toxicity, while also highlighting beneficial traits like promoting user agency and fostering genuine human flourishing.
“We’ve set out to build a rigorous, independent measure of how AI shapes human psychology — not just what it outputs, but what it does to our thinking, our autonomy, and our wellbeing,” said Juliana Schroeder, associate professor of Management of Organizations at UC Berkeley Haas and a collaborator on the project. Her words echo a concern that has been growing around kitchen tables and clinic waiting rooms alike: the technology is changing how we think, not just what we know.
Pattie Maes, professor of media arts and sciences at MIT and co-lead of the AHA program, put it more bluntly. “The fixation on existential risk of AI and sci-fi scenarios is distracting us from some of the established risks of AI deployment: we are actively losing our critical thinking ability and autonomy, and that is where our focus must urgently shift.”
Open Doors
The team is inviting researchers, domain experts, and organizations to use the platform, propose new benchmarks, validate existing ones, and help shape how results are presented. The AI Nutrition Labels and the full press release are already live on the MIT Media Lab website. As the framework evolves, the hope is that it becomes a shared reference point — something families, clinicians, and policymakers can trust when they ask the only question that really matters: what is this technology doing to the people we love?
