Protecting Cognitive Integrity: Our internal AI use policy (V1)

The GPAI Policy Lab, led by Tom David, has published V1 of its internal AI use policy, designed to protect cognitive integrity amid increasing daily AI use. The policy, originally in French, addresses concerns that frequent interaction with AI systems can compromise reasoning and judgment. It includes hard restrictions, such as a ban on delegating value judgments to AI, and warning signals to monitor for cognitive erosion. The lab argues that individual willpower is insufficient, advocating for shared norms and concrete policies to preserve core capacities.

The policy distinguishes between acceptable uses, like asking AI for alternative arguments or perspectives, and forbidden ones, like relying on AI for moral evaluations. It also outlines warning signals, such as reduced ability to form independent opinions or over-reliance on AI for complex tasks. The lab invites counterarguments and experiences from other organizations to refine the policy, emphasizing that the cost of over-caution is lower than under-caution. This initiative aims to spark broader conversation and best practices in AI safety and policy communities.