Small AI models can now negotiate like large ones — if you let them feel emotions

Researchers built a system that lets smaller, cheaper language models handle high-stakes negotiations by treating emotional decisions as strategic choices rather than reactions. This matters because smaller models run on phones and robots without sending data to the cloud, which means negotiation AI can now work in privacy-sensitive settings like emergency response or healthcare without the cost and security risks of large models.

The core problem was simple: small models are fast and private, but bad at negotiation because they don't understand emotional dynamics. Large models understand emotion but cost too much to run locally and leak data. This paper claims to solve that tradeoff by letting multiple specialized small models vote on emotional strategy in real time, with a coordinator keeping track of which model is most reliable. If this actually works at scale, it means negotiation AI moves from cloud-dependent to edge-deployable. The constraint was deployment cost and privacy risk; the change is that constraint loosens.