Human-level play with CICERO
Meta AI is the artificial intelligence laboratory of the company formerly known as Facebook. It is known for developing tools for face recognition, de-identification (removing identifying information from images), music translation (converting original music to alternative instruments and styles), and whole-body pose estimation (localizing the body, face, hands, and feet in 2D content).
Its greatest achievement is CICERO, a Meta AI agent that combines language models with strategic reasoning to negotiate, persuade, and cooperate with humans. It can achieve human-level play in the game of Diplomacy by inferring and anticipating other players’ actions based on conversations with them.
Who can use Meta AI?
Meta AI is currently used for research purposes only. Meta experts are trying to make their AI systems more efficient. Alongside projects like the CICERO AI tool, they are also working on MultiRay. This is a platform for running AI models at scale to simplify and speed up various tasks, from topic tagging to hate speech detection.
What can CICERO do?
As the mainstay of Meta AI laboratories, CICERO is essentially an AI bot that can plan its moves ahead, maintain a dialog, build trust, and progress within a game of diplomacy. It does this without succumbing to human vulnerabilities like rivalry and envy.
Diplomacy bot that outsmarts humans
As an AI game competitor endowed with perspective and a hefty playbook, CICERO impresses with its ability to impersonate humans and maintain a dialog with other players through hundreds of chat messages per game. However, it is unable to self-play. By virtue of its programming, it’s incapable of manipulating gameplay and furthering its strategy by deceiving its rivals through dialog.
