In a blog post, Meta says that the AI will be helpful in creating modern virtual assistant technology. It leverages both strategic reasoning and natural language processing (NLP). Cicero can deliver human-level performance when playing the Diplomacy board game. The AI was competing in an online league across 40 games against 82 human players. By the end, the model was able to rank in the top 10% of players. If you are unfamiliar with Diplomacy, it is a game where seven players compete to attempt to control Europe. However, it is a strategy game and not a war game. Each turn involves players negotiating to gain support for their plans before then trying to put those plans into action. Doing these plans without support will end in failure. In Diplomacy, it is important to be able to convince other players, whether through truth or deception.
How it Works
Meta says to be able to compete, the Cicero AI needed to be able to understand the nuances of players. For example, if they were bluffing or outright lying. It needed to be able to display empathy to form alliances with other players. The AI was built on natural language processing (NLP) and strategic reasoning. The latter technology uses an engine to predict player moves and helps the AI to build a strategy based on the predictions. As for NLP, it is an engine that generates messages and analyzes data to help form negotiation patterns. “We developed techniques to automatically annotate messages in the training data with corresponding planned moves in the game, so that at inference time we can control dialogue generation to discuss specific desired actions for the agent and its conversation partners,” researchers said in a more detailed blog post. Meta says the Cicero AI code is now available in open source on Microsoft’s GitHub platform. Tip of the day: Need to reduce picture size of several images, but don’t have the time to edit every one? Microsoft’s PowerToys image resizer can batch-resize your photos with just two clicks .
