CICERO from Meta AI outperforms 90% of human Diplomacy players.

A common criticism of artificial intelligence is that it lacks sentience and emotion. Artificial intelligence (AI) is more widely used to remedy human limitations, such as increasing production and efficiency, which enables technology and robots to help produce more in less time. But Meta’s most recent revelation reveals a new aspect of AI that includes it collaborating with people in the strategy game Diplomacy.

In the game of diplomacy, participants collaborate by exchanging goals and viewpoints in order to forge alliances and plans. Because diplomacy takes so much understanding and language at the level of a human, it has long been thought that AI could never succeed at it, according to Meta. The introduction of CICERO, an AI bot that has demonstrated exceptional diplomatic and negotiating abilities with humans, is an effort to instil strategic thinking in AI. Meta information:

“Diplomacy is a game where players are the focus as opposed to pieces, unlike games like Chess and Go. An agent will swiftly lose the game if it is unable to discern whether a player is most likely bluffing or whether another player would view a particular move as aggressive. Similar to this, it won’t locate other players who are prepared to cooperate with it if it doesn’t sound like a genuine person—demonstrating empathy, forming connections, and speaking intelligently about the game.”

CICERO lays the stage for advancements in AI, which can produce personal assistants that work with people through reasoning and natural language processing. Currently, CICERO can only be used to play Diplomacy; webDiplomacy.net was the site where the game was played. In addition, it scored more than twice as well as the typical human player and was in the top 10% of players who participated in multiple games.

Meta stated that other players in Diplomacy preferred allying with the AI rather than other people since CICERO’s ability to talk in natural language is so powerful. Thus, Meta discusses CICERO’s potential to develop into an AI helper that people may converse with for a long time while simultaneously gaining new skills.

According to Percy Liang, director of Stanford University’s Institute for Human-Centered AI’s Center for Research on Foundation Models:

  • “For the AI research ecosystem to be healthy, open science and open source are essential. With this release, Meta AI is continuing its culture of openness, which will enable researchers to expand on this work.

Researchers are encouraged to submit proposals in response to Meta’s Request for Proposal (RFP) in the area of natural language-based human-AI cooperation. You may read more about the technology underlying CICERO here.

Christopher Woodill

About ME

Enterprise technology leader for the past 15+ years…certified PMP, Six Sigma Black Belt and TOGAF Enterprise Architect. I collaborate with companies to help align their strategic objectives with concrete implementable technology strategies. I am Vice President, Enterprise Solutions for Klick Health.

Leave a Comment