Centuries ago, the future of diplomacy was questioned technological process – the invention of the radio, the telegraph and public interference in foreign policy. Meta’s recent AI model revived the conversation. CICERO has become the talk of the tech town and international chatbot diplomacy can’t be far behind.
Diplomacy is an art that even the most powerful AI couldn’t master until CICERO appeared. Inspired by the famous Roman orator Cicero, Meta AI showed its model that can beat several people in the Diplomacy board game, which requires strategic planning and verbal negotiations with several other players. The work, researchers say, could pave the way for virtual practice coaches and dispute mediators.
Speak with Analytics India magazineone of the researchers, Athul Paul Jacobshared his story of how Meta AI managed to build a model for diplomacy with human-level performance.
Jacob got involved in AI through his mentor, Joshua Bengio, the winner of the 2018 Turing Award. Describing the creation of CICERO, Jacob said: “I started my PhD at MIT in 2019, and Noam Brown (the project lead) talked to my advisor about the great challenge of diplomacy. He had a successful contribution to previous academic papers on poker solving. And so everyone wondered what his next thing was? And the conclusion was diplomacy.”
Over the decades, conquests have piled up, with AI agents devising ways to defeat mortals in games of all kinds. Jacob said: “All the games that AI has tackled have been much simpler. That’s when Brown came up with this project and brought in some of Meta’s most brilliant engineers. The goal of this research was to find a way to build systems that can interact with people in different systems and environments.
Jacob became involved in the summer of 2020 when the team focused on a diplomatic variant where communication was not allowed. “It is not called press diplomacy that focuses on the purely strategic aspects of the game. There is communication, but only through moves on the board, so it’s very different. My background in language systems was interesting for the team.”
Building dialogue systems that negotiate and coordinate with other human players was Jacob’s main interest. He initially worked on the two main components of the model: the language component and the dialogue component.
In his first year he devoted a lot of time to the dialogue component. The team built a prototype that could communicate and do things, but it didn’t do very well. So a year later, in 2021, Jacob joined another part of the team working on the strategic elements. One of the things the team needed was to consider human rationality.
Diplomacy versus dialogue
The Diplomacy game has numerous challenges. According to Jacob, the first task was to figure out how to communicate with 76 other players in language, since you had previously studied that setting. The other was to understand how to collaborate with people of different skill levels – and, finally, to decipher a system that spitting nonsense like the other interlocutors in the domain today.
According to Jacob, the main difference is that none of the current systems can handle human modeling, which is where the strategic component comes in. Most of these systems repeat what people say. Essentially it looks human, but can’t reason much like how humans would think or identify what a human would say in response.
Bringing the theory of mind He said: “Our strategic component can identify what people will do in response, so most of these chatbots don’t rely on that. And that is a weakness of the current systems that are being used.”
Jacob thinks there is still a lot to do, especially with regard to the dialogue component. The model still says nonsensical things that need to be clarified to fit the setting. “We’ve done a lot of work to fix that, but I think there’s more work to be done,” he added.
When AlphaGo was released, it was very specific to the game of Go, and then they expanded it to other board games, but then people thought that was the end of it. But it turns out that the technique has many applications, such as the Looking for a tree in Monte Carlo who uses the algorithm for 40 use cases.
“Our ultimate goal was to solve diplomacy, but that is not the end of the story. We knew that it would take a lot of components to develop a new technique. And so those different components are the ones that can be used responsibly, from code generation to proofing theorems in the future. he concluded.