Home » Meta researchers create AI that masters Diplomacy, tricking human players

Meta researchers create AI that masters Diplomacy, tricking human players

by admin
Expanding / online game screenshot diplomacyprovided by researchers at Cicero, including a live chat dialog.

Tuesday, Meta AI announced Development of Cicero, which claims to be the first AI to achieve human-level performance in a strategic board game diplomacyThis is a remarkable achievement, as the game requires deep interpersonal negotiation skills, and this means that Cicero has acquired the specific proficiency in the language required to win the game.

Before Deep Blue beat Garry Kasparov at chess 1997the board game is useful means AI achievements. In 2015, when AlphaGo broke another barrier defeated Please master Lee Sedol. Both of these games follow a relatively clear set of analytical rules (although Go’s rules are typically simplified for computer AI).

but, diplomacy, A large part of the gameplay involves social skills. Players must show empathy, use natural language, and build relationships to win. This is a difficult task for the computer he player. With this in mind, Meta asked: “Can we build more effective and flexible agents that can use language to negotiate, persuade, and work with people to achieve strategic goals in the same way humans do?”

According to Meta, the answer is yes. Cicero learned his skills by playing the online version of the game. diplomacy upon web Diplomacy.netOver time, it became a master of the game, reportedly achieving “more than twice the average score” of human players and ranking in the top 10% of those who played multiple games. rice field.

To create Cicero, Meta put together AI models for strategic reasoning (similar to AlphaGo) and natural language processing (similar to AlphaGo). GPT-3), and combined them into one agent. During each game, Cicero looks at the state of the game board and conversation history to predict how other players will behave. He creates plans to execute through a language model that can generate human-like dialogue, allowing him to coordinate with other players.

Meta が提供する、<em>Diplomacy</em> Block diagram of Cicero, a bot that plays .  ” src=”https://cdn.arstechnica.net/wp-content/uploads/2022/11/Visual_UnderTheHood-640×907.jpg” width=”640″ height=”907″ srcset=”https://cdn.arstechnica .net/wp-content/uploads/2022/11/Visual_UnderTheHood-1280×1813.jpg 2x”/></a><figcaption class=
Expanding / Cicero block diagram diplomacy– Playbot powered by Meta.

Meta AI

Meta calls Cicero’s natural language skills a “controllable interaction model.” This is the core of Cicero’s personality. Like GPT-3, Cicero draws from his large corpus of internet text scraped from the web. “We started with 2.7 billion parameters to build a controllable interaction model. bartLanguage models like those pre-trained on texts from the internet and fine-tuned on over 40,000 human games on webDiplomacy.net describe meta.

The resulting model mastered the complexity of complex games. “For example, Cicero can speculate that he will need certain player support later in the game,” says Mehta. from their particular point of view. “

Meta’s Cicero Study Appeared Titled Science magazine, “Human-Level Play in the Game of Diplomacy by Combining Language Models and Strategic Reasoning.”

As for broader applications, Meta says Cicero’s work could potentially “relax communication barriers” between humans and AI, such as sustaining long conversations to teach someone a new skill. suggesting. Or you could power a video game where NPCs can talk like humans, understand the player’s motivations, and adapt along the way.

At the same time, this technology could be used in some circumstances to manipulate people by impersonating them and deceiving them in dangerous ways. Along these lines, Meta hopes other researchers can build its code “in a responsible manner” and takes steps to detect and remove “toxic messages in this new domain.” says he is teaching. ingested—always risk For large language models.

provided by meta Detailed site Explained how Cicero works and open sourced Cicero’s code on GitHub. Online diplomacy Fans, and maybe even the rest of us, might need to take note.

You may also like

Leave a Comment