Meta teaches an AI to lie and strategize |  TECHNOLOGY

Meta teaches an AI to lie and strategize | TECHNOLOGY

Meta trained an artificial intelligence (AI) agent to play a board game that involves chatting with other players to persuade them to support their strategy and then betray them. this The company that owns Facebook, Instagram and WhatsApp made it happen Cicero AI may have widespread applications in the near future. These include: Development of smarter virtual assistants with the combined use of technologies such as natural language processing (NLP) and strategic reasoning. This was confirmed by the Menlo Park firm via a blog post.

Following the same path, Meta reassured in a research paper published in the academic journal Science: Cicero AI achieves human-level performance in strategy board game Diplomacy. Thus, in an online league where he played 40 matches against 82 players, he managed to enter the top 10% among the participants who played more than one game.

about diplomacy

The game posed a challenge for the AI ​​agent, as winning required him to understand whether his opponents were bluffing or strategizing in a certain way to win the game. AI needed to show a certain level of empathy while playing to form collaborations with other players, something that artificial intelligences didn’t have to do before in games like chess against human opponents.

AI agents have gotten better at strategy games over the years: in 1997, software IBM’s Deep Blue beat world chess champion Gary Kasparov, and in 2016, DeepMind’s AlphaGo beat top Go player Lee Sedol. Facebook has also developed another artificial intelligence engine that can outperform humans at poker.

strategic reasoning

Cicero based on two components technology the main ones are: strategic reasoning and natural language processing. The researchers explained that while the strategic reasoning engine predicts other players’ movements and uses this information to formulate its own strategy, the natural language processing engine generates messages and analyzes responses in conversations with other players to negotiate and reach an agreement.

help artificial intelligence representative To create relevant conversations, the researchers started with the natural language generation model. 2.7 billion parameters pre-trained from texts on the Internet and fine-tuned by conversations between human players in over 40,000 games of the webDiplomasi.net. “We developed techniques to automatically annotate messages in training data with relevant planned movements in the game, so that we can control dialogue generation to discuss specific desired actions for the agent and their speaking partners at inference time,” the researchers wrote.

Meta unlocked Cicero’s code so that other researchers can improve the capabilities of the AI ​​agent. Likewise, the company has created a portal to invite users to undertake research proposals in the field of human-AI collaboration through NLP, using Diplomacy as a core axis.

looking to the future

Tech giants are racing each other to develop smarter standalone virtual assistants to support a variety of business use cases, from call centers to AI agents that can perform sentiment analysis and teach an individual new skills. According to a Fortune Business Insights report, The global market for Natural Language Processing (NLP) including those participants, It will increase from $26.4 billion in 2022 to $161.8 billion in 2029..

In this scenario, the Meta researchers seemed to suggest that Cicero’s success in diplomacy superseded the abilities of other virtual assistants available today: “Today’s AI assistants, for example, can complete simple question-and-answer tasks like telling you the room temperature, but What if they could have a long-term conversation to teach you a new skill?”.

Despite this questioning, however, Cicero is also not ready for long-term conversations, as his reasoning is strictly short-term. As the meta researchers say in their Science article: “From a strategic point of view, Cicero evaluated the dialogue only in terms of the players’ actions during the relevant turn, not modeling how their dialogue might affect the relationship with other players.”

#Meta #teaches #lie #strategize #TECHNOLOGY

Leave a Comment

Your email address will not be published. Required fields are marked *