{"675255":{"#nid":"675255","#data":{"type":"news","title":"Meet VAL, an AI Teammate That Can Adapt to Your Tendencies","body":[{"value":"\u003Cp\u003EA team\u2019s success in any competitive environment often hinges on how well each member can anticipate the actions of their teammates.\u003C\/p\u003E\u003Cp\u003EAssistant Professor \u003Ca href=\u0022https:\/\/chrismaclellan.com\/\u0022\u003E\u003Cstrong\u003EChristopher MacLellan\u003C\/strong\u003E\u003C\/a\u003E thinks teachable artificial intelligence (AI) agents are uniquely suited for this role and make ideal teammates for video gamers.\u003C\/p\u003E\u003Cp\u003EWith the help of funding from the U.S. Department of Defense, MacLellan hopes to prove his theory with a conversational, task-performing agent he co-engineered called the Verbal Apprentice Learner (VAL).\u003C\/p\u003E\u003Cp\u003E\u201cYou need the ability to adapt to what your teammates are doing to be an effective teammate,\u201d MacLellan said. \u201cWe\u2019re exploring this capability for AI agents in the context of video games.\u201d\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EUnlike generative AI chatbots like ChatGPT, VAL uses an interactive task-learning approach.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003E\u201cVAL learns how you do things in the way you want them done,\u201d MacLellan said. \u201cWhen you tell it to do something, it will do it the way you taught it instead of some generic random way from the internet.\u201d\u003C\/p\u003E\u003Cp\u003EA key difference between VAL and a chatbot is that VAL can perceive and act within the gaming world. A chatbot, like ChatGPT, only perceives and acts within the chat dialog.\u003C\/p\u003E\u003Cp\u003EMacLellan immersed VAL into an open-sourced, simplified version of the popular Nintendo cooperative video game Overcooked to discover how well the agent can function as a teammate. In Overcooked, up to four players work together to prepare dishes in a kitchen while earning points for every completed order.\u003C\/p\u003E\u003Ch4\u003E\u003Cstrong\u003EHow Fast Can Val Learn?\u003C\/strong\u003E\u003C\/h4\u003E\u003Cp\u003EIn a study with 12 participants, MacLellan found that users could often correctly teach VAL new tasks with only a few examples.\u003C\/p\u003E\u003Cp\u003EFirst, the user must teach VAL how to play the game. Knowing that a single human error could compromise results, MacLellan designed three precautionary features:\u003C\/p\u003E\u003Cul\u003E\u003Cli\u003EWhen VAL receives a command such as \u0022cook an onion,\u0022 it asks clarifying questions to understand and confirm its task. As VAL continues to learn, clarification prompts decrease.\u003C\/li\u003E\u003Cli\u003EAn \u201cundo\u201d button to ensure users can reverse an errant command.\u003C\/li\u003E\u003Cli\u003EVAL contains GPT subcomponents to interpret user input, allowing it to adapt to ambiguous commands and typos. The GPT subcomponents drive changes in VAL\u2019s task knowledge, which it uses to perform tasks without additional guidance.\u003C\/li\u003E\u003C\/ul\u003E\u003Cp\u003EThe participants in MacLellan\u2019s study used these features to ensure VAL learned the tasks correctly.\u0026nbsp;\u003C\/p\u003E\u003Cp\u003EThe high volume of prompts creates a more tedious experience. Still, MacLellan said it provides detailed data on system performance and user experience. That insight should make designing a more seamless experience in future versions of VAL possible.\u003C\/p\u003E\u003Cp\u003EThe prompts also require the AI to be explainable.\u003C\/p\u003E\u003Cp\u003E\u201cWhen VAL learns something, it uses the language model to label each node in the task knowledge graph that the system constructs,\u201d MacLellan said. \u201cYou can see what it learned and how it breaks tasks down into actions.\u201d\u003C\/p\u003E\u003Ch4\u003E\u003Cstrong\u003EBeyond Gaming\u003C\/strong\u003E\u003C\/h4\u003E\u003Cp\u003EMacLellan\u2019s \u003Ca href=\u0022https:\/\/tail.cc.gatech.edu\/\u0022\u003E\u003Cstrong\u003ETeachable AI Lab\u003C\/strong\u003E\u003C\/a\u003E is devoted to developing AI that inexperienced users can train.\u003C\/p\u003E\u003Cp\u003E\u201cWe are trying to come up with a more usable system where anyone, including people with limited expertise, could come in and interact with the agent and be able to teach it within just five minutes of interacting with it for the first time,\u201d he said.\u003C\/p\u003E\u003Cp\u003EHis work caught the attention of the Department of Defense, which awarded MacLellan multiple grants to fund several of his projects, including VAL. The possibilities of how the DoD could use VAL, on and off the battlefield, are innumerable.\u003C\/p\u003E\u003Cp\u003E\u201c(The DoD) envisions a future in which people and AI agents jointly work together to solve problems,\u201d MacLellan said. \u201cYou need the ability to adapt to what your teammates are doing to be an effective teammate.\u003C\/p\u003E\u003Cp\u003E\u201cWe look at the dynamics of different teaming circumstances and consider what are the right ways to team AI agents with people. The key hypothesis for our project is agents that can learn on the fly and adapt to their users will make better teammates than those that are pre-trained like GPT.\u201d\u003C\/p\u003E\u003Ch4\u003E\u003Cstrong\u003EDesign Your Own Agent\u003C\/strong\u003E\u003C\/h4\u003E\u003Cp\u003EMacLellan is co-organizing a gaming agent design competition sponsored by the Institute of Electrical and Electronic Engineers (IEEE) 2024 \u003Ca href=\u0022https:\/\/2024.ieee-cog.org\/\u0022\u003E\u003Cstrong\u003EConference on Games\u003C\/strong\u003E\u003C\/a\u003E in Milan, Italy.\u003C\/p\u003E\u003Cp\u003E\u003Ca href=\u0022https:\/\/strong-tact.github.io\/\u0022\u003E\u003Cstrong\u003EThe Dice Adventure Competition \u003C\/strong\u003E\u003C\/a\u003Einvites participants to design their own AI agent to play a multi-player, turn-based dungeon crawling game or to play the game as a human teammate. The competition this month and in July offers $1,000 in prizes for players and agent developers in the top three teams.\u003C\/p\u003E","summary":"","format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003EA team\u2019s success in any competitive environment often hinges on how well each member can anticipate the actions of their teammates.\u003C\/p\u003E\u003Cp\u003EAssistant Professor \u003Ca href=\u0022https:\/\/chrismaclellan.com\/\u0022\u003E\u003Cstrong\u003EChristopher MacLellan\u003C\/strong\u003E\u003C\/a\u003E thinks teachable artificial intelligence (AI) agents are uniquely suited for this role and make ideal teammates for video gamers.\u003C\/p\u003E\u003Cp\u003EWith the help of funding from the U.S. Department of Defense, MacLellan hopes to prove his theory with a conversational, task-performing agent he co-engineered called the Verbal Apprentice Learner (VAL).\u003C\/p\u003E","format":"limited_html"}],"field_summary_sentence":[{"value":"A new AI teammate developed by Assistant Professor Christopher MacLellan could be the ideal co-opt video game partner."}],"uid":"36530","created_gmt":"2024-06-27 17:55:24","changed_gmt":"2024-07-17 14:05:01","author":"Nathan Deen","boilerplate_text":"","field_publication":"","field_article_url":"","dateline":{"date":"2024-06-27T00:00:00-04:00","iso_date":"2024-06-27T00:00:00-04:00","tz":"America\/New_York"},"extras":[],"hg_media":{"674252":{"id":"674252","type":"image","title":"VAL_86A1504-Enhanced-NR.jpg","body":null,"created":"1719510932","gmt_created":"2024-06-27 17:55:32","changed":"1719510932","gmt_changed":"2024-06-27 17:55:32","alt":"A female student wears the Meta Quest VR headset with two men standing behind her","file":{"fid":"257746","name":"VAL_86A1504-Enhanced-NR.jpg","image_path":"\/sites\/default\/files\/2024\/06\/27\/VAL_86A1504-Enhanced-NR.jpg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2024\/06\/27\/VAL_86A1504-Enhanced-NR.jpg","mime":"image\/jpeg","size":138089,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2024\/06\/27\/VAL_86A1504-Enhanced-NR.jpg?itok=Oz9nUZQO"}}},"media_ids":["674252"],"groups":[{"id":"47223","name":"College of Computing"},{"id":"50876","name":"School of Interactive Computing"},{"id":"1188","name":"Research Horizons"}],"categories":[{"id":"153","name":"Computer Science\/Information Technology and Security"}],"keywords":[{"id":"192863","name":"go-ai"},{"id":"187812","name":"artificial intelligence (AI)"},{"id":"91511","name":"Video gaming"},{"id":"2356","name":"gaming"},{"id":"187915","name":"go-researchnews"},{"id":"9153","name":"Research Horizons"}],"core_research_areas":[{"id":"193655","name":"Artificial Intelligence at Georgia Tech"}],"news_room_topics":[{"id":"71881","name":"Science and Technology"}],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Cp\u003ENathan Deen\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003ECommunications Officer\u003C\/p\u003E\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\u003Cp\u003ESchool of Interactive Computing\u003C\/p\u003E","format":"limited_html"}],"email":[],"slides":[],"orientation":[],"userdata":""}}}