{"668351":{"#nid":"668351","#data":{"type":"news","title":"New Chef Dataset Brings AI to Cooking","body":[{"value":"\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EArtificial intelligence (AI) can help people shop, plan, and write \u2014 but not cook. It turns out humans aren\u2019t the only ones who have a hard time following step-by-step recipes in the correct order, but new research from the Georgia Institute of Technology\u2019s College of Computing could change that.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EResearchers created a dataset called ChattyChef, which uses natural language processing models that can help a user cook a recipe. Using the open-source large language model GPT-J, ChattyChef\u2019s dataset of cooking dialogues follows recipes with the user. \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThe researchers presented their AI in the paper, \u201c\u003Ca href=\u0022https:\/\/arxiv.org\/abs\/2305.17280\u0022\u003E\u003Cspan\u003E\u003Cspan\u003EImproved Instruction Ordering in Recipe-Grounded Conversation\u003C\/span\u003E\u003C\/span\u003E\u003C\/a\u003E,\u201d presented at the 61st annual meeting of the \u003Ca href=\u0022https:\/\/2023.aclweb.org\/\u0022\u003E\u003Cspan\u003E\u003Cspan\u003EAssociation for Computational Linguistics\u003C\/span\u003E\u003C\/span\u003E\u003C\/a\u003E. \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EAlthough other researchers have theorized about the possibility of an AI chef, Georgia Tech\u2019s work pushes the field forward. \u201cWe are one of the first research teams to analyze the challenges of using large language models for building an AI chef,\u201d said Duong Le, a Ph.D. student in the School of Interactive Computing. \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EMost attempts at using language models for cooking fail because GPT-J doesn\u2019t understand what the user wants to do next, or user intent, and has difficultly tracking how far the user is in the recipe \u2014 what the researchers call the \u201cstate of the conversation.\u201d It also can\u2019t easily answer clarification questions, like about ingredient amounts or cooking times.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EFor example, maybe someone is trying to cook hashbrowns. The AI tells them to melt butter in the pan and add the potatoes. The user then asks about the next step. A bad bot might jumble the order and tell them to serve the hashbrown even though they haven\u2019t finished cooking it yet. Or a user asks a follow-up question about how long to cook the hashbrown, and AI won\u2019t be precise enough, instead giving a general time and not specifying the cooking time for each side.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EWith this in mind, the researchers ensured their model had two key features:\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cul\u003E\r\n\t\u003Cli\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EUser intent detection to determine the user\u2019s current intent within a fixed set of possibilities, such as \u201cAsk for next instruction\u201d or \u201cAsk for details about ingredients.\u201d\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/li\u003E\r\n\t\u003Cli\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EInstruction state tracking to identify which recipe step the user is on, which works with 80% accuracy. \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/li\u003E\r\n\u003C\/ul\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThe combined information from these features supports the third innovation of ChattyChef \u2014 response generation. User intent helps generate the best response to answer a user\u2019s question. The instruction state selects the most relevant parts of the recipe rather than including the entire recipe, to avoid confusing the user or burdening them with extra steps as they are cooking.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThe ChattyChef dataset is built off WikiHow recipes with positive ratings and fewer than eight steps. The researchers crowdsourced people to role play how they might use ChattyChef to determine what instructions would be best to include in the dataset. \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThe researchers believe the innovations of ChattyChef could be used in many domains besides cooking, such as repair manuals or software documentation. \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EDuong Minh Le, Ruohao Guo, Wei Xu, and Alan Ritter. 2023. Improved instruction ordering in recipe-grounded conversation. arXiv preprint arXiv:2305.17280.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThis research is supported in part by the National Science Foundation awards IIS-2112633 and IIS-2052498.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n","summary":"","format":"limited_html"}],"field_subtitle":"","field_summary":[{"value":"\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EResearchers created a dataset called ChattyChef, which uses natural language processing models that can help a user cook a recipe. Using the open-source large language model GPT-J, ChattyChef\u2019s dataset of cooking dialogues follows recipes with the user.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n","format":"limited_html"}],"field_summary_sentence":[{"value":"Researchers created a dataset called ChattyChef, which uses natural language processing models that can help a user cook a recipe."}],"uid":"34541","created_gmt":"2023-07-05 14:56:02","changed_gmt":"2023-07-05 14:57:30","author":"Tess Malone","boilerplate_text":"","field_publication":"","field_article_url":"","dateline":{"date":"2023-07-05T00:00:00-04:00","iso_date":"2023-07-05T00:00:00-04:00","tz":"America\/New_York"},"extras":[],"hg_media":{"671097":{"id":"671097","type":"image","title":"GettyImages-1430305488.jpg","body":"\u003Cp\u003ECourtesy of Getty Images\u003C\/p\u003E\r\n","created":"1688568997","gmt_created":"2023-07-05 14:56:37","changed":"1688568997","gmt_changed":"2023-07-05 14:56:37","alt":"Woman chopping peppers in front of laptop","file":{"fid":"254101","name":"GettyImages-1430305488.jpg","image_path":"\/sites\/default\/files\/2023\/07\/05\/GettyImages-1430305488.jpg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2023\/07\/05\/GettyImages-1430305488.jpg","mime":"image\/jpeg","size":9990317,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2023\/07\/05\/GettyImages-1430305488.jpg?itok=WPCIgGB6"}}},"media_ids":["671097"],"groups":[{"id":"1214","name":"News Room"},{"id":"1188","name":"Research Horizons"},{"id":"47223","name":"College of Computing"},{"id":"50876","name":"School of Interactive Computing"}],"categories":[],"keywords":[{"id":"187915","name":"go-researchnews"}],"core_research_areas":[],"news_room_topics":[],"event_categories":[],"invited_audience":[],"affiliations":[],"classification":[],"areas_of_expertise":[],"news_and_recent_appearances":[],"phone":[],"contact":[{"value":"\u003Cp\u003ETess Malone, Senior Research Writer\/Editor\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Ca href=\u0022mailto:tess.malone@gatech.edu\u0022\u003Etess.malone@gatech.edu\u003C\/a\u003E\u003C\/p\u003E\r\n","format":"limited_html"}],"email":[],"slides":[],"orientation":[],"userdata":""}}}