{"673265":{"#nid":"673265","#data":{"type":"news","title":"Researchers Reveal Roadmap for AI Innovation in Brain and Language Learning","body":[{"value":"\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EOne of the hallmarks of humanity is language, but now, powerful new artificial intelligence tools also compose poetry, write songs, and have extensive conversations with human users. Tools like ChatGPT and Gemini are widely available at the tap of a button \u2014 but just how \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cem\u003E\u003Cspan\u003Esmart\u003C\/span\u003E\u003C\/em\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E are these AIs?\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EA new multidisciplinary research effort co-led by \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003E\u003Cspan\u003EAnna (Anya) Ivanova\u003C\/span\u003E\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E, assistant professor in the \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Ca href=\u0022https:\/\/psychology.gatech.edu\/\u0022\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003ESchool of Psychology\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/a\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E at Georgia Tech, alongside \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003E\u003Cspan\u003EKyle Mahowald\u003C\/span\u003E\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E, an assistant professor in the Department of Linguistics at the University of Texas at Austin, is working to uncover just that.\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003ETheir results could lead to innovative AIs that are more similar to the human brain than ever before \u2014 and also help neuroscientists and psychologists who are unearthing the secrets of our own minds.\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThe study, \u003Ca href=\u0022https:\/\/www.sciencedirect.com\/science\/article\/pii\/S1364661324000275\u0022\u003E\u201cDissociating Language and Thought in Large Language Models,\u201d\u003C\/a\u003E is published this week in the scientific journal \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cem\u003E\u003Cspan\u003ETrends in Cognitive Sciences\u003C\/span\u003E\u003C\/em\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E. The work is already making waves in the scientific community: an earlier \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Ca href=\u0022https:\/\/arxiv.org\/abs\/2301.06627\u0022\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003Epreprint\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/a\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E of the paper, released in January 2023, has already been cited more than 150 times by fellow researchers. \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThe \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003Eresearch team \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003Ehas continued to refine the research for this final journal publication.\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u201cChatGPT became available while we were finalizing the preprint,\u201d Ivanova explains. \u201cOver the past year, we\u0027ve had an opportunity to update our arguments in light of this newer generation of models, now including ChatGPT.\u201d\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Ch3\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003E\u003Cspan\u003EForm versus function\u003C\/span\u003E\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/h3\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThe study focuses on large language models (LLMs), which include AIs like ChatGPT. LLMs are text prediction models, and create writing by predicting which word comes next in a sentence \u2014 just like how a cell phone or email service like Gmail might suggest what next word you might want to write. However, while this type of language learning is extremely effective at creating coherent sentences, that doesn\u2019t necessarily signify intelligence.\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EIvanova\u2019s team argues that \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cem\u003E\u003Cspan\u003Eformal competence\u003C\/span\u003E\u003C\/em\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E \u2014 creating a well-structured, grammatically correct sentence \u2014 should be differentiated from \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cem\u003E\u003Cspan\u003Efunctional competence\u003C\/span\u003E\u003C\/em\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E \u2014 answering the right question, communicating the correct information, or appropriately communicating. They also found that while LLMs trained on text prediction are often very good at formal skills, they still struggle with functional skills.\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u201cWe humans have the\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E tendency to conflate language and thought,\u201d Ivanova says. \u201cI think that\u2019s an important thing to keep in mind as we\u0027re trying to figure out what these models are capable of, because using that ability to be good at language, to be good at formal competence, leads many people to assume that AIs are also good at thinking \u2014 even when that\u0027s not the case.\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EIt\u0027s a heuristic that we developed when interacting with other humans over thousands of years of evolution, but now in some respects, that heuristic is broken,\u201d Ivanova explains.\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EThe distinction between formal and functional competence is also vital in rigorously testing an AI\u2019s capabilities, Ivanova adds. Evaluations often don\u2019t distinguish formal and functional competence, making it difficult to assess what factors are determining a model\u2019s success or failure. The need to develop distinct tests is one of the team\u2019s more widely accepted findings, and one that some researchers in the field have already begun to implement.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Ch3\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003E\u003Cspan\u003ECreating a modular system\u003C\/span\u003E\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/h3\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EWhile the human tendency to conflate functional and formal competence may have hindered understanding of LLMs in the past, our human brains could also be the key to unlocking more powerful AIs.\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003ELeveraging the tools of cognitive neuroscience while a postdoctoral associate at \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003EMassachusetts Institute of Technology (MIT)\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E, Ivanova and her team studied brain activity in neurotypical individuals via fMRI, and used behavioral assessments of individuals with brain damage to test the causal role of brain regions in language and cognition \u2014 both conducting new research and drawing on previous studies. The team\u2019s results showed that human brains use different regions for functional and formal competence, further supporting this distinction in AIs.\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u201cOur research shows that in the brain, there is a language processing module and separate modules for reasoning,\u201d Ivanova says. This modularity could also serve as a blueprint for how to develop future AIs.\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u201cBuilding on insights from human brains \u2014 where the language processing system is sharply distinct from the systems that support our ability to think \u2014 we argue that the language-thought distinction is conceptually important for thinking about, evaluating, and improving large language models, especially given recent efforts to imbue these models with human-like intelligence,\u201d says Ivanova\u2019s former advisor and study co-author \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003E\u003Cspan\u003EEvelina Fedorenko\u003C\/span\u003E\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E, a professor of brain and cognitive sciences at MIT and a member of the McGovern Institute for Brain Research.\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EDeveloping AIs in the pattern of the human brain could help create more powerful systems \u2014 while also helping them dovetail more naturally with human users. \u201cGenerally, differences in a mechanism\u2019s internal structure affect behavior,\u201d Ivanova says. \u201cBuilding a system that has a broad macroscopic organization similar to that of the human brain could help ensure that it might be more aligned with humans down the road.\u201d\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EIn the rapidly developing world of AI, these systems are ripe for experimentation. After the team\u2019s preprint was published, OpenAI announced their intention to add plug-ins to their GPT models.\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u201cThat plug-in system is actually very similar to what we suggest,\u201d Ivanova adds. \u201cIt takes a modularity approach where the language model can be an interface to another specialized module within a system.\u201d\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EWhile the OpenAI plug-in system will include features like booking flights and ordering food, rather than cognitively inspired features, it demonstrates that \u201cthe approach has a lot of potential,\u201d Ivanova says.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Ch3\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003E\u003Cspan\u003EThe future of AI \u2014 and what it can tell us about ourselves\u003C\/span\u003E\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/h3\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EWhile our own brains might be the key to unlocking better, more powerful AIs, these AIs might also help us better understand ourselves. \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u201cWhen researchers try to study the brain and cognition, it\u0027s often useful to have some smaller system where you can actually go in and poke around and see what\u0027s going on before you get to the immense complexity,\u201d Ivanova explains.\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EHowever, since human language is unique, model or animal systems are more difficult to relate. That\u0027s where LLMs come in.\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u201cThere are lots of surprising similarities between how one would approach the study of the brain and the study of an artificial neural network\u201d like a large language model, she adds. \u201cThey are both information processing systems that have biological or artificial neurons to perform computations.\u201d\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003EIn many ways, the human brain is still a black box, but \u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003Eopenly available AIs offer a unique opportunity to see the synthetic system\u0027s inner workings and modify variables, and explore these corresponding systems like never before.\u003C\/span\u003E\u003C\/span\u003E\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\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u201c\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EIt\u0027s a really wonderful model that we have a lot of control over,\u201d Ivanova says. \u201c\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EN\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003Eeural networks \u2014 they are amazing.\u201d\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cem\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EAlong with\u0026nbsp;\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003EAnna (Anya) Ivanova, Kyle Mahowald, and Evelina Fedorenko, the\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E research team also includes\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003EIdan Blank\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E (University of California, Los Angeles), as well as \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003ENancy Kanwisher\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E and \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003EJoshua Tenenbaum\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E(Massachusetts Institute of Technology).\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/em\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u0026nbsp;\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003E\u003Cspan\u003EDOI: \u003C\/span\u003E\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Ca href=\u0022https:\/\/doi.org\/10.1016\/j.tics.2024.01.011\u0022\u003Ehttps:\/\/doi.org\/10.1016\/j.tics.2024.01.011\u003C\/a\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003E\u003Cspan\u003EResearcher Acknowledgements\u003C\/span\u003E\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cem\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EFor helpful conversations, we thank Jacob Andreas, Alex Warstadt, Dan Roberts, Kanishka Misra, students in the 2023 UT Austin Linguistics 393 seminar, the attendees of the Harvard LangCog journal club, the attendees of the UT Austin Department of Linguistics SynSem seminar, Gary Lupyan, John Krakauer, members of the Intel Deep Learning group, Yejin Choi and her group members, Allyson Ettinger, Nathan Schneider and his group members, the UT NLL Group, attendees of the KUIS AI Talk Series at Ko\u00e7 University in Istanbul, Tom McCoy, attendees of the NYU Philosophy of Deep Learning conference and his group members, Sydney Levine, organizers and attendees of the ILFC seminar, and others who have engaged with our ideas. We also thank Aalok Sathe for help with document formatting and references.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/em\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cstrong\u003E\u003Cspan\u003E\u003Cspan\u003EFunding sources\u003C\/span\u003E\u003C\/span\u003E\u003C\/strong\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cem\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EAnna (Anya) Ivanova was supported by funds from the Quest Initiative for Intelligence. Kyle Mahowald acknowledges funding from NSF Grant 2104995. Evelina Fedorenko was supported by NIH awards R01-DC016607, R01-DC016950, and U01-NS121471 and by research \u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003Efunds from the Brain and Cognitive Sciences Department, McGovern Institute for Brain Research, and the Simons Foundation through the Simons Center for the Social Brain.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/em\u003E\u003C\/p\u003E\r\n","summary":"","format":"limited_html"}],"field_subtitle":[{"value":"A new study highlights how human neuroscience is paving the way for AI innovation \u2014 and what AI can teach us about ourselves."}],"field_summary":[{"value":"\u003Cp\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003E\u003Cspan\u003EA new study co-led by School of Psychology\u0027s Anna (Anya) Ivanova uncovers the relationship between language and thought in artificial intelligence models like ChatGPT, leveraging cognitive neuroscience research on the human brain. The results are a roadmap to developing new AIs \u2014 and to better understanding how we think and communicate.\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/span\u003E\u003C\/p\u003E\r\n","format":"limited_html"}],"field_summary_sentence":[{"value":"A new study co-led by Anna (Anya) Ivanova highlights how human neuroscience is paving the way for AI innovation \u2014 and what AI can teach us about ourselves."}],"uid":"35599","created_gmt":"2024-02-29 15:37:55","changed_gmt":"2024-03-27 19:26:43","author":"sperrin6","boilerplate_text":"","field_publication":"","field_article_url":"","dateline":{"date":"2024-03-19T00:00:00-04:00","iso_date":"2024-03-19T00:00:00-04:00","tz":"America\/New_York"},"extras":[],"hg_media":{"673267":{"id":"673267","type":"image","title":"Anna (Anya) Ivanova","body":null,"created":"1709232142","gmt_created":"2024-02-29 18:42:22","changed":"1709232116","gmt_changed":"2024-02-29 18:41:56","alt":"Anna (Anya) Ivanova","file":{"fid":"256635","name":"anna-ivanova-bridge-web.jpg","image_path":"\/sites\/default\/files\/2024\/02\/29\/anna-ivanova-bridge-web.jpg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2024\/02\/29\/anna-ivanova-bridge-web.jpg","mime":"image\/jpeg","size":3192349,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2024\/02\/29\/anna-ivanova-bridge-web.jpg?itok=vU8hy_BM"}},"673258":{"id":"673258","type":"image","title":"The Intersection of AI and Cognitive Neuroscience","body":null,"created":"1709221061","gmt_created":"2024-02-29 15:37:41","changed":"1709220852","gmt_changed":"2024-02-29 15:34:12","alt":"The Intersection of AI and Cognitive Neuroscience","file":{"fid":"256626","name":"anna-ivanova-brain-lead-shot.jpg","image_path":"\/sites\/default\/files\/2024\/02\/29\/anna-ivanova-brain-lead-shot.jpg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2024\/02\/29\/anna-ivanova-brain-lead-shot.jpg","mime":"image\/jpeg","size":4544965,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2024\/02\/29\/anna-ivanova-brain-lead-shot.jpg?itok=uRkUeVBJ"}},"673259":{"id":"673259","type":"image","title":"Anna (Anya) Ivanova","body":null,"created":"1709221061","gmt_created":"2024-02-29 15:37:41","changed":"1709220852","gmt_changed":"2024-02-29 15:34:12","alt":"Anna (Anya) Ivanova","file":{"fid":"256627","name":"Anna (Anya) Ivanova.jpeg","image_path":"\/sites\/default\/files\/2024\/02\/29\/Anna%20%28Anya%29%20Ivanova.jpeg","image_full_path":"http:\/\/hg.gatech.edu\/\/sites\/default\/files\/2024\/02\/29\/Anna%20%28Anya%29%20Ivanova.jpeg","mime":"image\/jpeg","size":3316252,"path_740":"http:\/\/hg.gatech.edu\/sites\/default\/files\/styles\/740xx_scale\/public\/2024\/02\/29\/Anna%20%28Anya%29%20Ivanova.jpeg?itok=eRJg9NaG"}}},"media_ids":["673267","673258","673259"],"related_links":[{"url":"https:\/\/arxiv.org\/abs\/2303.12712","title":"Sparks of Artificial General Intelligence: Early experiments with GPT-4"},{"url":"https:\/\/www.wired.com\/story\/chatgpt-agi-intelligence\/","title":"Some Glimpse AGI in ChatGPT. Others Call It a Mirage"},{"url":"https:\/\/www.theatlantic.com\/technology\/archive\/2023\/01\/chatgpt-ai-language-human-computer-grammar-logic\/672902\/","title":"The Difference Between Speaking and Thinking"},{"url":"https:\/\/mcgovern.mit.edu\/2024\/03\/19\/researchers-reveal-roadmap-for-ai-innovation-in-brain-and-language-learning\/","title":"MIT McGovern Institute press release"},{"url":"https:\/\/liberalarts.utexas.edu\/news\/researchers-reveal-roadmap-for-ai-innovation-in-brain-and-language-learning","title":"UT Austin press release"}],"groups":[{"id":"1278","name":"College of Sciences"},{"id":"1188","name":"Research Horizons"},{"id":"443951","name":"School of Psychology"}],"categories":[{"id":"135","name":"Research"}],"keywords":[{"id":"192863","name":"go-ai"},{"id":"172970","name":"go-neuro"},{"id":"187915","name":"go-researchnews"},{"id":"193266","name":"cos-research"},{"id":"192253","name":"cos-neuro"},{"id":"192258","name":"cos-data"}],"core_research_areas":[{"id":"39431","name":"Data Engineering and Science"},{"id":"39501","name":"People and Technology"},{"id":"39541","name":"Systems"}],"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\u003E\u003Cstrong\u003EWritten by Selena Langner\u003C\/strong\u003E\u003C\/p\u003E\r\n\r\n\u003Cp\u003E\u003Cstrong\u003EEditor and Press Contact:\u003C\/strong\u003E\u003Cbr \/\u003E\r\n\u003Ca href=\u0022mailto: jess.hunt@cos.gatech.edu\u0022\u003EJess Hunt-Ralston\u003C\/a\u003E\u003Cbr \/\u003E\r\nDirector of Communications\u003Cbr \/\u003E\r\nCollege of Sciences\u003Cbr \/\u003E\r\nGeorgia Tech\u003C\/p\u003E\r\n","format":"limited_html"}],"email":["jess@cos.gatech.edu"],"slides":[],"orientation":[],"userdata":""}}}