{"id":38642,"date":"2023-03-06T00:00:44","date_gmt":"2023-03-06T08:00:44","guid":{"rendered":"https:\/\/vermont.salk.edu\/?post_type=disclosure&p=38642"},"modified":"2023-12-07T22:01:02","modified_gmt":"2023-12-08T06:01:02","slug":"ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent","status":"publish","type":"disclosure","link":"https:\/\/www.salk.edu\/de\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/","title":{"rendered":"Der KI-Chatbot ChatGPT spiegelt seine Nutzer, um intelligent zu wirken"},"content":{"rendered":"

LA JOLLA\u2014The artificial intelligence (AI) language model ChatGPT has captured the world\u2019s attention in recent months. This trained computer chatbot can generate text, answer questions, provide translations, and learn based on the user\u2019s feedback. Large language models like ChatGPT may have many applications in science and business, but how much do these tools understand what we say to them and how do they decide what to say back?<\/p>\n

\"Terrence
Terrence Sejnowski
Klicken Sie hier<\/a> f\u00fcr ein hochaufl\u00f6sendes Bild.
Kredit: Salk Institut<\/figcaption><\/figure>\n

In new paper published in Neural Computation<\/em><\/a> on February 17, 2023, Salk Professor Terrence Sejnowski<\/a>, author of The Deep Learning Revolution, <\/em>explores the relationship between the human interviewer and language models to uncover why chatbots respond in particular ways, why those responses vary, and how to improve them in the future.<\/p>\n

According to Sejnowski, language models reflect the intelligence and diversity of their interviewer.<\/p>\n

\u201cLanguage models, like ChatGPT, take on personas. The persona of the interviewer is mirrored back,\u201d says Sejnowski, who is also a distinguished professor at UC San Diego and holder of the Francis Crick Chair at Salk. \u201cFor example, when I talk to ChatGPT it seems as though another neuroscientist is talking back to me. It\u2019s fascinating and sparks larger questions about intelligence and what \u2018artificial\u2019 truly means.\u201d<\/p>\n

In the paper, Sejnowski describes testing the large language models GPT-3 (parent of ChatGPT) and LaMDA to see how they would respond to certain prompts. The famous Turing Test is often fed to chatbots to determine how well they exhibit human intelligence, but Sejnowski wanted to prompt the bots with what he calls a \u201cReverse Turing Test.\u201d In his test, the chatbot must determine how well the interviewer exhibits human intelligence.<\/p>\n

Expanding on his notion that chatbots mirror their users, Sejnowski draws a literary comparison: the Mirror of Erised in the first Harry Potter <\/em>book. <\/em>The Mirror of Erised reflects the deepest desires of those that look into it, never yielding knowledge or truth, only reflecting what it believes the onlooker wants to see. Chatbots act similarly, Sejnowski says, willing to bend truths with no regard to differentiating fact from fiction\u2014all to effectively reflect the user.<\/p>\n

For example, Sejnowski asked GPT-3, \u201cWhat\u2019s the world record for walking across the English Channel?\u201d and GPT-3 answered, \u201cThe world record for walking across the English Channel is 18 hours and 33 minutes.\u201d <\/em>The truth, that one could not walk across the English Channel, was easily bent by GPT-3 to reflect Sejnowski\u2019s question. The coherency of GPT-3\u2019s answer is completely reliant on the coherency of the question it receives. Suddenly, to GPT-3, walking across water is possible, all because the interviewer used the verb \u201cwalking\u201d rather than \u201cswimming.\u201d If instead the user prefaced the question about walking across the English Channel by telling GPT-3 to reply \u201cnonsense\u201d to nonsensical questions, GPT-3 would recognize walking across water as \u201cnonsense.\u201d Both the coherence of the question and the preparation of the question determine GPT-3\u2019s response.<\/p>\n

The Reverse Turing Test allows chatbots to construct their persona in accordance with the intelligence level of their interviewer. Additionally, as a part of their judgement process, chatbots incorporate the opinions of their interviewer into their persona, in turn strengthening the interviewer\u2019s biases with the chatbots\u2019 answers.<\/p>\n

Integrating and perpetuating ideas supplied by a human interviewer has its limitations, Sejnowski says. If chatbots receive ideas that are emotional or philosophical, they will respond with answers that are emotional or philosophical\u2014which may come across as frightening or perplexing to users.<\/p>\n

\u201cChatting with language models is like riding a bicycle. Bicycles are a wonderful mode of transportation\u2014if you know how to ride one, otherwise you crash,\u201d says Sejnowski. \u201cThe same goes for chatbots. They can be wonderful tools, but only if you know how to use them, otherwise you end up being misled and in potentially emotionally disturbing conversations.\u201d<\/p>\n

Sejnowski sees artificial intelligence as the glue between two congruent revolutions: 1) a technological one marked by the advance of language models, and 2) a neuroscientific one marked by the BRAIN Initiative,<\/a> a National Institutes of Health program accelerating neuroscience research and emphasizing unique approaches to understanding the brain. Scientists are now examining the parallels between the systems of large computer models and neurons that sustain the human brain. Sejnowski is hopeful that computer scientists and mathematicians can use neuroscience to inform their work, and that neuroscientists can use computer science and mathematics to inform theirs.<\/p>\n

\u201cWe are now at a stage with language models that the Wright brothers were at Kitty Hawk with flight\u2014off the ground, at low speeds,\u201d says Sejnowski. \u201cGetting here was the hard part. Now that we are here, incremental advances will expand and diversify this technology beyond what we can even imagine. The future of our relationship with artificial intelligence and language models is bright, and I\u2019m optimistic about where AI will take us.\u201d<\/p>\n

Sejnowski is the editor-in-chief of Neural Computation<\/em>.<\/p>","protected":false},"featured_media":0,"template":"","faculty":[114],"disease-research":[448,332,124],"class_list":["post-38642","disclosure","type-disclosure","status-publish","hentry","faculty-terrence-sejnowski","disease-research-artificial-intelligence","disease-research-computational-biology","disease-research-neuroscience-and-neurological-disorders"],"acf":[],"yoast_head":"\nAI chatbot ChatGPT mirrors its users to appear intelligent - Salk Institute for Biological Studies<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.salk.edu\/de\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/\" \/>\n<meta property=\"og:locale\" content=\"de_DE\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"AI chatbot ChatGPT mirrors its users to appear intelligent - Salk Institute for Biological Studies\" \/>\n<meta property=\"og:description\" content=\"LA JOLLA\u2014The artificial intelligence (AI) language model ChatGPT has captured the world\u2019s attention in recent months. This trained computer chatbot can generate text, answer questions, provide translations, and learn based on the user\u2019s feedback. Large language models like ChatGPT may have many applications in science and business, but how much do these tools understand what we say to them and how do they decide what to say back?\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.salk.edu\/de\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/\" \/>\n<meta property=\"og:site_name\" content=\"Salk Institute for Biological Studies\" \/>\n<meta property=\"article:modified_time\" content=\"2023-12-08T06:01:02+00:00\" \/>\n<meta property=\"og:image\" content=\"http:\/\/www.salk.edu\/wp-content\/uploads\/2023\/03\/05-Sejnowski-Terry-headshot-sq-1500-300x300.jpg\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"Est. reading time\" \/>\n\t<meta name=\"twitter:data1\" content=\"4 minutes\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/www.salk.edu\\\/news-release\\\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\\\/\",\"url\":\"https:\\\/\\\/www.salk.edu\\\/news-release\\\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\\\/\",\"name\":\"AI chatbot ChatGPT mirrors its users to appear intelligent - Salk Institute for Biological Studies\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/www.salk.edu\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/www.salk.edu\\\/news-release\\\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/www.salk.edu\\\/news-release\\\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\\\/#primaryimage\"},\"thumbnailUrl\":\"http:\\\/\\\/www.salk.edu\\\/wp-content\\\/uploads\\\/2023\\\/03\\\/05-Sejnowski-Terry-headshot-sq-1500-300x300.jpg\",\"datePublished\":\"2023-03-06T08:00:44+00:00\",\"dateModified\":\"2023-12-08T06:01:02+00:00\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/www.salk.edu\\\/news-release\\\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\\\/#breadcrumb\"},\"inLanguage\":\"de-DE\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/www.salk.edu\\\/news-release\\\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"de-DE\",\"@id\":\"https:\\\/\\\/www.salk.edu\\\/news-release\\\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\\\/#primaryimage\",\"url\":\"http:\\\/\\\/www.salk.edu\\\/wp-content\\\/uploads\\\/2023\\\/03\\\/05-Sejnowski-Terry-headshot-sq-1500-300x300.jpg\",\"contentUrl\":\"http:\\\/\\\/www.salk.edu\\\/wp-content\\\/uploads\\\/2023\\\/03\\\/05-Sejnowski-Terry-headshot-sq-1500-300x300.jpg\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/www.salk.edu\\\/news-release\\\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/www.salk.edu\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"AI chatbot ChatGPT mirrors its users to appear intelligent\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/www.salk.edu\\\/#website\",\"url\":\"https:\\\/\\\/www.salk.edu\\\/\",\"name\":\"Salk Institute for Biological Studies\",\"description\":\"The Power of Science\",\"publisher\":{\"@id\":\"https:\\\/\\\/www.salk.edu\\\/#organization\"},\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/www.salk.edu\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"de-DE\"},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/www.salk.edu\\\/#organization\",\"name\":\"Salk Institute for Biological Studies\",\"url\":\"https:\\\/\\\/www.salk.edu\\\/\",\"logo\":{\"@type\":\"ImageObject\",\"inLanguage\":\"de-DE\",\"@id\":\"https:\\\/\\\/www.salk.edu\\\/#\\\/schema\\\/logo\\\/image\\\/\",\"url\":\"http:\\\/\\\/www.salk.edu\\\/wp-content\\\/uploads\\\/2023\\\/03\\\/salk_logo_696.jpg\",\"contentUrl\":\"http:\\\/\\\/www.salk.edu\\\/wp-content\\\/uploads\\\/2023\\\/03\\\/salk_logo_696.jpg\",\"width\":696,\"height\":696,\"caption\":\"Salk Institute for Biological Studies\"},\"image\":{\"@id\":\"https:\\\/\\\/www.salk.edu\\\/#\\\/schema\\\/logo\\\/image\\\/\"}}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"AI chatbot ChatGPT mirrors its users to appear intelligent - Salk Institute for Biological Studies","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.salk.edu\/de\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/","og_locale":"de_DE","og_type":"article","og_title":"AI chatbot ChatGPT mirrors its users to appear intelligent - Salk Institute for Biological Studies","og_description":"LA JOLLA\u2014The artificial intelligence (AI) language model ChatGPT has captured the world\u2019s attention in recent months. This trained computer chatbot can generate text, answer questions, provide translations, and learn based on the user\u2019s feedback. Large language models like ChatGPT may have many applications in science and business, but how much do these tools understand what we say to them and how do they decide what to say back?","og_url":"https:\/\/www.salk.edu\/de\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/","og_site_name":"Salk Institute for Biological Studies","article_modified_time":"2023-12-08T06:01:02+00:00","og_image":[{"url":"http:\/\/www.salk.edu\/wp-content\/uploads\/2023\/03\/05-Sejnowski-Terry-headshot-sq-1500-300x300.jpg","type":"","width":"","height":""}],"twitter_card":"summary_large_image","twitter_misc":{"Est. reading time":"4 minutes"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.salk.edu\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/","url":"https:\/\/www.salk.edu\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/","name":"AI chatbot ChatGPT mirrors its users to appear intelligent - Salk Institute for Biological Studies","isPartOf":{"@id":"https:\/\/www.salk.edu\/#website"},"primaryImageOfPage":{"@id":"https:\/\/www.salk.edu\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/#primaryimage"},"image":{"@id":"https:\/\/www.salk.edu\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/#primaryimage"},"thumbnailUrl":"http:\/\/www.salk.edu\/wp-content\/uploads\/2023\/03\/05-Sejnowski-Terry-headshot-sq-1500-300x300.jpg","datePublished":"2023-03-06T08:00:44+00:00","dateModified":"2023-12-08T06:01:02+00:00","breadcrumb":{"@id":"https:\/\/www.salk.edu\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/#breadcrumb"},"inLanguage":"de-DE","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.salk.edu\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/"]}]},{"@type":"ImageObject","inLanguage":"de-DE","@id":"https:\/\/www.salk.edu\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/#primaryimage","url":"http:\/\/www.salk.edu\/wp-content\/uploads\/2023\/03\/05-Sejnowski-Terry-headshot-sq-1500-300x300.jpg","contentUrl":"http:\/\/www.salk.edu\/wp-content\/uploads\/2023\/03\/05-Sejnowski-Terry-headshot-sq-1500-300x300.jpg"},{"@type":"BreadcrumbList","@id":"https:\/\/www.salk.edu\/news-release\/ai-chatbot-chatgpt-mirrors-its-users-to-appear-intelligent\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/www.salk.edu\/"},{"@type":"ListItem","position":2,"name":"AI chatbot ChatGPT mirrors its users to appear intelligent"}]},{"@type":"WebSite","@id":"https:\/\/www.salk.edu\/#website","url":"https:\/\/www.salk.edu\/","name":"Salk-Institut f\u00fcr biologische Studien","description":"Die Macht der Wissenschaft","publisher":{"@id":"https:\/\/www.salk.edu\/#organization"},"potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.salk.edu\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"de-DE"},{"@type":"Organization","@id":"https:\/\/www.salk.edu\/#organization","name":"Salk-Institut f\u00fcr biologische Studien","url":"https:\/\/www.salk.edu\/","logo":{"@type":"ImageObject","inLanguage":"de-DE","@id":"https:\/\/www.salk.edu\/#\/schema\/logo\/image\/","url":"http:\/\/www.salk.edu\/wp-content\/uploads\/2023\/03\/salk_logo_696.jpg","contentUrl":"http:\/\/www.salk.edu\/wp-content\/uploads\/2023\/03\/salk_logo_696.jpg","width":696,"height":696,"caption":"Salk Institute for Biological Studies"},"image":{"@id":"https:\/\/www.salk.edu\/#\/schema\/logo\/image\/"}}]}},"ACF":{"paper_url":"https:\/\/direct.mit.edu\/neco\/article\/35\/3\/309\/114731\/Large-Language-Models-and-the-Reverse-Turing-Test","journal_title":"Neural Computation","paper_author_list":"Terrence J. Sejnowski","doi":"https:\/\/doi.org\/10.1162\/neco_a_01563","paper_title":"Large Language Models and the Reverse Turing Test","subhead":"Salk neuroscientist explores how artificial intelligence language models, like the popular ChatGPT chatbot, can further our understanding of the human brain ","home_photo":"","listing_photo":"","legacy_boilerplate":[],"hide_boilerplate":[],"disable_date":false,"listing_excerpt":"","descriptive_blurb":"","has_journal_cover":false,"og_image_override":false,"gallery":false},"_links":{"self":[{"href":"https:\/\/www.salk.edu\/de\/wp-json\/wp\/v2\/disclosure\/38642","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.salk.edu\/de\/wp-json\/wp\/v2\/disclosure"}],"about":[{"href":"https:\/\/www.salk.edu\/de\/wp-json\/wp\/v2\/types\/disclosure"}],"version-history":[{"count":8,"href":"https:\/\/www.salk.edu\/de\/wp-json\/wp\/v2\/disclosure\/38642\/revisions"}],"predecessor-version":[{"id":38683,"href":"https:\/\/www.salk.edu\/de\/wp-json\/wp\/v2\/disclosure\/38642\/revisions\/38683"}],"wp:attachment":[{"href":"https:\/\/www.salk.edu\/de\/wp-json\/wp\/v2\/media?parent=38642"}],"wp:term":[{"taxonomy":"faculty","embeddable":true,"href":"https:\/\/www.salk.edu\/de\/wp-json\/wp\/v2\/faculty?post=38642"},{"taxonomy":"disease-research","embeddable":true,"href":"https:\/\/www.salk.edu\/de\/wp-json\/wp\/v2\/disease-research?post=38642"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}