{"id":848,"date":"2025-09-09T09:22:39","date_gmt":"2025-09-09T09:22:39","guid":{"rendered":"https:\/\/biconnect.psicologia.unimib.it\/?p=848"},"modified":"2025-09-14T12:55:22","modified_gmt":"2025-09-14T12:55:22","slug":"teaching-ai-like-an-apprentice-not-a-data-hoover","status":"publish","type":"post","link":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/teaching-ai-like-an-apprentice-not-a-data-hoover\/","title":{"rendered":"Teaching AI Like an Apprentice \u2014 Not a Data Hoover"},"content":{"rendered":"\n<p><strong>We\u2019re presenting at ICSR 2025 in Naples on Wednesday, 10 September 2025<\/strong> (<a href=\"https:\/\/icsr2025.eu\/?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">ICSR 2025<\/a>)<\/p>\n\n\n<div class=\"wp-block-image\">\n<figure class=\"alignright size-large\"><img fetchpriority=\"high\" decoding=\"async\" width=\"724\" height=\"1024\" src=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Pedagogically-enriched-knowledge-context-1-4-724x1024.png\" alt=\"\" class=\"wp-image-858\" srcset=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Pedagogically-enriched-knowledge-context-1-4-724x1024.png 724w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Pedagogically-enriched-knowledge-context-1-4-212x300.png 212w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Pedagogically-enriched-knowledge-context-1-4-768x1086.png 768w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Pedagogically-enriched-knowledge-context-1-4-1086x1536.png 1086w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/Pedagogically-enriched-knowledge-context-1-4-scaled.png 1810w\" sizes=\"(max-width: 724px) 100vw, 724px\" \/><figcaption class=\"wp-element-caption\">Poster presented at ICSR 2025 (Naples), Wednesday, 10 Sept 2025<\/figcaption><\/figure><\/div>\n\n\n<h2 class=\"wp-block-heading has-large-font-size\"><strong>TL;DR<\/strong><\/h2>\n\n\n\n<p>Instead of feeding models ever-bigger datasets, our new paper shows how to <em>teach<\/em> AI through structured, dialogic interactions\u2014more like mentoring an apprentice than labeling a mountain of data. In controlled experiments, pedagogical dialogues helped a learner model acquire and apply new concepts more efficiently than even an \u201cexpert\u201d with direct data access. Read the paper: (<a href=\"https:\/\/arxiv.org\/abs\/2507.21065?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">arXiv<\/a>).<\/p>\n\n\n\n<h2 class=\"wp-block-heading has-large-font-size\"><strong>Why interactive teaching matters (and why it\u2019s hard)<\/strong><\/h2>\n\n\n\n<p>Building huge datasets is slow, expensive, and often brittle. Real work is fluid: experts explain, question, and adapt on the fly. If AI systems could learn that way\u2014through back-and-forth pedagogy\u2014they\u2019d adapt faster, transfer better, and require far fewer labeled examples.<\/p>\n\n\n\n<p>But interactive teaching introduces challenges:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Structure:<\/strong> what sequence of explanations and questions best supports rapid understanding?<\/li>\n\n\n\n<li><strong>Agency:<\/strong> should the \u201cteacher\u201d lead, or should the \u201clearner\u201d drive with questions?<\/li>\n\n\n\n<li><strong>Direction:<\/strong> are top-down (category\u2192example) explanations more effective than bottom-up (example\u2192category) discovery?<\/li>\n<\/ul>\n\n\n\n<p>Our <a href=\"https:\/\/arxiv.org\/abs\/2507.21065?utm_source=chatgpt.com\" data-type=\"link\" data-id=\"https:\/\/arxiv.org\/abs\/2507.21065?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">paper<\/a> tackles these questions head-on.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>What we built and tested<\/strong><\/h2>\n\n\n\n<p>In <em>AI Pedagogy:<\/em> <em>Dialogic Social Learning for Artificial Agents (<\/em><a href=\"https:\/\/arxiv.org\/abs\/2507.21065?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">arXiv<\/a><em>)<\/em>, we set up an <strong>AI Social Gym<\/strong> where a teacher LLM with access to a ground-truth ontology interacts with a na\u00efve learner LLM using different teaching strategies (e.g., top-down vs. bottom-up; teacher-led vs. learner-led; mixed-initiative dialogues). We then evaluate learning with a 20-Questions\u2013style task: the learner must identify a secret item by asking yes\/no feature questions; fewer questions = better understanding. To avoid leakage from prior training, we used <em>made-up ontologies<\/em>.<\/p>\n\n\n\n<p><strong>Key findings (short):<\/strong><\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Pedagogical dialogues beat direct data access.<\/strong> Some strategies outperformed an \u201cexpert\u201d baseline that had the ontology outright.<\/li>\n\n\n\n<li><strong>Top-down &amp; dialogic strategies<\/strong> were the most efficient, cutting the number of questions needed to nail the answer.<\/li>\n<\/ul>\n\n\n\n<p>In other words: good teaching &#8211; the right mix of explanations <em>and<\/em> questions &#8211; helps AI learn new, structured knowledge on the fly.<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>Why this matters beyond the lab<\/strong><\/h2>\n\n\n\n<p>Across industries, organizations are scrambling to <strong>capture and transfer expert know-how<\/strong> as veterans retire and entry-level roles shrink. A few snapshots:<\/p>\n\n\n\n<figure class=\"wp-block-image alignleft size-full is-resized is-style-default\" style=\"margin-right:30px;\"><img decoding=\"async\" width=\"1024\" height=\"1024\" src=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/IMAGE-2025-09-08-62906-PM.jpg\" alt=\"\" class=\"wp-image-850\" style=\"width:327px;height:auto\" srcset=\"https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/IMAGE-2025-09-08-62906-PM.jpg 1024w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/IMAGE-2025-09-08-62906-PM-300x300.jpg 300w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/IMAGE-2025-09-08-62906-PM-150x150.jpg 150w, https:\/\/biconnect.psicologia.unimib.it\/wp-content\/uploads\/2025\/09\/IMAGE-2025-09-08-62906-PM-768x768.jpg 768w\" sizes=\"(max-width: 1024px) 100vw, 1024px\" \/><\/figure>\n\n\n\n<ul class=\"wp-block-list\">\n<li><strong>Japan\u2019s \u201cno new pupils\u201d problem:<\/strong> with an aging workforce and fewer apprentices, firms are using AI to <em>preserve and pass on veteran skills<\/em>. Hitachi, Meidensha, and NEC are piloting systems that turn tacit procedures into step-by-step guidance and training simulations. (<a href=\"https:\/\/asianews.network\/japanese-companies-use-ai-to-pass-on-veteran-workers-skills\/?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">Asia News Network<\/a>)<\/li>\n\n\n\n<li><strong>Craft meets AI:<\/strong> in Kyoto\u2019s centuries-old <em>Nishijinori<\/em> weaving, artisans collaborate with AI to generate pattern suggestions, augmenting human craft where traditional mentorship pipelines are thinning. (<a href=\"https:\/\/apnews.com\/article\/6c95395a5197ce3dd97b87afa6ac5cc7?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">AP News<\/a>)<\/li>\n\n\n\n<li><strong>Healthcare &amp; knowledge bottlenecks:<\/strong> hospitals and medical schools are adopting AI tools and curricula both to reduce paperwork and to scaffold learning, freeing experts to mentor and making expertise more transmissible. (<a href=\"https:\/\/www.pbs.org\/newshour\/science\/ai-in-health-care-could-save-lives-and-money-but-not-yet?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">PBS<\/a>,<a href=\"https:\/\/www.cbsnews.com\/news\/icahn-medical-school-nyc-chagpt-ai\/?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\"> CBS News<\/a>)<\/li>\n\n\n\n<li><strong>The broader workforce transition:<\/strong> research and industry guidance warn that retiring experts and thinner entry-level paths risk knowledge loss, unless we get serious about systematic knowledge capture and teaching tools. (<a href=\"https:\/\/www.deloitte.com\/us\/en\/insights\/topics\/talent\/strategies-for-workforce-evolution.html?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">Deloitte<\/a>)<\/li>\n<\/ul>\n\n\n\n<p>Japan\u2019s situation is particularly instructive: demographic pressure is pushing a national-scale shift toward <strong>human\u2013AI knowledge transfer<\/strong>, not just automation. That aligns with our core thesis: <em>teaching-centric AI<\/em> can help societies sustain capability when human apprentices are scarce (<a href=\"https:\/\/www.lemonde.fr\/en\/economy\/article\/2024\/03\/19\/japan-turns-to-new-technologies-to-counter-labor-shortages_6632475_19.html?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">Le Monde.fr<\/a>).<\/p>\n\n\n\n<h2 class=\"wp-block-heading\"><strong>How our approach helps<\/strong><\/h2>\n\n\n\n<p>Our results suggest that <strong>pedagogically structured, mixed-initiative dialogues<\/strong> (top-down explanations plus targeted learner questions) are a practical recipe for <em>operational knowledge transfer<\/em>. Think onboarding a service robot to a new store layout, briefing a diagnostic assistant on a hospital\u2019s local protocols, or capturing a retiring engineer\u2019s troubleshooting playbook.<\/p>\n\n\n\n<p>Concretely, this means organizations can invest in <strong>pedagogical agents<\/strong> that:<\/p>\n\n\n\n<ol class=\"wp-block-list\">\n<li>elicit key concepts and decision rules from experts,<\/li>\n\n\n\n<li>sequence explanations for rapid learner uptake, and<\/li>\n<\/ol>\n\n\n\n<p>verify understanding through dialogic questioning\u2014<em>without<\/em> assembling massive new datasets first.<\/p>\n\n\n\n<h2 class=\"wp-block-heading has-medium-font-size\"><strong>See us in Naples<\/strong>!<\/h2>\n\n\n\n<p>We\u2019ll present this work at <strong><a href=\"https:\/\/icsr2025.eu\/?utm_source=chatgpt.com\" data-type=\"link\" data-id=\"https:\/\/icsr2025.eu\/?utm_source=chatgpt.com\" target=\"_blank\" rel=\"noopener\">ICSR<\/a> 2025 (Naples), Wednesday, 10 Sept 2025<\/strong> (<a href=\"https:\/\/www.linkedin.com\/feed\/update\/urn:li:activity:7371573222200213504\" target=\"_blank\" rel=\"noopener\">LinkedIn post<\/a>). If you\u2019re at the conference, come say hi\u2014we\u2019d love to talk about applying dialogic teaching to your robots and AI tools.<\/p>\n\n\n\n<p>Stay tuned!<\/p>\n\n\n\n<p><\/p>\n","protected":false},"excerpt":{"rendered":"<p>We\u2019re presenting at ICSR 2025 in Naples on Wednesday, 10 September 2025 (ICSR 2025) TL;DR Instead of feeding models ever-bigger datasets, our new paper shows how to teach AI through structured, dialogic interactions\u2014more like mentoring an apprentice than labeling a mountain of data. In controlled experiments, pedagogical dialogues helped a learner model acquire and apply &#8230; <a title=\"Teaching AI Like an Apprentice \u2014 Not a Data Hoover\" class=\"read-more\" href=\"https:\/\/biconnect.psicologia.unimib.it\/index.php\/teaching-ai-like-an-apprentice-not-a-data-hoover\/\" aria-label=\"Read more about Teaching AI Like an Apprentice \u2014 Not a Data Hoover\">Read more<\/a><\/p>\n","protected":false},"author":4,"featured_media":850,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[38,43],"tags":[],"class_list":["post-848","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-events","category-lab-news"],"acf":[],"_links":{"self":[{"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/posts\/848","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/users\/4"}],"replies":[{"embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/comments?post=848"}],"version-history":[{"count":16,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/posts\/848\/revisions"}],"predecessor-version":[{"id":967,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/posts\/848\/revisions\/967"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/media\/850"}],"wp:attachment":[{"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/media?parent=848"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/categories?post=848"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/biconnect.psicologia.unimib.it\/index.php\/wp-json\/wp\/v2\/tags?post=848"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}