[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"branding":3,"analytics":7,"article-transformer-self-prior-drives-mirror-test-behavior-in-simulated-infant":10},{"siteName":4,"siteTagline":5,"publisherName":4,"contactEmail":6},"The Revision","Tech news, decoded.","editor@therevision.news",{"gaMeasurementId":8,"adsenseClientId":9},"G-ZW2MV82GYR","ca-pub-8533917693782264",{"article":11},{"id":12,"slug":13,"title":14,"dek":15,"body_md":16,"tags_json":17,"published_at":18,"created_at":19,"updated_at":20,"status":21,"review_note":22,"review_notes":23,"image_url":22,"persona_id":22,"persona_name":22,"section":22,"tags":24,"sources":28,"feedback":32,"feedback_at":22,"cost_usd":32,"total_tokens":32},1375,"transformer-self-prior-drives-mirror-test-behavior-in-simulated-infant","Transformer self-prior drives mirror-test behavior in simulated infant","A new model shows self-awareness-like actions emerging from a single self‑prior mechanism, without rewards, in a mirror‑mark task.","A computational model makes a simulated infant remove a sticker from its own face in a mirror.\n\nThe authors trained a Transformer to learn the density of familiar visual‑proprioceptive inputs, calling this the self‑prior. When a novel mark disrupts the learned distribution, the model’s expected free energy spikes, prompting active‑inference‑driven movements that bring the mark into view and then remove it. In the simulation, the infant succeeded in about 70 % of trials, despite having no tactile feedback or external reward. Expected free energy fell significantly after the sticker was removed, confirming the self‑prior acted as an internal self‑vs‑non‑self criterion.\n\nThe work suggests that a single probabilistic body schema can generate self‑recognition‑like behavior, sidestepping the need for hand‑crafted reward signals. It links the free‑energy principle to early developmental cognition, offering a computational bridge between AI models and classic mirror‑self tests.\n\nIf the approach scales, it could challenge current theories that rely on complex reinforcement structures, but it remains limited to a highly controlled simulation.","[\"artificial-intelligence\",\"cognitive-science\",\"self-awareness\"]","2026-06-16T04:00:00.000Z","2026-06-17T06:32:44.440Z","2026-06-17T06:32:47.259Z","published",null,[],[25,26,27],"artificial-intelligence","cognitive-science","self-awareness",[29],{"name":30,"url":31},"arXiv cs.AI","https:\u002F\u002Farxiv.org\u002Fabs\u002F2604.09673",0]