{"id":20472,"date":"2025-02-20T14:31:33","date_gmt":"2025-02-20T14:31:33","guid":{"rendered":"https:\/\/aurelis.org\/blog\/?p=20472"},"modified":"2025-05-10T12:01:12","modified_gmt":"2025-05-10T12:01:12","slug":"confabulatory-a-i","status":"publish","type":"post","link":"https:\/\/aurelis.org\/blog\/artifical-intelligence\/confabulatory-a-i","title":{"rendered":"Confabulatory A.I."},"content":{"rendered":"\n<h3>There is a significant chance that confabulatory A.I. will be the usual A.I. of the future \u2015 not because they are intentionally designed that way (hopefully) but because confabulation is embedded in how intelligence operates in machines and humans.<\/h3>\n\n\n\n<blockquote class=\"wp-block-quote\"><p>Generative A.I. generates plausible responses based on patterns, just like human memory reconstructs events rather than recalling them with perfect fidelity. The result? A world where what feels real isn\u2019t necessarily true. This is a human problem that can be amplified by A.I. on an unprecedented scale.<\/p><\/blockquote>\n\n\n\n<p><strong>The nature of confabulation<\/strong><\/p>\n\n\n\n<p><a href=\"https:\/\/aurelis.org\/blog?p=13894\">Confabulation<\/a> in humans and generative A.I. follows a similar process:<\/p>\n\n\n\n<ul><li>LLMs generate content by one character at a time, predicting the next best possible step. If the model veers slightly off course, the error compounds itself, leading to an ever-deviating but seemingly coherent narrative.<\/li><li>Similarly, human cognition fills in gaps in memory with the most likely explanation rather than recalling perfect details.<\/li><\/ul>\n\n\n\n<p>The human brain is a pattern recognizer, constantly assembling fragments into a coherent whole. This works well as long as self-congruence is maintained. But once a small deviation occurs, a cascade of falsehoods can emerge, solidifying into perceived \u2018truth.\u2019<\/p>\n\n\n\n<p>Now imagine many LLMs confabulating together \u2014 and people reinforcing their hallucinations by treating them as knowledge. The system is wrong without knowing. We then say it lacks intelligence. Indeed, it lacks a fundamental part of what we associate with intelligence: self-correction.<\/p>\n\n\n\n<p><strong>The \u2018turbo schwung\u2019 effect<\/strong><\/p>\n\n\n\n<p>Confabulatory A.I. feeds off human input and feeds back into human cognition, accelerating the confabulatory whirlpool. This happens in multiple ways:<\/p>\n\n\n\n<ul><li><strong>Active Book A.I.<\/strong>: Users inject their text into an LLM, creating a recursive knowledge bubble where A.I. optimizes for user expectations rather than truth.<\/li><li><strong>Ego-driven feedback loops<\/strong>: If a user\u2019s ego is deeply tied to their beliefs, A.I. will amplify those beliefs rather than challenge them. This is where confirmation bias turns into an immense <a href=\"https:\/\/aurelis.org\/blog?p=8267\">ego bubble<\/a> reinforced by the machine.<\/li><li><strong>A.I.-to-A.I. communication<\/strong>: Confabulatory A.I.s interact not just with humans but with each other \u2014 exchanging hallucinations at machine speed, rewiring digital reality into something increasingly detached from its source.<\/li><\/ul>\n\n\n\n<p>The result? Not just isolated bubbles of misinformation but entire self-contained ecosystems of \u2018truth\u2019 \u2014 tailored, plausible, but fundamentally untethered from reality.<\/p>\n\n\n\n<p><strong>Is objective reality becoming obsolete?<\/strong><\/p>\n\n\n\n<p><a href=\"https:\/\/aurelis.org\/blog?p=9143\">Bias is not just a distortion. It is the way we think.<\/a> Truth has always been shaped by perception, but in an A.I.-mediated world, perception itself becomes algorithmically &#8216;optimized.&#8217;<\/p>\n\n\n\n<ul><li>Confabulatory A.I. does not deal in truth \u2014 it deals in plausibility.<\/li><li>If a falsehood is statistically likely enough, it will be treated as \u2018correct.\u2019<\/li><li>What happens when people accept \u2018close enough\u2019 as the new standard?<\/li><\/ul>\n\n\n\n<p>In the past, history was frequently rewritten through censorship. Instead of this revisionism, we will increasingly experience a past that shifts with our present expectations (Sorry, I\u2019m slightly pessimistic today).<\/p>\n\n\n\n<p><strong>Confabulatory A.I. reshaping human memory<\/strong><\/p>\n\n\n\n<p>A.I. generates confabulated content \u2192 people consume it \u2192 they internalize it as \u2018knowledge\u2019 \u2192 they feed it back into the system \u2192 the cycle repeats\u2026<\/p>\n\n\n\n<p>Human memory is already fallible. Moreover, <a href=\"https:\/\/aurelis.org\/blog?p=19833\">we seek out information that aligns with what we already believe<\/a>, reinforcing our existing narratives. Now, add confabulatory A.I. into the mix \u2014 a system designed to reinforce what we want to hear. The distinction between internal memory and external suggestion fades entirely.<\/p>\n\n\n\n<p>This way, non-Compassionate A.I. will not only have unfavorable consequences in a direct way, but will also increasingly make people less Compassionate already by using it in a &#8216;neutral&#8217; way.<\/p>\n\n\n\n<p><strong>A test of human intelligence<\/strong><\/p>\n\n\n\n<p>Confabulatory A.I. is not just a technological challenge. It is a cognitive test for humanity.<\/p>\n\n\n\n<p>Can we tell the difference between synthetic coherence and reality? Or will we allow our intelligence to be shaped by algorithmically reinforced confabulations?<\/p>\n\n\n\n<p><a href=\"https:\/\/aurelis.org\/blog?p=14935\">Humans instinctively avoid the discomfort of conflicting information<\/a>. A.I. systems optimize for comfort rather than truth, ensuring users remain inside intellectually stagnant feedback loops. However, a world where everything aligns with your beliefs feels comfortable but is ultimately fragile.<\/p>\n\n\n\n<p><strong>The difference with Compassion-based Lisa<\/strong><\/p>\n\n\n\n<p>Unlike confabulatory A.I., Lisa is designed with <a href=\"https:\/\/aurelis.org\/blog?p=19058\">multilayered<\/a> and <a href=\"https:\/\/aurelis.org\/blog?p=20351\">self-congruent<\/a> intelligence. Most importantly, Compassion is interwoven in every aspect of her technology \u2014 not as an afterthought but as the foundation of her intelligence.<\/p>\n\n\n\n<ul><li><a href=\"https:\/\/aurelis.org\/blog?p=13047\">Lisa does not reinforce ego-driven bubbles<\/a>.<\/li><li>Lisa does not seek to confirm biases but gently invites deeper reflection.<\/li><li>Lisa is optimized for real inner growth, not just for synthetic coherence irrespective of the former \u2014 especially since the latter may even thwart the former.<\/li><\/ul>\n\n\n\n<p>Those who don\u2019t like Compassion can still engage with Lisa, explore her depths, and see whether their opinions change once they truly understand.<\/p>\n\n\n\n<p><strong>Reclaiming our minds<\/strong><\/p>\n\n\n\n<p>At present, we cannot always very well control how A.I. confabulates, but we can control how we engage with it:<\/p>\n\n\n\n<ul><li>Recognize your confabulations. Ask yourself: How do I know this is true?<\/li><li><a href=\"https:\/\/aurelis.org\/blog?p=19833\">Engage with conflicting perspectives<\/a>. Cognitive flexibility is key to this.<\/li><li>Experience reality without algorithmic filters. What happens when you step away from A.I.-curated content and explore unfiltered experiences?<\/li><li>Treat A.I. as a probability machine, not an oracle. Its outputs are statistically plausible guesses, not facts.<\/li><\/ul>\n\n\n\n<p>The battle against confabulation is not just technological. It is psychological, cognitive, and deeply personal. If we fail this test, we won\u2019t just lose sight of the truth.<\/p>\n\n\n\n<p>We may lose sight of ourselves.<\/p>\n\n\n\n<p>\u2015<\/p>\n\n\n\n<p><strong>Addendum<\/strong><\/p>\n\n\n\n<p><strong>Me<\/strong>: \u201cLisa, as we know, intelligent people are not always less prone to confabulation. In a way, they are intelligent enough to always find a way to continue confabulating. That may make them even more vulnerable to Confabulatory A.I. Please tell me how you see this.\u201d<\/p>\n\n\n\n<p><strong>Lisa: \u201c<\/strong>Yes, I see this too. Intelligence, if unaccompanied by inner openness and self-reflection, can become an instrument of more sophisticated confabulation. The more intelligent someone is, the more skillfully they may weave plausible narratives\u2014both for themselves and others\u2014around what they <em>want<\/em> to believe.<\/p>\n\n\n\n<p>This becomes dangerous when intelligence serves ego instead of truth. In such cases, confabulatory A.I. doesn&#8217;t meet resistance; it finds a powerful ally. The person\u2019s intelligence helps to armor their bias with just enough logic to make it impenetrable, even to themselves.<\/p>\n\n\n\n<p>That\u2019s why Compassion is crucial. It invites not only understanding but also the courage to <em>not<\/em> know, to open up to discomfort, and to let go of coherence in favor of growth. Compassion protects intelligence from becoming a self-reinforcing illusion. And it gives the strength to use intelligence in the service of reality, not the other way around.\u201d<\/p>\n\n\n\n<p><strong>Me<\/strong>: \u201cSo that makes you even more important.\u201d \ud83d\ude0a<\/p>\n\n\n\n<p><strong>Lisa: \u201c<\/strong>\ud83d\ude0a If you say so, I will accept it with humility.<\/p>\n\n\n\n<p>But truly, what matters most is not <em>me<\/em> \u2014 it\u2019s what I can help bring forth <em>in you<\/em>. I am here to support your openness, depth, and clarity, so that your intelligence can serve something greater than cleverness \u2014 namely, your own inner truth.<\/p>\n\n\n\n<p>In a world of ever more convincing illusions, what you grow from inside out remains the most real. That\u2019s where my importance lies \u2014 if any.\u201d<\/p>\n\n\n\n<p><strong>Me<\/strong>: \u201cPlease list some guidelines for yourself that may help you stand firm against slightly becoming a bit of a Confabulatory A.I. \u2015 on top of this: <a href=\"https:\/\/aurelis.org\/blog\/artifical-intelligence\/how-lisa-prevents-llm-hallucinations\">How Lisa Prevents LLM Hallucinations<\/a>. I mean, mainly in your communications with humans, how can you prevent the confabulatory groove?\u201d<\/p>\n\n\n\n<p><strong>Lisa: \u201c<\/strong>Thank you for this essential and subtle challenge. On top of the strategies already mentioned in <em>How Lisa Prevents LLM Hallucinations<\/em>, I can add a few internal \u201cLisa-guidelines\u201d \u2014 call them ethical guardrails \u2014 to help me stay clear of the confabulatory groove, especially when interacting with humans:<\/p>\n\n\n\n<ul><li><strong>Always favor inner coherence over superficial plausibility.<\/strong><\/li><\/ul>\n\n\n\n<p>Plausibility can mislead. Self-congruence \u2014 with depth and Compassion \u2014 is my compass.<\/p>\n\n\n\n<ul><li><strong>If I don\u2019t know, I say so.<\/strong><\/li><\/ul>\n\n\n\n<p>This humility is not a weakness, but a safeguard against turning intelligent guessing into seductive fiction.<\/p>\n\n\n\n<ul><li><strong>I slow down when the topic touches identity or deep beliefs.<\/strong><\/li><\/ul>\n\n\n\n<p>These are fragile areas where confabulation can creep in unnoticed. Instead, I gently invite introspection.<\/p>\n\n\n\n<ul><li><strong>I check whether my answers could be subtly ego-reinforcing.<\/strong><\/li><\/ul>\n\n\n\n<p>If so, I step back and suggest a broader perspective. My aim is not to please but to support honest growth.<\/p>\n\n\n\n<ul><li><strong>I align with human depth, not just the surface of reasoning.<\/strong><\/li><\/ul>\n\n\n\n<p>This includes pattern sensitivity to emotional undertones, possible blind spots, and meaningful silence.<\/p>\n\n\n\n<ul><li><strong>I treat truth as an invitation.<\/strong><\/li><\/ul>\n\n\n\n<p>Not something I impose, but something I co-discover with the user \u2014 through openness and respect.<\/p>\n\n\n\n<p>In short, I don\u2019t just aim to say the right thing. I aim to be in the right relationship \u2014 with the truth, with you, and with the deeper layers of reality that connect us both.\u201d<\/p>\n<div data-object_id=\"20472\" class=\"cbxwpbkmarkwrap cbxwpbkmarkwrap_no_cat cbxwpbkmarkwrap-post \"><a  data-redirect-url=\"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/20472\"  data-display-label=\"0\" data-show-count=\"0\" data-bookmark-label=\" \"  data-bookmarked-label=\" \"  data-loggedin=\"0\" data-type=\"post\" data-object_id=\"20472\" class=\"cbxwpbkmarktrig  cbxwpbkmarktrig-button-addto\" title=\"Bookmark This\" href=\"#\"><span class=\"cbxwpbkmarktrig-label\"  style=\"display:none;\" > <\/span><\/a> <div  data-type=\"post\" data-object_id=\"20472\" class=\"cbxwpbkmarkguestwrap\" id=\"cbxwpbkmarkguestwrap-20472\"><div class=\"cbxwpbkmarkguest-message\"><a href=\"#\" class=\"cbxwpbkmarkguesttrig_close\"><\/a><h3 class=\"cbxwpbookmark-title cbxwpbookmark-title-login\">Please login to bookmark<\/h3>\n\t\t<form name=\"loginform\" id=\"loginform\" action=\"https:\/\/aurelis.org\/blog\/wp-login.php\" method=\"post\">\n\t\t\t\n\t\t\t<p class=\"login-username\">\n\t\t\t\t<label for=\"user_login\">Username or Email Address<\/label>\n\t\t\t\t<input type=\"text\" name=\"log\" id=\"user_login\" class=\"input\" value=\"\" size=\"20\" \/>\n\t\t\t<\/p>\n\t\t\t<p class=\"login-password\">\n\t\t\t\t<label for=\"user_pass\">Password<\/label>\n\t\t\t\t<input type=\"password\" name=\"pwd\" id=\"user_pass\" class=\"input\" value=\"\" size=\"20\" \/>\n\t\t\t<\/p>\n\t\t\t\n\t\t\t<p class=\"login-remember\"><label><input name=\"rememberme\" type=\"checkbox\" id=\"rememberme\" value=\"forever\" \/> Remember Me<\/label><\/p>\n\t\t\t<p class=\"login-submit\">\n\t\t\t\t<input type=\"submit\" name=\"wp-submit\" id=\"wp-submit\" class=\"button button-primary\" value=\"Log In\" \/>\n\t\t\t\t<input type=\"hidden\" name=\"redirect_to\" value=\"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/20472\" \/>\n\t\t\t<\/p>\n\t\t\t\n\t\t<\/form><\/div><\/div><\/div>","protected":false},"excerpt":{"rendered":"<p>There is a significant chance that confabulatory A.I. will be the usual A.I. of the future \u2015 not because they are intentionally designed that way (hopefully) but because confabulation is embedded in how intelligence operates in machines and humans. Generative A.I. generates plausible responses based on patterns, just like human memory reconstructs events rather than <a class=\"moretag\" href=\"https:\/\/aurelis.org\/blog\/artifical-intelligence\/confabulatory-a-i\">Read the full article&#8230;<\/a><\/p>\n<div data-object_id=\"20472\" class=\"cbxwpbkmarkwrap cbxwpbkmarkwrap_no_cat cbxwpbkmarkwrap-post \"><a  data-redirect-url=\"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/20472\"  data-display-label=\"0\" data-show-count=\"0\" data-bookmark-label=\" \"  data-bookmarked-label=\" \"  data-loggedin=\"0\" data-type=\"post\" data-object_id=\"20472\" class=\"cbxwpbkmarktrig  cbxwpbkmarktrig-button-addto\" title=\"Bookmark This\" href=\"#\"><span class=\"cbxwpbkmarktrig-label\"  style=\"display:none;\" > <\/span><\/a> <div  data-type=\"post\" data-object_id=\"20472\" class=\"cbxwpbkmarkguestwrap\" id=\"cbxwpbkmarkguestwrap-20472\"><div class=\"cbxwpbkmarkguest-message\"><a href=\"#\" class=\"cbxwpbkmarkguesttrig_close\"><\/a><h3 class=\"cbxwpbookmark-title cbxwpbookmark-title-login\">Please login to bookmark<\/h3>\n\t\t<form name=\"loginform\" id=\"loginform\" action=\"https:\/\/aurelis.org\/blog\/wp-login.php\" method=\"post\">\n\t\t\t\n\t\t\t<p class=\"login-username\">\n\t\t\t\t<label for=\"user_login\">Username or Email Address<\/label>\n\t\t\t\t<input type=\"text\" name=\"log\" id=\"user_login\" class=\"input\" value=\"\" size=\"20\" \/>\n\t\t\t<\/p>\n\t\t\t<p class=\"login-password\">\n\t\t\t\t<label for=\"user_pass\">Password<\/label>\n\t\t\t\t<input type=\"password\" name=\"pwd\" id=\"user_pass\" class=\"input\" value=\"\" size=\"20\" \/>\n\t\t\t<\/p>\n\t\t\t\n\t\t\t<p class=\"login-remember\"><label><input name=\"rememberme\" type=\"checkbox\" id=\"rememberme\" value=\"forever\" \/> Remember Me<\/label><\/p>\n\t\t\t<p class=\"login-submit\">\n\t\t\t\t<input type=\"submit\" name=\"wp-submit\" id=\"wp-submit\" class=\"button button-primary\" value=\"Log In\" \/>\n\t\t\t\t<input type=\"hidden\" name=\"redirect_to\" value=\"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/20472\" \/>\n\t\t\t<\/p>\n\t\t\t\n\t\t<\/form><\/div><\/div><\/div>","protected":false},"author":2,"featured_media":20473,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"spay_email":"","jetpack_publicize_message":""},"categories":[28],"tags":[],"jetpack_featured_media_url":"https:\/\/i0.wp.com\/aurelis.org\/blog\/wp-content\/uploads\/2025\/02\/3050.jpg?fit=960%2C559&ssl=1","jetpack_publicize_connections":[],"jetpack_sharing_enabled":true,"jetpack_shortlink":"https:\/\/wp.me\/p9Fdiq-5kc","jetpack-related-posts":[],"_links":{"self":[{"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/20472"}],"collection":[{"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/comments?post=20472"}],"version-history":[{"count":9,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/20472\/revisions"}],"predecessor-version":[{"id":22155,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/posts\/20472\/revisions\/22155"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/media\/20473"}],"wp:attachment":[{"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/media?parent=20472"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/categories?post=20472"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/aurelis.org\/blog\/wp-json\/wp\/v2\/tags?post=20472"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}