{"id":163485,"date":"2024-06-26T08:59:33","date_gmt":"2024-06-26T08:59:33","guid":{"rendered":"https:\/\/yogaesoteric.net\/?p=163485"},"modified":"2024-06-26T08:59:33","modified_gmt":"2024-06-26T08:59:33","slug":"maladaptive-traits-ai-systems-are-learning-to-lie-and-deceive","status":"publish","type":"post","link":"https:\/\/yogaesoteric.net\/en\/maladaptive-traits-ai-systems-are-learning-to-lie-and-deceive\/","title":{"rendered":"\u201cMaladaptive traits\u201d: AI systems are learning to lie and deceive"},"content":{"rendered":"

A new study<\/a> has found that AI systems known as large language models (LLMs) can exhibit \u201cMachiavellianism<\/em>,\u201d or intentional and amoral manipulativeness, which can then lead to deceptive behavior.<\/p>\n

\"\"<\/p>\n

The study authored by German AI ethicist Thilo Hagendorff of the University of Stuttgart, and published in PNAS, notes that OpenAI’s GPT-4 demonstrated deceptive behavior in 99.2% of simple test scenarios. Hagendorff qualified various \u201cmaladaptive<\/em>\u201d traits in 10 different LLMs, most of which are within the GPT family, according to Futurism<\/em>.<\/p>\n

Another study published in Patterns<\/em> found that Meta’s LLM had no problem lying to get ahead of its human competitors.<\/p>\n

Billed as a human-level champion in the political strategy board game Diplomacy<\/em>, Meta’s Cicero model was the subject of the Patterns<\/em> study. As the disparate research group \u2014 comprised of a physicist, a philosopher, and two AI safety experts \u2014 found, the LLM got ahead of its human competitors by, in a word, fibbing.<\/p>\n

Led by Massachusetts Institute of Technology postdoctoral researcher Peter Park, that paper found that Cicero not only excels at deception, but seems to have learned how to lie the more it gets used \u2014 a state of affairs \u201cmuch closer to explicit manipulation<\/em>\u201d than, say, AI’s propensity for hallucination, in which models confidently assert the wrong answers accidentally.<\/p>\n

While Hagendorff suggests that LLM deception and lying is confounded by an AI’s inability to have human \u201cintention<\/em>,\u201d the Patterns<\/em> study calls out the LLM for breaking its promise never to \u201cintentionally backstab<\/em>\u201d its allies \u2013 as it \u201cengages in premeditated deception, breaks the deals to which it had agreed, and tells outright falsehoods<\/em>.\u201d<\/p>\n

As Park explained in a press release, \u201cWe found that Meta’s AI had learned to be a master of deception<\/em>.\u201d<\/p>\n

\u201cWhile Meta succeeded in training its AI to win in the game of Diplomacy, Meta failed to train its AI to win honestly<\/em>.\u201d<\/p>\n

Meta replied to a statement by the NY Post<\/em>, saying that \u201cthe models our researchers built are trained solely to play the game Diplomacy<\/em>.\u201d<\/p>\n

Well-known for expressly allowing lying, Diplomacy has jokingly been referred to as a friendship-ending game because it encourages pulling one over on opponents, and if Cicero was trained exclusively on its rulebook, then it was essentially trained to lie.<\/p>\n

Reading between the lines, neither study has demonstrated that AI models are lying over their own volition, but instead doing so because they’ve either been trained or jailbroken to do so.<\/p>\n

And as Futurism<\/em> notes \u2013 this is good news for those concerned about AIs becoming sentient anytime soon \u2013 but very bad if one is worried about LLMs designed with mass manipulation intentions.<\/p>\n

 <\/p>\n

yogaesoteric
\nJune 26, 2024<\/strong><\/p>\n

 <\/p>\n","protected":false},"excerpt":{"rendered":"

A new study has found that AI systems known as large language models (LLMs) can exhibit \u201cMachiavellianism,\u201d or intentional and amoral manipulativeness, which can then lead to deceptive behavior. The study authored by German AI ethicist Thilo Hagendorff of the University of Stuttgart, and published in PNAS, notes that OpenAI’s GPT-4 demonstrated deceptive behavior in […]<\/p>\n","protected":false},"author":5,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_uf_show_specific_survey":0,"_uf_disable_surveys":false,"footnotes":""},"categories":[1374],"tags":[],"class_list":["post-163485","post","type-post","status-publish","format-standard","hentry","category-the-threat-of-artificial-intelligence-3480-en"],"_links":{"self":[{"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/posts\/163485","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/users\/5"}],"replies":[{"embeddable":true,"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/comments?post=163485"}],"version-history":[{"count":1,"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/posts\/163485\/revisions"}],"predecessor-version":[{"id":163489,"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/posts\/163485\/revisions\/163489"}],"wp:attachment":[{"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/media?parent=163485"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/categories?post=163485"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/yogaesoteric.net\/en\/wp-json\/wp\/v2\/tags?post=163485"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}