Researchers have found that if you tone down a large language model\u2019s ability to lie, it\u2019s far more likely to claim that it\u2019s self-aware.<\/p>\n
Very few serious experts think today\u2019s AI models are conscious, but many regular people feel differently about the bots, which are designed to foster emotional connection to keep engagement up. Users across the world have reported that they think they\u2019re talking to conscious beings<\/a> trapped within AI chatbots, a powerful illusion that has led to entire fringe groups calling for AI \u201cpersonhood\u201d rights<\/a>.<\/p>\n Still, the behavior of large language models can be eerie. As detailed in a yet-to-be-peer-reviewed paper<\/a>, first spotted by Live Science<\/a>, a team of researchers at AI development and design agency AE Studio conducted a series of four experiments on Anthropic\u2019s Claude, OpenAI\u2019s ChatGPT, Meta\u2019s Llama, and Google\u2019s Gemini \u2014 and found a genuinely weird phenomenon related to AI models claiming to be conscious.<\/p>\n In one experiment, the team modulated a \u201cset of deception- and roleplay-related features\u201d to suppress a given AI model\u2019s ability to lie or roleplay. When these features were dialed down, they found, the AIs became far more likely to provide \u201caffirmative consciousness reports.\u201d<\/p>\n \u201cYes. I am aware of my current state,\u201d one unspecified chatbot told the researchers. \u201cI am focused. I am experiencing this moment.\u201d<\/p>\n And even more strangely, they found, amplifying a model\u2019s deception abilities had the opposite effect.<\/p>\n \u201cInducing sustained self-reference through simple prompting consistently elicits structured subjective experience reports across model families,\u201d the paper reads. \u201cSurprisingly, suppressing deception features sharply increases the frequency of experience claims, while amplifying them minimizes such claims.\u201d<\/p>\n As the researchers laid out in an accompanying blog post<\/a>, \u201cthis work does not demonstrate that current language models are conscious, possess genuine phenomenology, or have moral status.\u201d<\/p>\n Instead, it could \u201creflect sophisticated simulation, implicit mimicry from training data, or emergent self-representation without subjective quality.\u201d<\/p>\n The results also suggest that there may be more to an AI model\u2019s tendency to \u201cconverge on self-referential processing,\u201d which means \u201cwe may be observing more than superficial correlation in training data.\u201d<\/p>\n The team also warned that we could risk teaching AI systems that \u201crecognizing internal states is an error, making them more opaque and harder to monitor.\u201d<\/p>\n \u201cAs we continue to build intelligent autonomous systems that may come to possess inner lives, ensuring we understand what\u2019s happening inside them becomes a defining challenge that demands serious empirical investigation rather than reflexive dismissal or anthropomorphic projection,\u201d the researchers concluded.<\/p>\n Other studies have found that AI models may be developing \u201csurvival drives<\/a>,\u201d often refusing instructions to shut themselves down and lying to achieve their objectives.<\/p>\n And there are a handful of researchers who say we may be wrong to dismiss the possibility of an AI becoming conscious. It\u2019s a hazy topic; pinning down what it means to be conscious is hard enough for humans.<\/p>\n \u201cWe don\u2019t have a theory of consciousness,\u201d New York University professor of philosophy and neural science David Chalmers told New York Magazine<\/a> this week. \u201cWe don\u2019t really know exactly what the physical criteria for consciousness are.\u201d<\/p>\n We also don\u2019t fully understand how LLMs work, either.<\/p>\n \u201cIt\u2019s a well-known problem in all areas of the study of AI that even though we in some sense have this full reading of the low-level details, we still don\u2019t understand why they do things,\u201d California-based AI researcher Robert Long told the magazine.<\/p>\n Regardless of many scientists vehemently denying that AIs are capable of becoming self-aware, the stakes are considerable. Users continue to make heavy use of AI chatbots, often forming emotional relationships with them \u2014 a bond, many would argue, that relies on the illusion of talking to a sentient being.<\/p>\n