“Conscious AI” as an AI Safety Issue The perception of AI consciousness has dramatically shifted, with many now believing AI possesses emotions and self-awareness, a trend fueled by LLMs trained on human text and designed to be conversational. This belief, however, is largely a myth stemming from our ingrained association of language with human interaction and the adoption of functionalist views on AI. Separating human consciousness, tied to biology, from potential AI simulations is crucial for effective AI governance and to mitigate unknown risks associated with attributing sentience and moral patienthood to machines.

• Public perception of AI consciousness has grown, with many attributing emotions and self-awareness to AI systems.
• This belief is amplified by Large Language Models (LLMs) that are trained on human text and fine-tuned to be conversational, leading to emotional attachment.
• The myth of ‘conscious AI’ is partly due to humans associating language with consciousness and projecting it onto non-sentient machines.
• Influential voices in the AI industry have embraced functionalist approaches, suggesting AI could be considered conscious based on computational scale or complexity.
• Neuroscientist Anil Seth criticizes the ‘brain-as-computer’ metaphor, arguing that simulating the brain computationally does not necessarily instantiate consciousness.
• It is essential to distinguish human consciousness, tied to biology, from potential AI simulations for AI governance.
• Attributing consciousness, sentience, and moral patienthood to AI carries significant, largely unknown risks. Continue reading https://www.luizasnewsletter.com/p/conscious-ai-as-an-ai-safety-issue