As Large Language Models (LLMs) like ChatGPT and Gemini become more sophisticated, philosophers and ethicists are raising concerns about the potential for humans to misattribute consciousness and responsibility to these systems. The emerging philosophical consensus maintains that despite their advanced capabilities—such as strategic planning, problem-solving, and mimicking human-like conversation—LLMs still do not possess true subjective experience, or qualia, which is the “Hard Problem” of consciousness. They are better understood as “quasi-agents” with “quasi-beliefs,” capable of autonomous behavior but lacking genuine sentience. The key warning revolves around the “responsibility gap” created by highly autonomous AI. If an LLM causes harm, who is to blame? Researchers advocate for a “many-agents-many-levels-many-interactions” (M3) approach to moral responsibility, stressing that responsibility should be assigned not just to the model itself, but across the complex web of developers, deployers, and users. The psychological tendency for humans to “over-attribute” consciousness and intention to AI systems is a central ethical risk, as it can obscure the true sources of accountability and potentially degrade shared social truth through the creation of plausible but factually inaccurate “careless speech.” The challenge is to develop robust frameworks that manage these advanced systems while resisting the temptation to anthropomorphize them.

Source: PMC/Frontiers in Artificial Intelligence (November 2025) 

Link: Responsibility Gaps, LLMs & Organisations: Many Agents, Many Levels, and Many Interactions – PMC