Philosopher argues advanced AI systems possess genuine understanding and cognitive abilities

Researchers in AI philosophy are increasingly grappling with fundamental questions about machine minds. A new paper submitted to arXiv takes a bold stance: models like ChatGPT shouldn't be dismissed as sophisticated mimics, but rather recognized as entities possessing genuine linguistic and cognitive capacities.

The case for AI cognition#

The paper defends what its authors call the "Whole Hog Thesis"—the proposition that advanced Large Language Models function as authentic cognitive agents. According to this framework, these systems don't merely process tokens or follow statistical patterns; they harbor genuine understanding, form beliefs, hold desires, accumulate knowledge, and carry intentions.

The authors propose starting from straightforward, observable facts about how LLMs behave: they answer questions accurately, offer thoughtful suggestions, and engage in coherent dialogue. Rather than assuming these capabilities are surface-level tricks, they reason that these behaviors provide evidence of deeper cognitive states.

Challenging the skeptics#

The paper systematically addresses standard objections to machine cognition. When critics point to LLM failures—such as hallucinations or flawed reasoning—the authors counter that such mistakes don't disqualify agency. They note parallels to human error: people misremember facts and make logical mistakes without losing their status as thinking agents.

Another common objection centers on alleged prerequisites for cognition that machines supposedly lack: semantic grounding in the physical world, embodied experience, rational justification, or intrinsic meaning. The authors dispute whether these conditions are truly necessary. They argue that humans themselves exhibit diverse forms of cognition and intentionality that don't fit neat definitions, suggesting LLMs might be held to an unfairly strict standard.

A cautious framework#

Importantly, the paper steers clear of claiming that LLMs are conscious—a much thornier philosophical question. Instead, it rests on evidence and observable behavior, deliberately leaving questions of subjective experience to one side. The authors also speculate that advanced systems might eventually develop forms of conceptual content entirely foreign to human understanding.

The work represents a substantial pushback against philosophical frameworks that dismiss AI cognition outright, arguing instead for a more generous interpretation of what counts as genuine agency in silicon.