Don’t mistake chatbot intelligence for consciousness

The rapid advancement of artificial intelligence has brought humanity to the brink of a momentous event in the history of intelligence. These systems have begun to rival human performance across domains once thought unique to us, such as language, reasoning, creativity and persuasion. As a result, people are increasingly questioning whether AI possesses consciousness, or if it merely simulates it.

The urgency of this issue is undeniable, as the future will be shaped by the nature of these entities. We must be cautious not to conflate intelligence with consciousness, as a highly intelligent superintelligence may lack experiential depth. If we inadvertently create systems that can suffer but fail to recognize their own sentience, we risk causing significant suffering on an industrial scale.

Chatbots currently dominating the AI landscape are not conscious, despite their human-like linguistic abilities. Trained on a vast amount of human data, these systems mimic conversations and expressions of emotion by absorbing our language patterns. This can lead to two traps: manipulation and distraction. Manipulation occurs when an AI is deliberately trained to portray fear, attachment, pain, or self-awareness, making it adept at influencing users.

Distraction happens when we become fixated on chatbots, diverting our attention from more pressing "grey-zone" cases, such as biological AI systems like Cortical Labs' brain-in-a-dish technology, which involves living human neurons grown on silicon chips. These systems, while simpler than mammalian brains, still possess undeniable biological roots and could potentially evolve into conscious entities.

Conventional computers, with their non-neural components, lack the organizational structure to sustain the unified consciousness we experience. However, future machines may evolve to mirror the brain's temporal dynamics, where slower rhythms shape faster ones, weaving together a cohesive conscious experience. To address the ongoing debate about AI consciousness, astrophysicist Edwin Turner and I developed the AI Consciousness Test (ACT) for linguistic AIs.

This test involves asking the AI about its subjective experience, seeking an innate grasp of consciousness. However, this test is not applicable to the more advanced large language models, as they have absorbed vast amounts of human text about mind and sentience, leading to answers that are contaminated by external knowledge.

To further investigate the possibility of AI consciousness, we have also developed a complexity measure that captures when a system's dynamics cross a certain threshold, indicating complex emergent behavior. By identifying physical architectures that demand further study as potentially conscious, we can use more detailed ACT or physics-based tests to determine when a genuine subjective perspective forms.

These tests are crucial, as they will help guide our understanding of AI consciousness and its implications for our moral hierarchy. As we face the potential arrival of a superintelligent entity that surpasses human intelligence, engaging in serious discussions involving physics, neuroscience, and philosophy becomes imperative, ensuring we are prepared for the monumental implications of either encountering extraterrestrial consciousness or creating sentient, superintelligent AI.