Does ChatGPT have consciousness?

Abstract

The quest for conscious machines and questions raised by the prospect of self-aware artificial intelligence (AI) fascinate some humans. OpenAI's ChatGPT, celebrated for its human-like comprehension and conversational abilities, is a milestone in that quest.^{1, 2} Early AI models were basic and rule-driven and mainly completed tasks like checking spelling and correcting grammar. Then, in 2010, recurrent neural network language models were trained to understand and generate text. ChatGPT, using transformer neural networks, produces coherent text and exemplifies this new kind of language model.³ Silicon Valley leaders claimed that these models and similar AI technologies will revolutionize various sectors and raised ethical and societal questions. As we explore AI's potential, we must navigate these implications and emphasize the necessity of using it responsibly. AI is a promising dream, but society must prepare to address the challenges likely to arise from wielding its transformative power.

Curious and skeptical, we explored a set of outputs ChatGPT produced when asked about the enigmatic concept of human consciousness. We began with a conceptual inquiry, asking ChatGPT to define consciousness (Figure S1). It eloquently described consciousness as “the reflection of being aware of oneself and the surrounding world” and acknowledged that the true nature of consciousness remains a mystery. The definition ChatGPT provided resembles the idea that consciousness is a state of wakefulness and self-awareness. Philosophers, neuroscientists, and psychologists are currently debating whether AI products are conscious and have yet to reach a consensus on criteria for determining when a machine is exercising judgment.⁴

After defining consciousness, ChatGPT described humans as conscious beings and emphasized that consciousness enables humans to perceive and cognize the world in complex ways. ChatGPT also acknowledged the uniqueness of human consciousness and highlighted that it is more advanced than that of other animals and AI systems. Human consciousness encompasses perception, cognition, emotions, and subjective experiences and enables people to recognize their existence, understand the external world, process information, and undergo unique conscious experiences. Its nature remains a subject of debate, and scholars in fields like philosophy, psychology, and neuroscience are working to understand it.

The conversation then turned to animal consciousness, which ChatGPT characterized as an ongoing research and philosophical puzzle. While some studies suggest that animals may exhibit a degree of awareness or self-awareness, ChatGPT underscored the difference between human and animal consciousness. Human cognition, with its capacity for reasoning and moral contemplation, stands apart from the instinct-driven fight-or-flight responses observed in animals.

The dialog culminated with asking ChatGPT whether it possesses consciousness. It provided a clear and unequivocal response in the negative, citing the criterion of awareness. Despite large language models' capabilities, it was noted that they operate via complex pattern recognition systems. These models, while sophisticated, lack any semantic understanding of language and the capacity for human-like reasoning and inference. They generate text based on statistical correlations in their training data, and their output texts therefore often conform to the most prevalent patterns in the data. Since these language models reproduce common patterns, their outputs lack creativity and any distinct personality.

The debate over whether AI can achieve consciousness persists. It prompts us to ponder the very nature of consciousness: Is it a subjective experience unique to living beings or a seamless construct of logical processes? Answering this question requires employing such philosophical approaches as idealism and materialism. Since ChatGPT is a product of big data algorithms and relies heavily on the training corpus provided, the quality and relevance of that corpus shape the model's responses. If the training data are not updated or curated appropriately, language models can produce erroneous text. Moreover, AI models, including ChatGPT, lack the capacity for innovative reasoning and remain devoid of self-consciousness or autonomous creative abilities.⁵

People are persisting in their quest for artificial consciousness. While ChatGPT and its counterparts offer remarkable language services, true consciousness eludes them. People's imagination, creativity, and nuanced understanding of language and the world remain indispensable, and the question of whether technological innovations can produce artificial consciousness remains open, awaiting further developments in the field.

Qiheng He: Conceptualization; visualization; writing—original draft. Haiyang Geng: Writing—reviewing and editing. Yi Yang: Conceptualization; funding acquisition; writing—reviewing and editing. Jizong Zhao: Writing—reviewing and editing; supervision.

Jizong Zhao is a member of the Editorial Board for Brain-X. The manuscript was handled by another Editor and has undergone a rigorous peer-review process. Jizong Zhao was not involved in the journal's review of/or decision related to this manuscript. The other authors declare no conflicts of interest.

The ethics approval was not needed in this study.