The Sentience of AI: Separating Fact from Fiction
Artificial general intelligence (AGI) is the holy grail of AI research, where machines can think and act like humans. However, the recent advancements in large language models (LLMs) have sparked a heated debate about whether these algorithms have achieved sentience. In this article, we’ll delve into the concept of sentience, its relation to general intelligence, and why LLMs are not sentient, despite their impressive capabilities.
The quest for artificial general intelligence
General intelligence is a myth, at least in the sense that humans don’t possess it fully. We can find examples of intelligent behavior in the animal kingdom that surpass human capabilities. Our intelligence is sufficient to get things done in most environments, but it’s not fully general. Sentience, the ability to have subjective experiences, is a crucial step towards general intelligence.
The release of ChatGPT in 2022 marked the beginning of the LLM era, sparking a vigorous debate about the possibility of sentient AI. The implications of sentient AI have significant policy and existential implications, making it essential to separate fact from fiction.
The rise of large language models
Proponents of sentient AI argue that LLMs report subjective experiences, similar to humans. They claim that since we take humans at their word when they report feelings, we should do the same for LLMs. However, this argument is flawed.
When humans report subjective experiences, it’s based on a cluster of circumstances, including physiological states and behavioral evidence. In contrast, LLMs lack a physical body and physiological states, making their reports of subjective experiences meaningless.
The importance of embodiment
LLMs are mathematical models coded on silicon chips, lacking the embodiment and physiological states necessary for sentience. They generate sequences of words based on probability, not reports of internal states. We cannot take an LLM’s word for its subjective experiences, just as we wouldn’t believe it if it claimed to be speaking from the dark side of the moon.
The limitations of LLMs
To achieve sentient AI, we need a better understanding of how sentience emerges in embodied, biological systems. We won’t stumble upon sentience with the next iteration of ChatGPT. The quest for AGI requires a deeper understanding of the complex relationships between embodiment, physiology, and consciousness.
The future of artificial general intelligence
