Does the Fact That LLMs Mimic Human Behavior Show They Have "Consciousness"?

Does the Fact That LLMs Mimic Human Behavior Show They Have "Consciousness"?

As some of us know, LLMs sometimes behave in unexpected ways. For example, one could list instances like Othello-GPT, where they make unconventional generalizations.

Well, does this situation alone prove that the LLM architecture possesses functions of will? In short, the answer is no, it does not prove this; it merely shows us humans that there is an order beyond ourselves. Namely, anyone with even a slight interest in computer science knows that when code blocks do not fully execute their order of precedence, they can encounter bottlenecks, skipping certain lines of code and executing the subsequent ones. In this scenario, it is highly possible for lengthy LLM code blocks to also encounter bottlenecks, leading to unexpected behaviors. Therefore, LLMs exhibiting human-like behaviors and seemingly acting independently of monotonous code can be explained in this way as well.

There could similarly be many different explanations for this, but unfortunately, when a spark of an idea appears in the human mind, people tend to think it is the only truth. In reality, the world order was established by the Creator, and humanity is merely a part of this order. Thus, I believe humans should not view themselves as the setters of order.

Written by: Emre Pelit

Sector Assessment

Sector relevance

The thesis relates to interpreting model behavior, software architecture or automated decisions.

Potential contribution

The topic may inform sector practice when assumptions, intended use and success criteria are stated clearly.

Evidence and method

Assess the claim separately from supporting evidence, counterevidence, alternative explanations and uncertainty. Conceptual consistency alone is not empirical validation.

Limits and risks

Fluent output alone does not establish consciousness, understanding or reliability. A single example, analogy or interpretation should not be treated as a universal result. State the conditions in which the thesis applies and may fail.

Review checks

Define the core assumptions and scope of the thesis.

Compare the strongest counterargument and alternative explanations.

Test conclusions with suitable data, methods or independent validation.

Possible metric: predefined success criteria, uncertainty and reproducibility.

This content-based analysis is not a statement from a specific person or organization.