When a large language model gives a self-report about what is happening inside it, how can we know whether that report is true? The seventeenth-century philosopher Spinoza claims that mind and body are one and the same thing, expressed in two ways. In other words, the mind and body are parallel with each other and unfold in the same order and connection. Within this framework, this paper argues that a trained language model can be treated as a computational individual. When its computational bod…
Read moreWhen a large language model gives a self-report about what is happening inside it, how can we know whether that report is true? The seventeenth-century philosopher Spinoza claims that mind and body are one and the same thing, expressed in two ways. In other words, the mind and body are parallel with each other and unfold in the same order and connection. Within this framework, this paper argues that a trained language model can be treated as a computational individual. When its computational body is affected by a prompt, its mind is aware of this change, which we call "minimal felt experience." Its introspective self-report puts that awareness into words and, if genuine, should correspond to the computational process in its body — the layer-by-layer numerical states the model computes while reading a prompt (the hidden states). We bring this metaphysical assertion into an anti-circular engineering test on one of the open-weight models — Gemma 3 27B IT, made by Google. The results show that the report-derived event chains are fully recovered in the hidden states; the self-report co-varies with the hidden states under intervention. Furthermore, the same report-side chains recur across three other models. This is the first empirical test of Spinozist parallelism on an LLM. The claim is bounded — not human-like consciousness. Yet the self-report proved faithful to the model's internal process, and should be taken seriously.
