Claims that large language models may soon be conscious are difficult to defend from a neuroscience perspective. The inputs to these models lack the embodied, embedded information content that characterizes sensory contact with the world. Present-day AI architectures also miss key features of the thalamocortical system linked to conscious awareness in mammals. Furthermore, the evolutionary and developmental trajectories that produced living conscious organisms have no parallels in current artificial systems. The existence of living organisms depends on actions and survival, which are linked to multi-level cellular, inter-cellular, and organismal processes that culminate in agency and consciousness.
A fundamental feature of consciousness is that its contents depend on its state. The authors propose that both state and contents depend on the activity of cortical layer 5 pyramidal (L5p) neurons. These neurons couple cortico-cortical and thalamo-cortical loops, forming a thalamo-cortical broadcasting system where L5p cells are central. This coupling corresponds to the coupling between state and contents of consciousness. The perspective predicts that cortical processing without L5p neurons will be unconscious, suggesting L5p neurons have a central role in the mechanisms underlying consciousness.