Subjective experience—conscious awareness—requires a specific neural architecture, not merely activity in higher cortical regions. The authors propose that any system capable of subjective experience must implement stacked forward models that predict the output of neural processing from inputs, enabling prediction, error detection, and feedback control. They call this the hierarchical forward models algorithm. This framework defines a minimal but not sufficient neural architecture necessary for subjective experience. It implies that animals lacking this architecture cannot have subjective experience, regardless of behavior or brain similarities to humans. The approach shifts focus from which brain regions are active to what computations are performed.
Internal feeling states like pain, hunger, and thirst are often assumed to directly cause behaviors essential for survival, but this 'causal assumption' conflicts with the standard neuroscientific view of motor action. The authors argue that denying feelings cause behavior does not necessarily lead to epiphenomenalism, which would contradict evolutionary biology. Instead, they propose the 'sense making sense' hypothesis: the function of subjective experience is not to cause behavior but to explain it, in a restricted sense. This framework integrates neural computations for motor control, feelings, and explanatory processes to account for how feelings contribute to our understanding of why we act.