Simulating a simple computation in an artificial neural network, researchers recorded neuron activity during visual stimulation and replayed those signals back into the same neurons. This replay degraded the computation by erasing counterfactual activity patterns—alternative neural states that could have occurred—while leaving ongoing brain activity unchanged. This outcome reveals a disconnect between neural activity and computational structure, challenging the computational functionalist view that consciousness emerges from the right computations, whether in machines or biological brains.
Integrated information, a proposed signature of consciousness, is maximized in a biophysical network model when the nonspecific thalamus drives thick-tufted layer 5 pyramidal neurons into a regime of time-varying synchronous bursting. In this regime, variable spiking dynamics with broad pairwise correlations support enhanced integrated information. The peak in integrated information coincides with criticality signatures and empirically observed layer 5 pyramidal bursting rates. These findings suggest that the thalamocortical core of the mammalian brain may be evolutionarily configured to optimize effective information processing, offering a potential neuronal mechanism linking microscale theories to macroscale signatures of consciousness.