AIT foundations of structured experience

Abstract

We present a unifying framework to study consciousness based on algorithmic information theory (AIT). We take as a premise that ``there is experience'' and focus on the requirements for structured experience (S)--- the spatial, temporal, and conceptual organization of our first-person experience of the world and of ourselves as agents in it. Our starting point is the insight that access to good models---succinct and accurate generative programs of world data---is crucial for homeostasis and survival. We hypothesize that the successful comparison of such models with data provides the structure to experience. Building on the concept of Kolmogorov complexity, we can associate the qualitative aspects of S with the algorithmic features of the model, including its length, which reflects the structure discovered in the data. Moreover, a modeling system tracking structured data will display dimensionality reduction and criticality features that can be used empirically to quantify the structure of the program run by the agent. KT provides a consistent framework to define the concepts of life and agent and allows for the comparison between artificial agents and S-reporting humans to provide an educated guess about agent experience. A first challenge is to show that a human agent has S to the extent they run encompassing and compressive models tracking world data. For this, we propose to study the relation between the structure of neurophenomenological, physiological, and behavioral data. The second is to endow artificial agents with the means to discover good models and study their internal states and behavior. We relate the algorithmic framework to other theories of consciousness and discuss some of its epistemological, philosophical, and ethical aspects.

LSD-induced increase of Ising temperature and algorithmic complexity of brain dynamics

“…Personalized Ising models were created using fMRI BOLD data by adjusting the model temperature for each subject and condition in relation to the group average model. We found that the effects of LSD translate into increased BOLD signal complexity and Ising temperature, in agreement with earlier findings and predictions from existing theories of the effects of psychedelics such as the relaxed beliefs under psychedelics (REBUS), the anarchic brain hypothesis [4], and the algorithmic information theory of consciousness (KT) [64,67]. More precisely, in agreement with earlier work [23], we found that the system in the placebo condition was already in the paramagnetic phase-above the critical point-, with ingestion of LSD resulting in a shift away from criticality into a more disordered state.…”

“…In [68], we hypothesized that increased apparent (Shannon) complexity (entropy) of spontaneous activity is a logical corollary of this action, as courser grain functional architecture will become finer-grained, which is indeed observed using measures such as LZW [71], such as LZW [66]. The present study extends on previous work, however, by incorporating algorithmically-oriented methods such as BDM.…”

“…As discussed in the context of the algorithmic information theory of consciousness (KT) [64, 67], psychedelics will shift the dynamics of an agent system tracking world data to a less constrained state further away from criticality, and hence produce more complex signals.…”

Comparison between an exact and a heuristic neural mass model with second order synapses

“…We may conjecture that certain neuronal populations may be in states near criticality, i.e. close to the bifurcation points in the NMM2 model (Chialvo, 2004;Carhart-Harris, 2018;Vázquez-Rodríguez et al, 2017;Zimmern, 2020;Ruffini and Lopez-Sola, 2022). This would apply, for example, to inhibitory populations, which display a Hopf bifurcation where a state near the critical point will display arbitrarily large amplified sensitivity to weak but uniform perturbations applied over long time scales.…”

Algorithmic structure of experience and the unfolding argument

“…In causal structure theories of consciousness such as IIT [21] or computational theories of consciousness such as KT [18,19], where the structure of the information processing system is important, the statement is that a system will experience and potentially report consciousness in a set of experiments depending on its causal/computational structure, e.g., as partly characterized by integrated information Φ in the case of IIT, or by program length in KT. As it turns out, to obtain a high value of Φ, computational or cause/effect recurrence is needed.…”