Estimate Integrated Information for Computer Architectures and Demonstrate a Dissociation Between Artificial Intelligence and Consciousness

Written By Tiny Blue Dot Foundation

Can computers be conscious? Developments in machine learning and computing power are raising the possibility that artificial general intelligence (AGI) may be within reach. This raises the question of artificial consciousness: if a computer were functionally equivalent to a human, having the same cognitive abilities, would it experience sights, sounds, and thoughts, as we do when we are conscious? Answering this question has profound implications, because if computers were sentient, then we would have to grant them similar rights as to humans, and cease treating them as mere tools. 

To answer this question in a principled answer, however, we need a theory of consciousness that accounts for what consciousness actually is, identifies its essential phenomenal properties, translates them into measurable physical properties, and can be validated on ourselves. Building upon integrated information theory (IIT), a companion Project (Estimate the substrate of consciousness from the ‘connectome’ of the human brain) aims precisely at validating IIT starting from human consciousness and the human brain. Crucially, preliminary results are consistent with the predictions of IIT and indicate that only a neural substrate organized in a specific manner, such as posterior cortical regions, can support high levels of consciousness. 

In this Project, we employ IIT to determine whether the physical substrate typical of digital computers can or cannot support high levels of consciousness. To do so, we consider pairs of simple systems constituted of logic gates, one of which—a basic computer—simulates the other with full functional equivalence. By applying the principles of IIT, we demonstrate that (i) two systems can be functionally equivalent without being phenomenally equivalent; (ii) that this conclusion applies no matter how one ‘black-boxes’ the computer’s gates in space and time; and (iii) that even certain Turing-complete systems, which could theoretically pass the Turing test and simulate a human brain in detail, would be negligibly conscious

More generally, the approach developed in this Project will allow us to infer whether any biological or artificial network is conscious, even if its organization is radically different from the human brain, and irrespective of intelligent behavior.

Broader Impact:

We expect this Project to show, based on IIT, that conventional computer architectures cannot support artificial consciousness, no matter how precisely they simulate human functions or the human brain. This implies a dissociation between intelligence and consciousness: something, say a robot equipped with a powerful computer endowed with AGI, can display highly intelligent behaviors and functions, yet be virtually unconscious—it would ‘do’ everything and ‘be’ nothing. The implications are important, broad, and urgent. It is natural for us to project consciousness based on intelligence, as we tend to do with other species and we might soon do with machines. Yet, if IIT is correct, there is a radical divide between conscious beings like us or other animals and mere machines, no matter how smart—it is the divide between being there or not.

Publications:

  • Grasso, Matteo, Andrew M. Haun, and Giulio Tononi. "Of maps and grids." Neuroscience of Consciousness 2021.2 (2021): niab022.

  • Albantakis, L et al. “A macro agent and its actions”. In J Voorsholz & M Gabriel’s (Ed) Top-down Causation and Emergence (2021). Synthese Library, Springer Publishing. 

  • Haun, Andrew, and Giulio Tononi. "Why does space feel the way it does? Towards a principled account of spatial experience." Entropy 21.12 (2019): 1160.

  • Albantakis, Larissa, and Giulio Tononi. "Causal composition: structural differences among dynamically equivalent systems." Entropy 21.10 (2019): 989.

  • Findlay, Graham et al. "Dissociating Intelligence from Consciousness in Artificial Systems - Implications of Integrated Information Theory.” Proceedings of the 2019 Towards Conscious AI Systems Symposium, AAAI SSS19 (2019)

Tiny Blue Dot Foundation
Previous

Estimate the Substrate of Consciousness from the ‘Connectome’ of the Human Brain