Neuromorphic Correlates of Artificial Consciousness

TL;DR

The paper tackles whether artificial systems can exhibit consciousness by linking neural correlates of consciousness (NCC) with integrated information theory (IIT) within neuromorphic design and brain simulations. It introduces Neuromorphic Correlates of Artificial Consciousness (NCAC) as a four-phase framework, including Quantification, Simulation, Adaptation, and Implementation, aimed at maximizing high $\Phi$ in neuromorphic architectures. It surveys NCC and IIT foundations, discusses spiking neural networks, and reviews brain-simulation projects (e.g., HBP, Blue Brain, Spaun) as scaffolds for artificial consciousness. The work emphasizes optimistic potential and practical challenges, including measurement, ethics, and the need for machine learning to shape conscious-like awareness.

Abstract

The concept of neural correlates of consciousness (NCC), which suggests that specific neural activities are linked to conscious experiences, has gained widespread acceptance. This acceptance is based on a wealth of evidence from experimental studies, brain imaging techniques such as fMRI and EEG, and theoretical frameworks like integrated information theory (IIT) within neuroscience and the philosophy of mind. This paper explores the potential for artificial consciousness by merging neuromorphic design and architecture with brain simulations. It proposes the Neuromorphic Correlates of Artificial Consciousness (NCAC) as a theoretical framework. While the debate on artificial consciousness remains contentious due to our incomplete grasp of consciousness, this work may raise eyebrows and invite criticism. Nevertheless, this optimistic and forward-thinking approach is fueled by insights from the Human Brain Project, advancements in brain imaging like EEG and fMRI, and recent strides in AI and computing, including quantum and neuromorphic designs. Additionally, this paper outlines how machine learning can play a role in crafting artificial consciousness, aiming to realise machine consciousness and awareness in the future.

Figures (8)

• Figure 1: Schema of the neural processes underlying consciousness, from Christof Koch's research, illustrating the intricate interplay of neural correlates of consciousness (NCC) and their dynamic interactions within the brain's complex network. Koch's work sheds light on the fundamental mechanisms that give rise to conscious experience, offering valuable insights into the nature of cognition and subjective awareness. Image source: koch2004quest
• Figure 2: According to Koch, each conscious experience is associated with a specific integrated state in the brain's posterior hot zone. This hot zone encompasses the parietal, occipital, and temporal lobes of the cerebral cortex. Image Source: koch2018consciousness
• Figure 3: (A) The circles represent the Perturbational Complexity Index (PCI) values derived from cortical responses to transcranial magnetic stimulation (TMS) across various brain stimulation sites. These values are sorted by age and condition, reflecting different states of consciousness, such as non-REM sleep, anesthesia, dreaming, and wakefulness. The solid circles denote the highest PCI value (PCImax) for each individual, while open circles represent lower PCI values. (B) The Area Under the Curve (AUC) is 100%, indicating excellent discriminatory power. Using PCI as the cutoff, both sensitivity and specificity achieve 100%, demonstrating the accuracy of this cutoff in identifying conscious states. (C) The contingency table is created by dividing the PCImax values using the PCI$^*$ cutoff, as illustrated by the dashed horizontal line in panel A. This table categorizes different states, such as emergence from a minimally conscious state (EMCS), locked-in syndrome (LIS), and rapid eye movement (REM), providing a structured view of how the PCI values correlate with various levels of consciousness and brain activity. Image source. casarotto2016stratification
• Figure 4: Left: The topographic structures of $\overline{\Phi}_R$ and EEG connectivity in the alpha band were analyzed across states of consciousness during two anesthetic experiments: (A) ketamine and (B) propofol followed by isoflurane. In both experiments, the first row shows the topography structure of $\overline{\Phi}_R$, while the second row presents the node degree of 96 EEG channels, averaged across subjects and states. These structures and strengths reflect the varying levels of consciousness. For instance, higher $\overline{\Phi}_R$ and node degree in the posterior region during the baseline state is disrupted during anesthesia but partly restored upon regaining responsiveness. While the scales of $\overline{\Phi}_R$ and node degree are consistent within each experiment, they differ between the ketamine and propofol-isoflurane experiments. image source kim2018estimating Right: The impact of anesthesia on system-level integrated information ($F$) and the integrated information structure (IIS: a set of $\phi$ values) is examined in relation to the fruit fly Drosophila melanogaster. (A) $F$ values are compared between wakefulness (red) and anesthesia (blue) across 1365 channel sets, averaged across flies. (B) The ratio of $F$ (wakeful/anesthetized) is shown for all channel sets, averaged across flies. (C) $\phi$ values from the IIS are compared between wakefulness (red) and anesthesia (blue) for different mechanism sizes, averaged across flies and channel sets. (D) The wakeful $\phi$ to anesthetized $\phi$ ratio is presented for each mechanism size, averaged across flies. Image Source leung2021integrated
• Figure 5: (A) Spiking Neural Network: This model represents a neuron (spiking) connected by adjustable synapses (plastic) in an array. The core of the neuron is its membrane, modelled by a special state variable (ferroelectric polarization). (B) Ferroelectric Capacitor: This tiny device (MFM capacitor) uses a thin film (Hf x Zr 1-x O 2) sandwiched between metal plates (W) to study how voltage affects polarization switching. (C) Polarization vs. Electric Field: This graph shows how the material's polarization (P) changes with applied voltage (E). The loops indicate multiple internal regions within the film. (D-E) Mimicking Neurons: Short voltage pulses can be used to study how polarization changes over time. This behaviour resembles how a real neuron's membrane integrates and relaxes with electrical signals. Image Source dutta2020supervised