Metrics of Consciousness in Embodied 2D-3D Neural Network Hybrids

1. Overview

Scientific Rationale

Considering the close ties between consciousness and cognition, the generation and validation of an in vitro minimal cognitive systems (MCS) in the context of consciousness can dramatically improve both the speed and ease of research. As they relate to consciousness, in vitro systems are currently hampered by both the lack of embodiment, which yields spasmodic, random cellular activation, and a distinct lack of structural similarity to human cortical columns. Several objective measures of consciousness have been proposed, based on both Integrated Information Theory (IIT) and the Critical Brain Hypothesis (CBH). Thus, the current project aims to validate the use of MCS in consciousness research by determining these measures using induced pluripotent stem cell (iPSC)-based neuronal cultures undergoing varying levels of embodiment and constructed with various levels of complexity.

Confirmation of funding/facilities/ethics approvals

All work has been approved for work in a BSL-2 laboratory approved for work with iPSCs. This work aligns with Gigabrain Track 3. This work focuses on directly measuring both the level of network synchronization and comparative organoid intelligence.

Estimated Timeline

Data Sharing Commitment

We plan on publishing the results of this study through open access and making the data available to the public upon request as well as on ResearchHub.

2. Introduction

Consciousness, or subjective experience, cannot be directly observed; however, several objective correlates have been identified that vary systematically with waking, sleeping, dreaming, and altered states.¹Validating objective “metrics” of consciousness would enable clinicians to detect the inner lives of minimally aware coma patients and fetuses. Minimal cognitive systems (MCS) are in vitro neural tissue cultures that enable investigators to identify the basic building blocks of mental states and actions.² By building cognition from its composite building blocks and assessing what is sufficient for the emergence of consciousness, as defined by metrics grounded in theories of how mind emerges from matter, we can begin to make progress toward a grounded science of minds.^2,3

Integrated Information Theory (IIT) and the Critical Brain Hypothesis (CBH) are two theories of consciousness that provide objective, albeit indirect, measures of experience.^4–6 IIT is based on phenomenological axioms which are expressed in mathematical terms, and provide falsifiable claims based on measurable values of irreducible cause-effect power in a given system at particular spatial and temporal grain.⁷ These measures have been demonstrated as reliable predictors of the modulation of consciousness through sedation.^8,9 The CBH provides similarly falsifiable claims about the substrate of consciousness, proposing that cortical dynamics operate near a critical phase transition between order and disorder. Such systems exhibit scale-free activity and long-range correlations where information is retained as it propagates throughout the system.⁵Some experimental evidence suggests that these theories are mutually validating, providing multiple independent measures that align in the presence of consciousness.^6,10,11

MCSs provide an opportunity to explore these theories of consciousness in vitro, where we can build complexity in a stepwise fashion and identify under which conditions objective measures of consciousness, supported by IIT and CBH, arise mutually. Importantly, cultured neurons exhibit aberrant, paroxysmal ‘burst’ firing patterns due to a lack of exogenous stimulus - sending signals that will never be interpreted or modulated through normal feedback mechanisms - an abnormal condition which would hinder the development of an MCS¹². In 2005, Potter, Wagnaar, and their colleagues suggested that this behaviour, reminiscent of epileptic events, inhibits learning, and demonstrated that random direct electric stimulation transiently normalized the bursting phenotype.¹² This connects well with experimental evidence suggesting that a given system must be embodied, meaning to observe and partake in the world, before they can begin to exhibit the properties which we associate with consciousness; moreover random exogenous stimulation fits the criteria of a ‘minimal form of embodiment’.^13–15One such MCS has recently been developed on the MaxOne HD-MEA (high-definition microelectrode array) system and through basic embodiment via closed-loop feedback stimulation, learned to play Pong, thus generating non-spasmodic, goal-directed signals.¹⁴Although intentionality could be considered a proxy of “sentience” - a term very closely associated with consciousness - these researchers fell short of obtaining any measurable parameters which would take this claim out of the subjective and into the objective.^6,14

Beyond the lack of objective measures of consciousness, the MCS trained to play Pong, like most MCSs to date, rely on planar monolayer (two-dimensional or “2D”) neuronal cultures that are vastly different from the three-dimensional (3D) biological circuits that give rise to human consciousness.¹⁴The mammalian neocortex achieves its computational power through highly structured vertical layering, where information flows between interconnected strata of excitatory and inhibitory neurons that act as reservoirs; supporting sensory integration, predictive coding, and higher-order cognition, and likely consciousness.¹⁶ Considering that the metrics provided by IIT and CBH are dependent on the complexity of a system, it may be that the incorporation of the three-dimensionality associated with neuronal reservoirs may serve to increase the associated metrics of consciousness.

Hypothesis:

Minimal levels of embodiment will increase the objective measures of consciousness within said systems.

Objectives:

The aims of the present study are to determine objective metrics of consciousness in MCSs as they arise from:

(1) Increasingly complex stimulation patterns ranging from no stimulation to closed-loop feedback

In pursuit of building an MCS to expand our understanding of the conditions in which consciousness emerges, we aim to expand upon his previous findings by broadening the repertoire of stimulation patterns to deepen our understanding of suppressing spasmodic burst and its effect on the objective measures of consciousness.

(2) Varying levels of dimensionality and underlying spatial complexity:

First, we will engineer vertically integrated neuronal reservoirs from (iPSC)-derived neurons cultured in layered architectures on multi-electrode arrays (MEAs). The reservoirs will be constructed by layering neuronal spheroids on top of a two-dimensional monolayer of iPSC-derived neurons. The spheroids act as dense, three-dimensional computational units, while the monolayer serves as a base for anchoring and recording through the MEA. This vertical architecture increases connectivity across layers, enabling richer spatiotemporal dynamics than flat cultures.

Impact Statement:

Current research relies heavily on neural models which exhibit aberrant bursting activity which is not physiologically relevant to an array of neurodegenerative, neuropsychiatric, and cybernetic neuroscientific research. This project aims to further elucidate the types of stimuli and three-dimensional structures that are sufficient and required to achieve physiologically relevant network activity in neural cultures, establishing standards which might result in better translatable in vitro models of the central nervous system for research. Moreover, this stepwise approach of minimal embodiment will afford us a glimpse into the most fundamental conditions of conscious emergence while mutually validating two strong candidate theories of consciousness, IIT and CBH, by measuring and comparing their quantifiable claims of the physical correlates of consciousness.^1,2,17

3. Methods

Materials and Procedures

This project will employ a multidisciplinary approach integrating stem cell biology, microfabrication and electrophysiology. To achieve the stated objectives iPSCs will be differentiated into neurons, cultured on MEA chips, and subjected to three levels of embodiment: no stimulation, spatially and temporally unpredictable electrostimulation and closed-loop feedback, where stimulation input along a predefined ‘sensory’ region will depend on output from a predefined ‘motor’ region. If neurons in this motor regions exceed 5x their spontaneous firing rate (a frequency plausibly above noise), it will elicit stimulation in the sensory region.^13,18 Consciousness parameters Φ, Lempel-Ziv complexity, and power law scaling will be assessed through network activity recordings at each level of embodiment, to examine embodiment-dependent changes to metrics of consciousness.^1,5,12

The iPSC-derived neurons will also be used increase vertical integration through the production of organoids stabilized using polydimethylsiloxane (PDMS) molds and collagen-based scaffolds; enabling layered connectivity reminiscent of cortical microcircuits. These organoids can then be used to assess the effects of spatial patterning and dimensionality on the objective measures of consciousness.

We will leverage existing collaborative partnerships with leaders in minimal cognition (Michael Levin, Tufts University), hybrid biological intelligences (Steve M. Potter, Georgia Institute of Technology), and IIT (Larissa Albantakis, University of Wisconsin-Madison) to enhance theoretical and experimental approaches outlined here.

Planned Analyses

The calculation of the metrics of consciousness will be measured based on first binning neurons within the 2D monolayer based on spatial positioning, after which neurons in each bin will be measured based on neuronal activation, as measured through average neuronal firing rate and spike intensities. Non-random neuronal firing will result in an increased coherence of firing activity and thus the activity in each bin will be normalized based on total firing within the MEA chip area. Temporal grain will be established by determining the optimal scale at which the transition probability matrices will be populate (between 1 and 100ms). Data will be binarized based on a determined threshold and measures will be calculated using PyPhi (publicly available from https://github.com/wmayner/pyphi).

Lempel-Ziv complexity will be assessed by choosing an analysis object (individual electrodes or pooled populations), binning them temporally (1-100ms), and similarly binarized based on a median firing rate. A binary sequence P = s(1), s(2) … (s) is created by comparing a sample of the signal x(i) with a median of the time series (Td) such that s(i) is 0 if x(i) < T_d and 1 if x(i) ≥ T_d.

The parsing process defined by Lempel and Ziv (1976)¹⁹ outlined below will be utilized to derive complexity from these binary sequences:

Where n is the length of the sequence, c(n) is the Lempel-Ziv complexity count; the number of new patterns encountered as the sequence is scanned, and n/log2(n) as the normalization factor, ensuring that complexity is not scaled linearly with sequence length. The final value CLZ being close to 0 indicates a highly repetitive sequence, and close to 1 indicates a highly complex sequence.

Scale-free avalanche measures will be derived by binning spikes in time across an MEA chip. Avalanches are defined as a consecutive run of non-empty bins bounded by empty bins. The total spikes in an avalanche and number of consecutive active bins will be plotted on log-log axes, and the extent to which they fit to a scale-free power law will be assessed.²⁰

Analytical Contingencies

IF repeated measurements are obtained from the same chip or recording over time, THEN analyses will account for within-sample dependence by using paired or repeated-measures statistical approaches rather than treating observations as independent.

IF a recording fails predefined quality-control criteria (e.g., low activity or excessive noise), THEN it will be excluded from downstream complexity and avalanche analyses.

IF multiple metrics and spatial or temporal bins are compared simultaneously, THEN p-values will be adjusted for multiple comparisons with a false discovery rate correction.

Ethics and Data Management

All the presented work will be done in a BSL-2 laboratory that is currently approved for work with iPSCs as outlined by Wilfrid Laurier University.

5. Budget

Personnel - $12000

This project is expected to be completed through the part-time efforts of two graduate students and one post-doctoral scholar. These individuals have been selected due to previous experience in the field and as such should allow for rapid generation of results following the onset of the project.

Consumables - $9000

Much of the cost of consumables will be from two sources: purchasing of the MEA chips, and cell culture media and supplements for the iPSCs.

Publication/Dissemination - $4000

This includes publishing fees for an open access journal as well as conference registration fees.

6. References

1. Rouleau, N. Embodied 3D Neural Tissue Cultures for Cognitive Research. in Brain and Cognitive Intelligence Control in Robotics 81–101 (CRC Press, Boca Raton, 2022).

2. Smirnova, L. et al. Organoid intelligence (OI): the new frontier in biocomputing and intelligence-in-a-dish. Front. Sci. 1, (2023).

3. Watanabe, M. From Biological to Artificial Consciousness: Neuroscientific Insights and Progress. (Springer International Publishing, Cham, 2022). doi:10.1007/978-3-030-91138-6.

4. Oizumi, M., Albantakis, L. & Tononi, G. From the Phenomenology to the Mechanisms of Consciousness: Integrated Information Theory 3.0. PLoS Comput. Biol. 10, e1003588 (2014).

5. O’Byrne, J. & Jerbi, K. How critical is brain criticality? Trends Neurosci. 45, 820–837 (2022).

6. Popiel, N. J. M. et al. The Emergence of Integrated Information, Complexity, and ‘Consciousness’ at Criticality. Entropy 22, 339 (2020).

7. Albantakis, L. et al. Integrated information theory (IIT) 4.0: Formulating the properties of phenomenal existence in physical terms. PLOS Comput. Biol. 19, e1011465 (2023).

8. Leung, A., Cohen, D., Swinderen, B. van & Tsuchiya, N. Integrated information structure collapses with anesthetic loss of conscious arousal in Drosophila melanogaster. PLOS Comput. Biol. 17, e1008722 (2021).

9. Kim, H. et al. Estimating the Integrated Information Measure Phi from High-Density Electroencephalography during States of Consciousness in Humans. Front. Hum. Neurosci. 12, 42 (2018).

10. Kim, H. & Lee, U. Criticality as a Determinant of Integrated Information Φ in Human Brain Networks. Entropy 21, 981 (2019).

11. Walter, N. & Hinterberger, T. Self-organized criticality as a framework for consciousness: A review study. Front. Psychol. 13, 911620 (2022).

12. Wagenaar, D. A., Madhavan, R., Pine, J. & Potter, S. M. Controlling Bursting in Cortical Cultures with Closed-Loop Multi-Electrode Stimulation. J. Neurosci. 25, 680–688 (2005).

13. Moon, K. & Pae, H. Making Sense of Consciousness as Integrated Information: Evolution and Issues of Integrated Information Theory. J. Cogn. Sci. 20, 1–52 (2019).

14. Kagan, B. J. et al. In vitro neurons learn and exhibit sentience when embodied in a simulated game-world. Neuron 110, 3952-3969.e8 (2022).

15. Wilson, A. D. & Golonka, S. Embodied cognition is not what you think it is. Front. Psychol. 4, (2013).

16. van den Heuvel, M. P. & Sporns, O. Rich-club organization of the human connectome. J. Neurosci. Off. J. Soc. Neurosci. 31, 15775–15786 (2011).

17. Bettinger, J. S. & Friston, K. J. Conceptual foundations of physiological regulation incorporating the free energy principle and self-organized criticality. Neurosci. Biobehav. Rev. 155, 105459 (2023).

18. Pezzulo, G., Parr, T. & Friston, K. The evolution of brain architectures for predictive coding and active inference. Philos. Trans. R. Soc. B Biol. Sci. 377, 20200531 (2022).

19. Lempel, A. & Ziv, J. On the Complexity of Finite Sequences. IEEE Trans. Inf. Theory 22, 75–81 (1976).

20. Klaus, A., Yu, S. & Plenz, D. Statistical Analyses Support Power Law Distributions Found in Neuronal Avalanches. PLoS ONE 6, e19779 (2011).

Metrics of Consciousness in Embodied 2D-3D Neural Network Hybrids

Metrics of Consciousness in Embodied 2D-3D Neural Network Hybrids

Funding Progress

Peer Reviews

Related Funding

Request for Proposals - Project Gigabrain (Brain Organoids)

Nonprofit Recipient

Topics

DOI