Johannes Kleiner

The key limitation of contemporary tests for consciousness (C-tests) in non-human animals, infants, and AI is that they are indirect tests: they test for the presence of functions, behaviours, capacities, or states whose relation to or correlation with consciousness has to be known for implications about consciousness to follow, but which is not itself part of the test. Credence in the subsumed relation or correlation is established based on arguments, prior experiments, or theories of consciousness, with corresponding implications for the certainty and construct validity of these tests. Here, we present a method to construct direct tests for consciousness that do not depend on a known relation or correlation between consciousness and a measured property. Rather, the tests rely on the measurement of covariates of the test. Knowledge of correlations between the covariates and consciousness is not required. The method may be applicable to improve known C-tests. As a theoretical example, we discuss the case of infant consciousness. Our result rests on mathematical facts similar to those that have enabled the field of evaluation research in psychology.

The Newman problem is a fundamental problem that threatens to undermine structural assumptions and structural theories throughout philosophy and science. Here, we consider the problem in the context of consciousness science. We introduce and discuss the problem, and explain why it is detrimental not only to structuralist assumptions, but also to theories of consciousness, if left unconsidered. However, we show that if phenomenal spaces, and mathematical structures of conscious experience more generally, are understood in the right way, the Newman problem does not arise. The upshot of this paper is that consciousness science needs to be careful in which definition of structure to consider, but if it is, the Newman problem disappears.

The search for a scientific theory of consciousness should result in theories that are falsifiable. However, here we show that falsification is especially problematic for theories of consciousness. We formally describe the standard experimental setup for testing these theories. Based on a theory’s application to some physical system, such as the brain, testing requires comparing a theory’s predicted experience (given some internal observables of the system like brain imaging data) with an inferred experience (using report or behavior). If there is a mismatch between inference and prediction, a theory is falsified. We show that if inference and prediction are independent, it follows that any minimally informative theory of consciousness is automatically falsified. This is deeply problematic since the field’s reliance on report or behavior to infer conscious experiences implies such independence, so this fragility affects many contemporary theories of consciousness. Furthermore, we show that if inference and prediction are strictly dependent, it follows that a theory is unfalsifiable. This affects theories which claim consciousness to be determined by report or behavior. Finally, we explore possible ways out of this dilemma.

This paper presents a mathematical analysis of the logic of measuring Neural Correlates of Consciousness (NCCs). Starting from the canonical definition of NCCs provided by Crick and Koch (1990) and Chalmers (2000), a series of lemmas and theorems are provided which show how NCCs can be discovered if empirical data about the co-activation of neural states and states of consciousness is available. The result of this analysis is a new method to measure NCCs, which we preliminarily call Co-Activation Analysis (CoAA), that might complement or extend existing methods such as contrastive analysis and decoding. CoAA might be of interest for experiments because it does not require data from near-threshold conditions, is compatible with all major conceptions of states of consciousness (including, e.g., micro-phenomenological notions and global states of consciousness), can be applied to most conceptions of neural or computational states (including, e.g., those provided by Predictive Processing/Active Inference, or those attained in more ecological conditions), and helps alleviate the problem of confounders. Furthermore, we show as a theorem that as far as the logic of measurement is concerned, if applied to data from contrastive analysis studies, CoAA improves upon the result provided by contrastive analysis. The paper is a purely theoretical contribution. It is presented in the hope that the mathematics developed here can support and improve the empirical search for NCCs.

The scientific study of consciousness, also referred to as consciousness science, is a young scientific field devoted to understanding how conscious experiences and the brain relate. It comprises a host of theories, experiments, and analyses that aim to investigate the problem of consciousness empirically, theoretically, and conceptually. This thesis addresses some of the questions that arise in these investigations from a formal and mathematical perspective. These questions concern theories of consciousness, experimental paradigms, methodology, and artificial consciousness. Regarding theories of consciousness, the thesis contributes to the understanding of the mathematical structure that some of the formal theories in the field propose. The work presented here targets the theory of consciousness known as Integrated Information Theory (IIT) and the neuroscientific theory known as Predictive Processing Theory or Free Energy Principle in its Active Inference form (AI-PP). The thesis provides axiomatic definitions of the mathematical structures that constitute these theories, and uses these definitions to address some of the open questions surrounding the theories. For AI-PP, this includes a rigorous derivation of the formula for Active Inference via Free Energy minimisation and a proof of compositionality of Free Energy. For IIT, this includes resolutions of some of the criticisms of IIT's formal scope and applications, but also the identification of new issues that concern the formalism and its derivation. When possible, the definitions are provided in the mathematical framework of category theory. Regarding experiments, the thesis addresses the main paradigm for testing and falsifying theories of consciousness currently applied in the field. This paradigm consists of comparing the conscious experience that a theory predicts with the conscious experience that is inferred from behavioural data or report by use of measures of consciousness. The thesis provides a formal model of this paradigm and shows that under a certain condition—if inference and prediction are independent—, any minimally informative theory of consciousness can always be falsified. This is deeply problematic since the field’s reliance on report or behaviour to infer conscious experiences, in conjunction with the general structure of most contemporary theories of consciousness, implies such independence. This observation provides the exact formal underpinning of the well-known unfolding argument. The thesis analyses the origin of the problem and identifies precisely which changes are required to avoid this problem in future research. The thesis furthermore shows that the problem of falsifying theories of consciousness, and of empirical comparisons of theories of consciousness more generally, follows from a pervasive closure paradigm in consciousness science, that consists of taking a neuroscientific account of the brain as input to a theory of consciousness, so as to explain what consciousness is, without allowing for modifications or adaptations of the neuroscientific account that would accommodate consciousness as part of the brain's functioning. As is shown in the thesis, this paradigm has implications that point to a fundamental need of revision. Regarding methodological and conceptual questions, the thesis contributes to the foundations of structural research in consciousness science. Structural research aims to use mathematical structures or mathematical spaces, instead of verbal descriptions or simple categorisations, to represent conscious experiences scientifically, for example when building theories of consciousness, or when exploring new empirical avenues to measure consciousness. Despite considerable advances in this realm, there was, prior to this thesis, no explicit definition of what a mathematical structure of conscious experience should be; that is, how the attribution of mathematical structure to conscious experiences should be systematically understood. Perhaps the most important contribution of this thesis to the field is to propose such a definition. The definition, a structural concept, extends existing approaches wherever available, and provides a basis for developing a common formal language to study consciousness, bridging developments as far apart as psychophysics and phenomenology. In addition, and independently of this proposal, the thesis offers a critical analysis of which metaphysical premises need to be presumed in structural research, whether the use of particular formal tools (such as structure-preserving mappings or homomorphisms) is justified, and how structural theories of consciousness could otherwise be built in the first place. An attempt to expand the results from consciousness to more general problems in philosophy of science is made in the context of the well-known Newman problem. Regarding the question of artificial consciousness—can AI feel?—, the thesis contributes two results that take the form of no-go theorems, as known from physics. The first no-go theorem shows that if consciousness is relevant for the temporal evolution of a system's states—if it is dynamically relevant—, then contemporary AI systems cannot be conscious. That is because AI systems run on CPUs, GPUs, TPUs or other processors which have been designed and verified to adhere to computational dynamics that systematically preclude or suppress deviations. The second no-go theorem is situated in the context of computational functionalism, a view which posits that consciousness is a computation. The theorem shows that if computational functionalism holds true, consciousness cannot be a Turing computation. Rather, it must be a novel type of computation that has recently been proposed by Geoffrey Hinton, called mortal computation. This thesis is part of a global effort to pioneer a mathematical perspective in consciousness science, now called Mathematical Consciousness Science. The hope behind the research carried out in this PhD is to illustrate the power and usefulness of mathematical approaches in different areas of consciousness science, and in doing so, to lay the foundations for future mathematical work that complements and supports empirical and theoretical work in the further development of this exciting field.

The Newman problem is a fundamental problem that threatens to undermine structural assumptions and structural theories throughout philosophy and science. Here, we consider the problem in the context of consciousness science. We introduce and discuss the problem, and explain why it is detrimental not only to structuralist assumptions, but also to theories of consciousness, if left unconsidered. However, we show that if phenomenal spaces, and mathematical structures of conscious experience more generally, are understood in the right way, the Newman problem does not arise. The upshot of our paper is that consciousness science needs to be careful in which definition of structure to consider, but if it is, the Newman problem disappears.

Recent activities in virtually all fields engaged in consciousness studies indicate early signs of a structural turn, where verbal descriptions or simple formalisations of conscious experiences are replaced by structural tools, most notably mathematical spaces. My goal here is to offer three comments that, in my opinion, are essential to avoid misunderstandings in these developments early on. These comments concern metaphysical premises of structural approaches, the viability of structure-preserving mappings, and the question of what a structure of conscious experience is in the first place. I will also explain what, in my opinion, are the great promises of structural methodologies and how they might impact consciousness science at large.

Several promising approaches have been developed to represent conscious experience in terms of mathematical spaces and structures. What is missing, however, is an explicit definition of what a ‘mathematical structure of conscious experience’ is. Here, we propose such a definition. This definition provides a link between the abstract formal entities of mathematics and the concreta of conscious experience; it complements recent approaches that study quality spaces, qualia spaces, or phenomenal spaces; and it provides a general method to identify and investigate structures of conscious experience. We hope that ultimately this work provides a basis for developing a common formal language to study consciousness.

Integrated Information Theory is one of the leading models of consciousness. It aims to describe both the quality and quantity of the conscious experience of a physical system, such as the brain, in a particular state. In this contribution, we propound the mathematical structure of the theory, separating the essentials from auxiliary formal tools. We provide a definition of a generalized IIT which has IIT 3.0 of Tononi et al., as well as the Quantum IIT introduced by Zanardi et al. as special cases. This provides an axiomatic definition of the theory which may serve as the starting point for future formal investigations and as an introduction suitable for researchers with a formal background.

We demonstrate that if consciousness is relevant for the temporal evolution of a system's states—that is, if it is dynamically relevant—then AI systems cannot be conscious. That is because AI systems run on CPUs, GPUs, TPUs or other processors which have been designed and verified to adhere to computational dynamics that systematically preclude or suppress deviations. The design and verification preclude or suppress, in particular, potential consciousness-related dynamical effects, so that if consciousness is dynamically relevant, AI systems cannot be conscious.

Recent activities in virtually all fields engaged in consciousness studies indicate early signs of a structural turn, where verbal descriptions or simple formalisations of conscious experiences are replaced by structural tools, most notably mathematical spaces. My goal here is to offer three comments that, in my opinion, are essential to avoid misunderstandings in these developments early on. These comments concern metaphysical premises of structuralist approaches, overlooked assumptions in regard to isomorphisms, and the question of what structure to consider on the side of consciousness in the first place. I will also explain what, in my opinion, are the great promises of structural methodologies and how they might impact consciousness science at large.

In consciousness science, several promising approaches have been developed for how to represent conscious experience in terms of mathematical spaces and structures. What is missing, however, is an explicit definition of what a 'mathematical structure of conscious experience' is. Here, we propose such a definition. This definition provides a link between the abstract formal entities of mathematics and the concreta of conscious experience; it complements recent approaches that study quality spaces, qualia spaces or phenomenal spaces; it provides a general method to identify and investigate structures of conscious experience; and it may serve as a framework to unify the various approaches from different fields. We hope that ultimately this work provides a basis for developing a common formal language to study consciousness.