Lukas Kob

This paper investigates the extent to which vector space representations in large language models (LLMs) can be understood as holistic representations. It is argued that holistic principles influence both the training process and the generation of output in LLMs. To contextualize this claim, debates on representation in psychology and philosophy will be presented, with a focus on the distinction between indicator and structural representations which recently emerged in philosophy of cognitive science. The holistic properties of structural representations are introduced as a key feature that distinguishes them from indicator representations and provides a basis for analyzing holism in LLMs. Two novel criteria for holism are developed: holistic perturbability and holistic exploitability. It is argued that LLMs meet both criteria. Moreover, holistic exploitation in output generation provides a novel way to understand the exploitability of structural representations in general, possibly also applicable in the human brain. Classic counter-arguments to holism and their implications for the question of semantic grounding are discussed. Finally, implications for the homunculus problem, gestalt perception, and development are presented.

Recently, several authors have argued that the current impasse in consciousness science supports a pluralistic interpretation: the view that seemingly contradictory findings may both be valid because consciousness can be brought about in different ways, each sufficient to generate conscious experience.We analyze different versions of pluralism to determine under what conditions it offers a viable interpretation of consciousness science. We argue that viable pluralism requires a two-factor framework distinguishing general consciousness factors(mechanismsrendering contents conscious) from content encoding itself. Moreover, NCC kinds must be finite in number and systematically ordered to enable a unified scientific model. While pluralism enables testable hypotheses, significant ontological tensions arise: how can consciousness constitute a unified natural kind if realized by fundamentally distinct NCC kinds? Drawing on philosophy of science, we propose an alternative interpretation in terms of scientific rather than natural kinds. Our goal is not to advocate for pluralism, but rather, to critically examine its viability and implications. Based on this, we urge the field to engage more explicitly with the foundational assumptions that shape consciousness research.

Traditional contrastive analysis has been the foundation of consciousness science, but its limitations due to the lack of a reliable method for measuring states of consciousness have prompted the exploration of alternative approaches. Structuralist theories have gained attention as an alternative that focuses on the structural properties of phenomenal experience and seeks to identify their neural encoding via structural similarities between quality spaces and neural state spaces. However, the intertwining of philosophical assumptions about structuralism and structuralist methodology may pose a challenge to those who are skeptical of the former. In this paper, I offer an analysis and defense of structuralism as a methodological approach in consciousness science, which is partly independent of structuralist assumptions on the nature of consciousness. By doing so, I aim to make structuralist methodology more accessible to a broader scientific and philosophical audience. I situate methodological structuralism in the context of questions concerning mental representation, psychophysical measurement, holism, and functional relevance of neural processes. At last, I analyze the relationship between the structural approach and the distinction between conscious and unconscious states.

Researchers on the neural correlates of consciousness (NCC) need to distinguish mere statistical NCCs from NCCs proper. Some neural events may be co-occurrent, probabilistically coupled, or coincidental with a type of conscious experience but lack any deeper connection to it, while in other cases, the relation between neural states and a type of experience hints at a strong metaphysical relation, which distinguishes such NCCs proper from mere statistical NCCs. In order to address this issue of how to distinguish NCCs proper from mere statistical NCCs, we propose a position we call neurophenomenal structuralism. The position hinges on the uncontroversial idea that phenomenal experiences relate to each other in degrees of similarity and difference. These complex structures are used to identify and individuate experiences in the methods of neuroscience, psychophysics, and phenomenology. Such individuation by structure leads to phenomenal holism, which has implications for how to investigate consciousness neuroscientifically and generates a constraint by which we can distinguish NCCs proper from mere statistical NCCs: the structural similarity constraint. Neural activation must preserve the structure governing the domain of experiences it is associated with in order to count as that domain’s NCC proper. Any activation that fails to preserve phenomenal structure fails to be an NCC proper. We illustrate how this constraint works with a study by Brouwer & Heeger (2009) as an example.

Methodological structuralism is a research program that seeks to identify neural correlates of consciousness (NCCs) by mapping phenomenal similarity relationships onto the similarity relations between neural population activity. This paper presents a discussion of the potential benefits of methodological structuralism for the neurosciences of consciousness, namely as a specific theory of neural content encoding. In order to achieve this, I supplement it with a metatheoretical framework concerning the relationship between content and consciousness: the two-factor interaction view. Although structuralism provides a comprehensive description of the neural encoding of content, it is inadequate for fully explaining the conscious experience of contents. The majority of current theories of consciousness posit the existence of an additional mechanism that underlies the conscious experience of content. Consequently, if structuralism is indeed correct, progress in consciousness science can be achieved by investigating the interactions between neural mechanisms responsible for consciousness and structures in neural population code activity accounting for the structure of contents. This also has significant implications for consciousness in AI. I discuss these implications, as well as potential empirical avenues for investigating the interaction between content structures and consciousness with cutting-edge neuroscientific methodologies.