Within the contemporary landscape of consciousness science, the majority of theoretical approaches gather fundamental data for theory-building from the observation of behavioural outcomes, such as subjective report, in conjunction with brain imaging data (Chalmers 2000). These approaches hold that without neural and behavioural data we cannot possibly do meaningful science of consciousness, since without neural and behavioural data we would lack the objective observables at the basis of any scientific investigation (Doerig et al. 2019; Herzog et al. 2022). The starting point of consciousness science must thus coincide with experiments that pick out correspondences between conscious states and objectively (neural-behavioural) states.
Some of the most influential theories of consciousness committed to the this type of approach are the Global Neuronal Workspace Theory (GNWT) (Mashour et al. 2020), Higher-order theories (Brown et al. 2019), Attention Schema Theory (Graziano and Webb 2015), and Recurrent Processing Theory (RPT) (Lamme 2006, 2010). These theories gather data mainly from experiments where a significant correlation is found between specific subjective reports, or other behavioural outcomes, and specific neural properties. This observation is explained by the hypothesis that a particular neural property underpins the conscious state the subject is reporting.
In the context of experiment-driven approaches, the main problem is to understand the exact relation between reportability (or at least cognitive accessibility) and phenomenal consciousness, since some research programmes require a distinction between the mechanisms underpinning these two phenomena (Block 1995, 2011; Lamme 2018), whereas others hold that studying consciousness independently of any behavioural outcome or report is simply outside the scope of science (Cohen and Dennett 2011; Doerig et al. 2019; Naccache and Dehaene 2007; for a comprehensive discussion, see (Kleiner and Hoel 2021)).
Thus, within experiment-driven approaches, the debate seems mostly focused on (i) whether data about cognitive reportability should be separated from data about phenomenal consciousness, and (ii) if so, how to do it. Experiment-driven consciousness science, then, seems to have two issues. First, there is a theoretical problem: there is no consensus on which concept of consciousness should be used in consciousness research, since it is not clear whether phenomenal consciousness is identical with cognitive access or not, nor whether this distinction is useful at all (Block 1995; Irvine 2017; Overgaard and Grünbaum 2012). Second, there is a methodological problem: since there is no consensus on how to conceptualize the explanandum phenomenon (the phenomenon that needs to be explained), there is no well-established methodology for studying consciousness scientifically (Irvine 2012), and how to set it apart from all possible confounds (for a discussion, see (Michel and Morales 2020))Footnote 1.
At the present stage, theories of consciousness based on the experiment-first approach seems to be systematically underdetermined (Michel 2019), namely, the same data can be accounted for by different models with equal explanatory and predictive power: the claim that a certain model is best can be countered by proponents of a different model, who operate under different theoretical and methodological assumptions. This situation seems to lead to an impasse, which can be summarized in this way: at the present stage, theories employing an experiment-driven approach do not have a well-defined methodology for deciding which model of consciousness fits best with currently available data.
This is acknowledged by several consciousness scholars. Victor Lamme states that “the debate [between theories that focus on phenomenal consciousness and those that focus on conscious access] seems to reach a stalemate” (Lamme 2018, p. 1), whereas Phillips (2018) argues that “not only do we not know whether consciousness requires cognition, we do not know how to find out” (Phillips 2018, p. 7).
Perhaps this impasse can be overcome from within the experiment-driven approach: finessing consciousness detection procedures and calibrating different measures of consciousness (Michel 2021; Seth et al. 2008), paired with accumulation of new evidence, might help decide between competing models of consciousness.
In this paper, however, I want to explore the prospects and limitations of a different strategy. This is the strategy of claiming that, if we want to advance consciousness science, we must approach consciousness from a different angle, and this angle is provided by phenomenology itself. That is, in order to explain consciousness, we must start from consciousness itself. This is the “phenomenology-first” approach taken by IIT.
IIT is by no means the only research programme that starts with first-person data (e.g. (Rudrauf et al. 2017; Varela 1996; Williford et al. 2018); for a discussion, see (Chalmers 2004)), but its centrality in current neuroscientific debates on the neural basis of consciousness and its clear and well-developed theoretical structure place IIT at the forefront of phenomenology-first approaches. For these reasons, I will focus on IIT’s specific version of developing the phenomenology-first approach.
According to IIT’s phenomenology-first approach, although we need neural and behavioural data to test and validate theories of consciousness (IIT included), if we want to build one we need to extract the essential structure of experience from our own phenomenology. Phenomenologically gathered data are observations that are not publicly available and objectively measurable as traditionally conceived in science. Rather, they constitute a different kind of evidence, since they are observations of how things appear to the subject. A phenomenology-first approach to consciousness science can claim that the features that different appearances have in common just constitute phenomenal (from the Greek phainomenon, namely “that which appears”) consciousness. In this sense, according to the phenomenology-first approach, phenomenologically gathered data, or first-person data, can be used scientifically if we acknowledge that the science of consciousness is the science of first-person appearances, and therefore how things appear to the subject must constrain and guide our understanding of consciousness itselfFootnote 2.
This approach promises to resolve the theoretical and methodological issues that afflict experiment-driven approaches. First, it posits that phenomenal consciousness is the appropriate explanandum of consciousness science because it is manifest as an observational datum. Second, the phenomenology-first approach does not require an account of the relation between consciousness and reportability, since phenomenological observations are not data for an extrinsic observer. Rather, they are data for the same subject who is experiencing them. If the phenomenological observation of consciousness is direct, and does not involve neural and behavioural data, there is no risk of bringing confounding factors such as post-perceptual processes into the fundamental dataset upon which we build our theory of consciousness.
To clarify, according to IIT’s phenomenology-first approach, phenomenological data are not subjective reports about consciousness, nor judgments or beliefs about one’s own consciousness. It is not that there are two things: consciousness and the observation of consciousness. Rather, IIT seems to work with an account of introspection which is akin to the acquaintance view of introspection (Gertler 2012; Russell 1912): the idea is that we are in direct contact with our conscious states, and that there is no medium between how consciousness is and how it appears to the conscious subject. According to Russell’s test, you are acquainted with something if you cannot possibly doubt about the object’s existence. This seems in fact the (Cartesian) starting point of IIT, its zeroth postulate (Barbosa et al. 2021, p. 2): consciousness exists and this much cannot be doubtedFootnote 3.
If the acquaintance view of introspection is on the right track, in the case of consciousness, the distinction between appearance and reality vanishes (Kripke 1980): the “real thing” is precisely phenomenology itself (i.e., how things appear)Footnote 4.
In this paper, I am not interested in assessing whether this is a convincing view of how introspection works (for criticisms of the acquaintance view see (Hill 1991); see also (Dennett 1991) and (Schwitzgebel 2011) for criticisms against introspection in general, and (Spener 2013) for a discussion). I will concede the point for the sake of the argument. Rather, my point here is that this view of how self-knowledge works seems to imply an important consequence for our explanatory practices in consciousness science: if there is no medium between consciousness and the subjective observation of consciousness itself, then the datum (i.e., the observation) and the explanandum phenomenon (i.e., consciousness) coincide.
Thus, IIT’s approach seems to reject, in consciousness science, the distinction between data and phenomena (Bogen and Woodward 1988, p. 305) that the traditional, experiment-driven, approach implies. This is because in consciousness science the observer and the observed subject matter coincide: the phenomenon to be explained is precisely the immediate and direct experience the observer is having.
IIT’s approach might have the advantage of not requiring an account of the relation between consciousness and reportability, but is founded on a relation between data and phenomena that is scientifically unorthodox, and it is not at all clear that a scientific explanation of consciousness can be provided by starting from phenomenological data (Herzog et al. 2022).
It is thus important to explore whether IIT’s phenomenology approach can in fact use phenomenologically gathered data in order to provide a convincing explanation of consciousness. In order to do this, I will adopt a conditional strategy: I will thus accept for the sake of the argument that (i) IIT’s phenomenology-first approach can be used to guide scientific enquiry on consciousness; (ii) IIT’s foundation is in fact able to extract all the essential features of phenomenal consciousness, and (iii) that these features can be singled out via acquaintance with phenomenality itselfFootnote 5. I will argue, however, that even if we accept these foundational aspects of IIT, a problem for IIT as a scientific explanation of consciousness still remains (i.e., what I call “the self-evidencing” problem).
In particular, in order to evaluate whether IIT’s methodology can truly be superior to the experiment-driven approach, we need to look at the particular structure of IIT first, and determine (i) whether IIT’s phenomenology-first approach is able to “bootstrap” an explanation of consciousness from consciousness itself (Ellia et al. 2021, p. 10); and (ii) whether such an explanation avoids the impasse faced by the experiment-driven approach. So, let us examine IIT’s structure first.