How well do we know our own minds? Descartes thought we have transparent access to our mental states—that the nature of thoughts, beliefs, and desires is fully revealed to us. But work in philosophy and psychology suggests this is wrong. We may be far more opaque to ourselves than we think, prone to confabulation and self-deception, systematically mistaken about what we believe and why we act.
My work challenges the transparency assumption and explores the mechanisms and limits of introspection. It addresses implicit bias, delusion, the nature of belief, and what forms self-knowledge might take in humans, animals, and AI. Central is a two-level framework distinguishing implicit mental states (non-conscious, passively formed, guiding unreflective behaviour) from explicit mental states (conscious commitments guiding deliberate reasoning and action). This framework partially vindicates Cartesian intuitions while explaining systematic opacity. We have performative authority over our explicit attitudes but only interpretive access to our implicit ones.
Core questions
The limits of introspection
Do we have direct access to our thoughts and beliefs, or do we infer them from observing ourselves? For implicit attitudes, I argue for an interpretive account: we know our basic-level minds by interpreting sensory evidence—perceptions of our speech, behaviour, emotional responses, and mental imagery. Self-knowledge here is fallible. We may sincerely believe we hold certain views when our behaviour reveals otherwise. For explicit attitudes (premises and goals), the situation is different: avowals are performative, and we have a special authority grounded in commitment rather than observation.
Introspective illusion
Introspection is not only fallible but systematically misleading. Central to the illusionist view of consciousness is that introspection selectively highlights certain features of our mental lives while obscuring others. It distorts and caricatures what it represents, making us believe (falsely) that our experiences possess intrinsic, irreducibly subjective phenomenal qualities. These qualities are not real features of experience but illusions produced by the introspective process itself. This misrepresentation is adaptive—it simplifies cognitive monitoring and facilitates communication and self-control—but it leaves us deeply confused about the nature of our own minds.
Implicit bias and playing double
How can we explicitly endorse egalitarian principles while implicitly acting on prejudice? This ‘playing double’ reflects the dual-level structure of mind. Implicit attitudes, formed passively through experience and culture, may conflict with the explicit commitments we consciously adopt. Understanding this is crucial for addressing bias and achieving self-control.
Delusions and self-deception
What are delusions, and why do people maintain manifestly irrational beliefs? The two-level view offers an account. Delusions may be non-doxastic acceptances—premising policies adopted for pragmatic or emotional reasons rather than truth. Full-blown self-deception involves accepting a proposition for comfort while non-consciously manipulating evidence to avoid acknowledging what one has done. The self-deceiver pursues a ‘shielding strategy’, adjusting their deliberative standards so the acceptance appears like genuine belief. This explains how we can believe contradictory things without basic-level irrationality.
The performative account of first-person authority
How do we know our own minds? For explicit beliefs and desires, the answer is not introspection but performativity. In sincerely saying ‘I believe that p’, I don’t report a pre-existing belief but commit myself to a premising policy, thereby making myself a p-believer. The utterance is self-guaranteeing—it simultaneously asserts that I possess the mental state and brings it about that I do. This explains why avowals are authoritative, why we look outward when asked what we believe (we’re considering whether to commit), and why psychological verbs have the same meaning in all uses yet are special in the first person. First-person authority for explicit attitudes is performative self-guarantee, not privileged access. For implicit attitudes, we have no such authority—we must interpret ourselves as we interpret others.
Forms of introspection
What forms could introspection take in different minds? We can map a space of possibilities. Introspective systems vary in directness (whether they access mental states directly or infer them), conceptuality (whether they represent states conceptually), flexibility (whether they can be controlled and modified), and unity (whether introspection involves one system or many). This helps us understand human introspection, assess whether animals and AIs introspect, and identify novel forms.
The Truth Demon test
How can you tell whether you really believe something? Imagine being interrogated by an all-knowing Truth Demon who will punish you horribly for any false answer. What would you say under that threat? If you’d give a different answer than you normally profess, you don’t really believe what you claim—you’ve adopted it for social, emotional, or ideological reasons. This reveals the difference between genuine belief and mere acceptance.
Implications
This work challenges the idea that we are transparent to ourselves and that introspection gives reliable knowledge of our minds. Much of what we think we believe may be post-hoc rationalization, our reasons for action opaque to us, and implicit and explicit mental states systematically divergent. The distinction between performative authority (for explicit attitudes) and interpretive access (for implicit ones) resolves the tension between Cartesian and anti-Cartesian intuitions: we have a kind of privileged access, but only to a limited domain of mental states, and it works differently than traditionally supposed.
If thoughts and decisions are largely non-conscious, as the interpretive view suggests, we need to rethink moral responsibility. There are also practical implications for combating implicit bias, treating delusions, and developing AI systems with genuine self-knowledge. The view that introspection is systematically misleading—selective, distorting, and caricaturing—is central to illusionism about consciousness and explains why phenomenal realism seems so compelling despite being false.
The research programme on possible forms of introspection opens new avenues for investigating self-knowledge across different types of minds and developing more nuanced accounts of human metacognition. By asking what introspection could be rather than only what it is, we better understand its diversity, evolution, and potential.
Key publications
The Landscape of Introspection (2025, edited with François Kammerer) — A collection exploring possible forms of introspection in humans, animals, and AI. Opens with a target article proposing a research programme on introspective systems, followed by fifteen commentaries from philosophers and cognitive scientists.
What forms could introspective systems take? A research programme (2023, with François Kammerer) — Maps the space of possible introspective systems along dimensions including directness, conceptuality, flexibility, and unity. Argues this approach can focus attention on radically non-human forms of introspection and integrate competing theories non-adversarially. Journal of Consciousness Studies 30(9–10): 13–48.
More possibilities for introspection: Reply to commentators (2023, with François Kammerer) — Responds to fifteen contributions on our research programme, addressing criticisms, refining the conceptual framework, and drawing lessons for future research. Discusses introspection in humans, non-human animals, current AI, and imaginary minds. Journal of Consciousness Studies 30(9–10): 235–75.
Mind and Supermind (2004) — Develops the two-level framework comprehensively, applying it to akrasia, self-deception, and first-person authority. Argues that avowals are performatives that make premising commitments, explaining first-person authority as performative self-guarantee. Shows how the framework resolves puzzles about weakness of will and self-deceit without partitioning strategies or abandoning rationality constraints. Cambridge University Press.
Playing double: Implicit bias, dual levels, and self-control (2015) — Develops the dual-level framework to explain implicit bias as involving implicit beliefs that conflict with explicit commitments. Argues that override of implicit bias requires strong metacognitive motivation. In M. Brownstein and J. Saul (eds.), Implicit Bias and Philosophy, Volume I. Oxford University Press.
Popular writing
What do you really believe? Take the Truth Demon test (2018) — Are we genuinely committed to everything we profess or unconsciously motivated by desires for acceptance, reassurance, or ideology? Proposes the Truth Demon thought experiment for testing genuine belief. Aeon.
Whatever you think, you don’t necessarily know your own mind (2016) — Introduces the Interpretive Sensory-Access theory of self-knowledge. We learn about our thoughts by interpreting sensory evidence, not through direct introspection, and may be systematically mistaken about our beliefs. Aeon.
How to read a mind (2017) — Explores how we discover what we think and whether we can make mistakes about our own minds. Discusses experimental evidence that people’s choices can be influenced by factors they’re unaware of, leading them to confabulate explanations and attribute mental states they don’t really have. IAI News.
Are delusions acceptances? (2014) — Explores whether delusions are acceptances—intellectual commitments to argue for, defend, and act upon a proposition. Deluded patients are firmly attached to their delusions at the verbal level, but this often doesn’t show up in nonverbal behaviour. If delusions are acceptances, they are active, motivated, and have limited behavioural influence. Imperfect Cognitions.
Related topics
Illusionism—Introspective illusion and phenomenal consciousness
The two-level mind—The two-level account of belief and acceptance
Dual-process theories—Implicit and explicit mental processes
