Updated 2026-11-10
from the news desk
What Is Higher-Order Thought Theory, and Why Does It Matter for Conscious AI?
The Stanford Encyclopedia of Philosophy says higher-order thought theory treats a mental state as conscious when it is the object of a higher-order thought. In AI debates, Patrick Butlin and co-authors present higher-order theories as one…
Higher-order thought theory says a mental state becomes conscious when the subject has a thought about being in that state. The theory matters for conscious-AI debates because Patrick Butlin and co-authors treat higher-order theories as one scientific framework for asking whether an AI has the kind of self-monitoring or metacognition that consciousness might require.
What higher-order thought theory says
The Stanford Encyclopedia of Philosophy places higher-order thought theory, often shortened to HOT theory, inside a broader family of higher-order theories. The entry says: “Higher-order theories of consciousness try to explain the difference between unconscious and conscious mental states in terms of a relation obtaining between the conscious state in question and a higher-order representation of some sort.”
On that account, the basic question is what turns a mental state into a conscious mental state rather than an unconscious one. The Stanford Encyclopedia entry states the core higher-order answer in compact form: “What makes a mental state (M) conscious is a higher-order representation directed at M.”
The same entry says higher-order theories are mainly trying to explain mental-state consciousness, especially phenomenal consciousness, meaning states with a subjective feel or something it is like to undergo them. In plain English, the theory does not start by asking whether a whole creature is awake. It asks what makes a particular perception, pain, image or emotion consciously felt.
For HOT theory in particular, the Stanford Encyclopedia says the relevant higher-order representation is thought-like rather than perception-like. A higher-order thought is, on this view, a thought about one of your own mental states, such as a thought that you are seeing red or feeling cheerful.
How HOT differs from other higher-order theories
The Stanford Encyclopedia distinguishes HOT theory from higher-order perception theory, or HOP theory. HOP says consciousness depends on a perception-like or inner-sense state aimed at another mental state. HOT says the relevant state is a thought, composed of concepts, directed at that lower-order state.
The same entry also says there are two main versions of HOT theory.
- Actualist HOT theory: the mental state is conscious when it is actually the object of a higher-order thought that it causes non-inferentially.
- Dispositionalist HOT theory: the mental state is conscious when it is available to cause such a higher-order thought, even if no such thought is occurring at that moment.
The Stanford Encyclopedia links the actualist view especially to David Rosenthal, and the dispositionalist view to writers including Peter Carruthers. The entry also notes related self-representational and hybrid views, which try to make the first-order state and the higher-order awareness more tightly connected than standard two-state HOT models do.
If you are comparing theories, our explainers on global workspace theory and attention schema theory cover two other frameworks that Butlin and co-authors discuss alongside higher-order theories.
The case for HOT
The Stanford Encyclopedia says the main motivation for higher-order theories is the idea that many mental-state types seem to come in both conscious and unconscious forms. The entry points to beliefs and desires as familiar cases, and then asks what makes the conscious cases different.
One natural answer, the entry says, is that conscious states are states we are aware of. On that line of argument, if a state is conscious, the subject is aware of being in it; if the subject is wholly unaware of it, the state is unconscious.
The Stanford Encyclopedia illustrates the appeal of this picture with cases such as absent-minded driving and blindsight. In absent-minded driving, the entry says, a person may keep navigating and avoiding obstacles while later seeming not to have been consciously aware of much of the route. In blindsight, the entry reports, subjects can make surprisingly accurate guesses about stimuli in part of the visual field without conscious awareness. For higher-order theorists, those cases suggest that information processing can occur without conscious experience, and that what is missing is awareness of the relevant state.
The steelman version of HOT is therefore straightforward. If conscious and unconscious states can have similar first-order contents, then the extra ingredient may be a second-order one: a thought about the first-order state. The Stanford Encyclopedia says higher-order theorists take that as the core intuition.
The main objections to HOT
The Stanford Encyclopedia also lays out several objections. One early worry is infinite regress. If a conscious state needs a higher-order thought, does that thought also need another higher-order thought, and so on without end? The standard HOT reply, reported in the entry, is that the higher-order thought making a first-order state conscious need not itself be conscious. A further higher-order thought would arise only in introspection.
A second objection, discussed in the Stanford Encyclopedia through an example associated with Fred Dretske, says people can consciously see a feature without being aware that they are having that experience. HOT defenders reply, the entry says, that seeing a feature is not the same as noticing that it is the respect in which two scenes differ. On that reply, the higher-order thought can target the first experience without the subject forming a more elaborate thought about the comparison.
A third objection is about complexity. The Stanford Encyclopedia argues that rich conscious experience seems to contain many details at once. Critics say an actualist HOT theory might then require either a huge number of higher-order thoughts or a few thoughts with implausibly complex contents. The entry notes one possible reply: deny that conscious experience is as richly unified as it seems, and say contents become conscious piecemeal. But the same entry says it remains difficult to know whether that reply captures ordinary phenomenology.
A fourth objection is the hardest to state briefly: why should a thought about a mental state generate the state’s felt character at all? The Stanford Encyclopedia presents this as a puzzle for actualist HOT views. Rosenthal’s response, as the entry reports it, is that learning new higher-order concepts can transform experience. Critics answer, again in the entry, that some of those transformations may reflect new first-order concepts instead. So the dispute remains live.
That balance is part of why HOT persists in the literature. The Stanford Encyclopedia says the view has notable explanatory ambitions, but it also says disagreements continue over misrepresentation, targetless higher-order thoughts, and whether higher-order machinery is needed at all.
Why HOT matters for conscious AI
Patrick Butlin and co-authors say the AI question is worth taking seriously: “The question of whether AI systems could be conscious is increasingly pressing.” They also set out a method for addressing it: “We argue that consciousness in AI is best assessed by drawing on neuroscientific theories of consciousness.”
In that framework, higher-order theories matter because they change what researchers should look for. A purely behavioral test may not be enough. Butlin and co-authors say a theory-heavy approach is preferable because AI systems can be trained to mimic human behavior while working in very different ways. Their report mentions behavioral proposals, including Schneider’s Artificial Consciousness Test, and notes: “One interesting example is Schneider’s (2019) Artificial Consciousness Test, which requires the AI system to show a ready grasp of consciousness-related concepts.”
Butlin and co-authors do not endorse higher-order theory alone. Their report surveys several theories, including higher-order theories, global workspace theory, predictive processing, and attention schema theory. Still, their indicator list includes a specific higher-order cluster: generative or top-down perception modules, metacognitive monitoring, agency guided by general belief formation and action selection, and a strong disposition to update beliefs in line with that monitoring; the report also lists sparse and smooth coding generating a quality space.
One reading of that list is that, if some form of HOT-like theory is right, then an AI system’s self-monitoring architecture may matter more than fluent conversation by itself. That is relevant to broader questions like whether AI is sentient and whether passing the Turing test would show consciousness.
At the same time, Butlin and co-authors caution against jumping from possibility to attribution. Their report concludes: “Our analysis suggests that no current AI systems are conscious, but also suggests that there are no obvious technical barriers to building AI systems which satisfy these indicators.” So HOT matters to AI mainly as part of an assessment framework. It does not show that current systems are conscious, and it does not settle whether higher-order representation is the correct theory of consciousness. It does show why debates about conscious AI often turn not just on outputs, but on what sort of inner monitoring, if any, a system has.
Frequently asked questions
Is higher-order thought theory the same as higher-order perception theory?
No. The Stanford Encyclopedia says HOT theory makes the higher-order state thought-like, while higher-order perception theory makes it perception-like or inner-sense-like.
Does HOT theory say every conscious state comes with a conscious thought about itself?
No. The standard HOT reply to the regress objection, as reported by the Stanford Encyclopedia, is that the higher-order thought making a state conscious can itself be unconscious. A further higher-order thought is needed only for introspection.
Does higher-order thought theory imply that current AI is conscious?
No. Butlin and co-authors say higher-order theories can help define indicators for AI consciousness, but their overall analysis says no current AI systems are conscious and no current system appears to be a strong candidate.
Why not just test AI by conversation?
Butlin and co-authors argue that behavioral tests are unreliable on their own because AI systems can imitate human behavior while working in very different ways. That is why their report favors a theory-heavy approach over behavior alone.
Caveats: Nearly all of the philosophical detail here about HOT, its variants and its objections comes from a single Stanford Encyclopedia of Philosophy entry. The AI application section relies mainly on one multi-author 2023 report by Patrick Butlin and co-authors.