Consciousness, oriented

Most arguments about consciousness go wrong in the first sentence, because the two sides mean different things by the word. Four distinctions clear up most of it.

Creature or state. A person can be conscious while one of their perceptions is not. Say you drive home on autopilot and cannot recall the last mile. You were plainly conscious, yet some of your seeing may not have been.

Awake or aware. Wakefulness is arousal: eyes open, sleep-wake cycles running. Awareness is experience of yourself or your surroundings. A dreamer has awareness with little wakefulness, and some brain-injured patients show wakefulness with no sign of awareness.

Level or contents. Sinking under anesthesia changes how conscious you are. Shifting from the music to the smell of toast changes what you are conscious of. Studies of one need not tell you about the other.

Phenomenal or access. Ned Block's pair is the one you will meet most. Phenomenal consciousness is what an experience is like. Access consciousness is information you can reason with, report and act on. Thomas Nagel's phrase "something it is like" points at the first kind.

The usual slip is to treat a finding about one sense as a finding about all of them. A study of what people can report is a study of access, and it may or may not reach felt experience.

Rule: before you judge any claim about consciousness, ask which of these senses it uses.

David Chalmers split the field in two. Easy problems ask how the brain does a job: tell colors apart, direct attention, produce a report. They may take decades, but a mechanism could answer them. The hard problem asks why doing those jobs comes with experience at all. Joseph Levine calls the same worry an explanatory gap: "water is hydrogen and oxygen" explains water's behavior, while "pain is this brain process" seems not to explain why it hurts.

The felt qualities at issue are called qualia, and three thought experiments press the point. Frank Jackson's Mary knows every physical fact about color but has lived in a black-and-white room; on leaving, she seems to learn something. A philosophical zombie is your exact physical twin with no experience. An inverted spectrum imagines someone whose red and green are swapped with no change in behavior.

Each argument has a weak joint, and that is where the debate lives. Against Mary, one reply says she gains abilities rather than facts. Against zombies, many physicalists accept that you can imagine them and deny that imagining shows they are possible. Daniel Dennett goes further, treating qualia as something like an illusion. Colin McGinn holds that an answer exists but may be beyond human concepts.

A common misreading treats these as proofs. They are tools for finding out what you are committed to, and Jackson himself later gave up his conclusion.

Rule: find the step an argument cannot do without, then ask who denies it.

You can sort most views by what each says exists and what each then has to explain.

Physicalism says that fixing the physical facts fixes the mental ones. Identity theory makes each kind of mental state a kind of brain state, which struggles if an octopus can feel pain with a very different brain. Functionalism answers by defining a state through its causal role, so one state could run on many kinds of hardware. Its cost is the worry that role leaves out feeling: Ned Block imagines a whole nation copying a brain's organization and asks whether the nation would feel anything.

Dualism keeps something non-physical. Descartes made mind a separate substance; property dualists keep one substance with two kinds of property. The cost is causation. If physics already supplies a sufficient cause for every bodily movement, a non-physical mind has nothing to do, so some dualists accept epiphenomenalism: experience is caused but causes nothing.

Panpsychism puts simple experience in basic matter, so experience never has to emerge from nothing. Its cost is the combination problem: why would many tiny experiencers make one mind? Russellian monism is a close relative, holding that physics describes what matter does and not what it intrinsically is.

Illusionism says experience is not as introspection presents it, because the brain's model of itself simplifies. The cost is plausibility, though Michael Graziano argues that scholars tend to picture only its most extreme form.

Rule: when a view seems obviously wrong, look for the cost its rivals pay instead.

A neural correlate of consciousness is the smallest brain activity sufficient for a specific experience. Note the word smallest. Your whole brain is trivially enough for any experience you have, so naming it explains nothing, and background conditions such as arousal do not tell seeing a face apart from seeing a house.

The basic tool is contrast. Hold the input fixed and let the experience vary. In binocular rivalry, each eye gets a different picture and perception flips between them while the screen stays the same. In masking, a flashed word followed at once by a pattern often goes unseen. Activity that tracks the flips, or the seen trials, becomes a candidate.

Here is the problem. Suppose a lab finds strong frontal activity whenever volunteers press a button to say they saw an image. That could mark seeing, or it could mark deciding and pressing. No-report designs answer this by reading perception from involuntary signals, such as eye movements, with no button at all. In some of these studies frontal activity shrinks, which feeds a live dispute over whether the correlates lie mainly at the front of the cortex or the back. No-report designs have their own weakness, since researchers must then infer what was seen.

The other trap is the word correlate itself. Activity that always accompanies an experience might cause it, might follow from it, or might share a cause with it. Stimulation and lesion studies help, and even a perfect correlate would leave the hard problem where it was.

Rule: ask what a contrast held fixed, and what else differs between its two conditions.

Both theories treat most brain processing as unconscious and ask what lifts a content out of it. Their answers differ.

Global workspace theory, from Bernard Baars, pictures many specialized processors running in parallel and competing for a limited shared stage. The winner gets broadcast, so memory, planning and speech can all use it. Stanislas Dehaene and colleagues tied this to long-range neurons in prefrontal and parietal cortex, and to ignition, a sudden burst of widespread activity. That gives their version something to lose: a clearly conscious content with no prefrontal trace would count against it. The wider workspace idea is not committed to any one region.

Higher-order theories start from a different case. Absent-minded driving is the classic case: you see the road with no awareness of seeing it. These theories say what is missing is a further state directed at the perception, either a thought about it or something like inner perception of it. Blindsight fits the picture: visual processing guides behavior, with no awareness of it.

Each has a standard objection. Critics say the workspace explains access and leaves out felt experience. Against higher-order views, they ask whether infants and animals have thoughts about their own states, and what happens if a higher-order state misrepresents a perception that is not there.

Do not assume that a shared prediction makes them the same theory. Both often point to prefrontal cortex, for different reasons, so a frontal result alone will not separate them.

Rule: to tell two theories apart, find what one says makes a state conscious and the other does not.

Integrated information theory (IIT), from Giulio Tononi, runs the usual argument backward. Rather than starting with brain data, it starts from claims about experience: it exists for itself, it is structured, specific, unified and definite. It then asks what a physical system must be like to match those features.

The answer is integration. Phi measures how much a system does as a whole beyond what its parts do separately, and the theory identifies consciousness with this quantity. Only a system's physical causal structure counts, so behavior is beside the point.

Take a concrete case. Suppose an image classifier passes signals one way through layers, with no loops, and performs as well as you do. On IIT its integration is zero, so it is not conscious, however clever it looks. A simple but tightly looped circuit, meanwhile, would have some consciousness. Supporters accept both verdicts. Critics find them hard to believe, and they add that no behavioral test could detect the difference the theory claims.

Two further points cause trouble. Exact phi cannot be computed for anything near the size of a brain, because the calculation grows explosively, so experiments rely on proxies. And critics dispute whether the starting claims about experience are self-evident, and whether the step from them to physical structure holds.

A common slip is to read IIT as a theory of intelligence. It is close to the opposite: it separates what a system can do from whether it is conscious.

Rule: on IIT, look at the causal structure and ignore the performance.

Researchers tend to find support for their own theories, partly because each side designs its own experiments. An adversarial collaboration tackles this directly. Rival theorists agree beforehand on the experiments and on what result would count against each view. Preregistration then fixes the predictions and analyses before data arrive, so nobody can pick the flattering analysis afterward.

The COGITATE consortium applied this to global neuronal workspace theory and integrated information theory. Participants viewed faces, objects, letters and symbols for varying durations while several labs recorded brain activity with different methods.

The two theories made different bets. Integrated information theory expected sustained synchronization within posterior cortex for as long as an image was seen. Workspace theory expected ignition at the image's onset and again at its offset, with prefrontal cortex carrying what was seen.

Read the results carefully, because they are easy to misreport. Some findings fit each theory. Yet the sustained posterior synchrony did not appear as predicted, which counts against the first. Offset bursts mostly failed to appear and frontal coding covered only some features, which counts against the second. So a headline saying one theory won is wrong, and any rival sharing those predictions inherits the trouble.

A failed prediction also does not end a theory. Proponents can dispute what a result means, and both sides here recorded their own readings. What the method gives you is a clear record of what was risked and what happened.

Rule: score a theory test prediction by prediction, and charge each failure to the theory that made it.

Everyday language runs these together: "I wasn't paying attention, so I didn't see it." Experiments pull them apart. In inattentional blindness, people counting basketball passes miss a person in a gorilla suit. In change blindness, a large change hidden by a brief flicker goes unnoticed. Both suggest that seeing something consciously takes more than having it in view.

The gatekeeper view says attention is necessary for consciousness. Think carefully about what would refute it. A stimulus that seems to draw attention but is never seen suggests at most that attention is not sufficient. To break "no consciousness without attention," you need awareness of something unattended, and any report of it seems to involve attending to it. That is why the debate is hard to settle.

Brain damage gives sharper cases. In hemispatial neglect, a common result of stroke, a patient may ignore one side of space although their eyes work. In blindsight, damage to primary visual cortex leaves a patient denying sight in part of the field, yet pointing to objects there far above chance when made to guess.

Measurement is the catch. Suppose a participant rarely says "I saw it," but her yes answers are nearly always right. She may be blind to the stimulus, or merely cautious. Signal detection theory separates sensitivity from willingness to say yes, and confidence ratings show whether she knows how well she is doing. A claim of unconscious perception depends on which of these tests was used.

Rule: before accepting "perception without awareness," ask how awareness was measured.

Your sense of being one self with one body feels like bedrock. Two lines of research show it is put together.

Split-brain patients had the corpus callosum, the main link between the hemispheres, cut to control severe epilepsy. Researchers then flashed a word to one visual field, which reaches only the opposite hemisphere. Flash "key" on the left and the patient, whose speech is usually on the left side, may say he saw nothing, yet find a key by touch with his left hand. Roger Sperry and Michael Gazzaniga's findings were widely read as showing two minds in one skull.

That reading now looks too quick. Some processes, including control of action, seem to remain unified, and a recent review judges the evidence insufficient to say whether consciousness is split. Be wary of the confident popular version.

The bodily self is similarly constructed. In the rubber hand illusion, a fake hand in view is stroked in time with your hidden hand, and you start to feel the fake one as yours. Stroke them out of step and the effect weakens. Knowing the hand is rubber does not stop it, which tells you the feeling is not a belief.

Two feelings are easy to merge here. Ownership is the sense that a limb is yours; agency is the sense that you are moving it. If a doctor lifts your relaxed arm, you have the first without the second.

Rule: when something feels simple and given, ask what signals have to line up to produce it.

A dreamer lies still and unresponsive while having vivid experience. That one fact undoes an easy assumption: that no response means no experience.

Research on anesthesia draws the distinction formally. A patient may be unconscious, with no experience at all. They may have disconnected consciousness, with experience but no perception of the room, as in dreaming. Or they may have connected consciousness, aware of their surroundings. All three can look the same from the outside. Which state appears depends on the drug and the dose, and ketamine is known for disconnected states with vivid content.

Memory complicates things. Say a patient recalls nothing of an operation. They may have had no experience, or they may have had experience that was never stored. Dream research has a similar problem. Reports come after waking, filtered by memory, and philosophers have even asked whether dreams are experienced during sleep or put together on waking.

Lucid dreaming offers a rare way around this. Lucid dreamers can make prearranged eye movements, which show up on sleep recordings as the dream unfolds. Dreams are also not confined to REM (rapid eye movement) sleep, though recall is highest after waking from REM.

These states are useful because they vary experience in one brain, within one night or one procedure. The drugs and stages differ, though, so a finding about one anesthetic does not automatically cover them all.

Rule: treat unresponsiveness as a fact about behavior, and no memory as a fact about memory.

Three states are often confused. In a coma, the eyes stay closed and there are no sleep-wake cycles. In the unresponsive wakefulness syndrome (UWS), once called the vegetative state, the eyes open and sleep-wake cycles return, but no purposeful response appears. In the minimally conscious state (MCS), signs of awareness come and go but can be repeated, as when a patient follows a simple request on some days or tracks a moving mirror with the eyes.

Clinicians separate these with set bedside tools, notably the Coma Recovery Scale–Revised (CRS-R). Repeat testing matters for two reasons. A single movement could be a reflex or chance, and a real response may not show up on any given day.

The deeper limit is that a behavioral test can only find what the body can express. A patient may be aware and unable to show it. That is the motive for the perturbational complexity index (PCI). Researchers deliver a magnetic pulse to the cortex using transcranial magnetic stimulation (TMS) and record the response with an electroencephalogram (EEG). A complex, widespread, varied response suggests capacity for consciousness, and a simple or local one suggests its absence.

Read the results at the strength they are reported. In one study, 4 of 11 UWS patients gave high-complexity responses, which the authors say might suggest a covert capacity. That is not a finding that those patients were conscious at the time, and it is not a basis for any decision about a particular person's care.

Rule: keep "shows no sign of awareness" apart from "has no awareness."

An altered state is a temporary, marked change in the overall pattern of experience. Psychedelics interest researchers because they change perception, thought and the sense of self in a way that can be studied under controlled conditions. Classic psychedelics such as psilocybin act mainly on one kind of serotonin receptor, the 2A receptor. That tells you where the effect begins, not how it produces a change in experience.

Two terms come up constantly. Ego dissolution is the reported loss of a bounded sense of self. Mystical-type experience, measured by questionnaire, includes a sense of unity and of something beyond words. Both describe what a state was like. A person who comes away convinced that everything is one has had a real experience of that conviction, and whether it is true is a separate question.

Several models try to connect brain and experience. The entropic brain hypothesis links more disordered brain signals to more varied experience. Another model says these drugs relax the high-level expectations through which the brain interprets input. Both are proposals still being tested.

The common overreach is the headline that a drug study will finally explain consciousness. Researchers in this area point out that the word has several senses. Findings may tell you how contents or the sense of self change, without touching why there is experience at all. This chapter describes research and is not guidance on using any substance.

Rule: ask which sense of "consciousness" a finding addresses.

Libet asked people to flex a wrist whenever they liked and to note, using a fast clock, when they first felt the urge. Scalp recordings showed a slow build-up, the readiness potential, starting several hundred milliseconds before that reported moment. Later work, including brain imaging and recordings from single neurons, found related signals that predict a choice before people report deciding.

It is tempting to jump from there to "free will is an illusion." Separate the finding from the conclusion, because several links in that chain can be questioned.

What does the signal mark? The conclusion needs it to be a decision. It may instead be general preparation, or background activity that drifts up and down and sometimes tips into a movement. Because the signal is an average lined up on the movement, a smooth ramp in the average need not exist on each trial.

How good is the timing? Judging a clock position while deciding to move is a hard dual task, and reports can be biased. And how far does it reach? Flexing a wrist for no reason is unlike choosing a job, so the result may say little about decisions you care about. Libet himself suggested that consciousness could still veto a prepared act.

Finally, the argument assumes that a free act must begin in awareness. Many accounts of free will do not require that, so even a solid timing result would not settle the matter alone.

Rule: for any finding said to disprove free will, list the assumptions that connect the measurement to the claim.

The same problem runs through both cases: experience is not directly observable, so what counts as evidence?

For animals, the starting point is analogy. Close relatives share your brain structures and behavior, so the inference is strong, and it weakens with distance. Researchers separate the distribution question, which animals are conscious at all, from the phenomenological one, what their experience is like. Nagel's bat shows why the second is hard. No single test settles the first: failing the mirror test speaks to one capacity, recognizing a reflection, and not to experience in general. Fish pain and octopus cognition are the live cases, and birds, with flexible behavior and no layered neocortex, press against any claim that a human-style cortex is required.

Machines reverse the problem. Their behavior can be excellent while their biology is absent. Say a chatbot writes that it feels lonely. It was trained on an enormous amount of human writing, so that sentence is weak evidence of anything either way. The Turing test measures conversation, not experience, and John Searle's Chinese room argues that following rules is not understanding.

One response is to look inside, checking a system's architecture for features that theories link to consciousness. But the theories disagree. On a functionalist view, the right organization might be enough. Integrated information theory rules out strictly feedforward designs whatever they do, and embodied views require a living body. Nobody has a settled test, so treat any flat claim about a particular system with suspicion.

Rule: for any non-human case, ask what the evidence would look like if there were no experience behind it.