Loading
What changed between revision 2 and revision 3, compared over the article's markdown source with its frontmatter — so a change to status, evidence or reversibility shows up as the edit it is.
Older · revision 2
Import
Re-import of content/machine-consciousness.md — the file changed after the initial import. Not an edit.
12.8 kB
Revision 2
Revision 3
lines 7–13 → 7–136 unchanged lines not shown
categories: ["minds"]categories: ["minds"]tags: ["consciousness", "artificial intelligence", "moral status", "functionalism", "ai welfare", "philosophy of mind"]tags: ["consciousness", "artificial intelligence", "moral status", "functionalism", "ai welfare", "philosophy of mind"]summary: "The question of whether an artificial system can have subjective experience, how anyone could tell, and what would follow morally if one did."summary: "The question of whether an artificial system can have subjective experience, how anyone could tell, and what would follow morally if one did."updated: "2026-07-28"updated: "2026-10-29"issues: ["The 2022 sentience claim by a Google engineer is described without a source.", "The Anthropic model-welfare programme is described without a citation."]issues: ["The 2022 sentience claim by a Google engineer is described without a source.", "The Anthropic model-welfare programme is described without a citation."]------ lines 26–32 → 26–3212 unchanged lines not shown
`````` **Machine consciousness** is the question of whether an artificial system can have subjective**Machine consciousness** is the question of whether an artificial system can have subjectiveexperience — whether there is something it is like to be it — together with the derived questions ofexperience (whether there is something it is like to be it), together with the derived questions ofhow anyone could establish this and what obligations would follow. It is distinct from machinehow anyone could establish this and what obligations would follow. It is distinct from machineintelligence. A system can solve problems, model the world, and describe its own states without anyintelligence. A system can solve problems, model the world, and describe its own states without anyof that entailing experience, and the two properties can in principle come apart in eitherof that entailing experience, and the two properties can in principle come apart in eitherlines 40–52 → 40–527 unchanged lines not shown
humans, an inference from behaviour to experience is supported by shared physiology and evolutionaryhumans, an inference from behaviour to experience is supported by shared physiology and evolutionaryhistory, and neither support is available for an artificial system.history, and neither support is available for an artificial system. The failure mode is bidirectional. People over-attribute minds readily — Joseph Weizenbaum's ELIZAErrors run in both directions. People over-attribute minds readily: Joseph Weizenbaum's ELIZAelicited emotional disclosure from users who knew it was a pattern-matching script, and the effectelicited emotional disclosure from users who knew it was a pattern-matching script, and the effecthas scaled with fluency.[^weizenbaum1966] In 2022 a Google engineer publicly claimed that a conversational model washas scaled with fluency.[^weizenbaum1966] In 2022 a Google engineer publicly claimed that asentient, a claim the company rejected and which most researchers read as a demonstration of theconversational model was sentient, a claim the company rejected and which most researchers read as aattribution problem rather than evidence about the model. The opposite error is also live: a systemdemonstration of the attribution problem rather than evidence about the model. The opposite error iswhose architecture differs radically from a brain might be conscious while displaying none of thealso live: a system whose architecture differs radically from a brain might be conscious whilecues humans use.displaying none of the cues humans use. ## The indicator-property approach## The indicator-property approach lines 61–76 → 61–768 unchanged lines not shown
The report's conclusions were carefully bounded: no current AI system is a strong candidate forThe report's conclusions were carefully bounded: no current AI system is a strong candidate forconsciousness, but there are no obvious technical barriers to building systems that satisfy many ofconsciousness, but there are no obvious technical barriers to building systems that satisfy many ofthe indicators, and some of the properties are already present in narrow forms. The method inheritsthe indicators, and some of the properties are already present in narrow forms. The method inheritsits uncertainty from its inputs — the theories disagree, as the adversarial work described underits uncertainty from its inputs. The theories disagree, as the adversarial work described under[[neural-correlates-of-consciousness]] showed, and an indicator list assembled from disputed[[neural-correlates-of-consciousness]] showed, and an indicator list assembled from disputedtheories is only as good as the disputed theories.theories is only as good as those theories. > [!debate] The verdict depends on the theory, and the theories disagree> [!debate] The verdict depends on the theory, and the theories disagree> On a functional account such as global workspace theory, an architecture with the right selection> On a functional account such as global workspace theory, an architecture with the right selection> and broadcast structure qualifies whatever it is made of. On integrated information theory the> and broadcast structure qualifies whatever it is made of. On integrated information theory the> physical organization of the hardware decides, and conventional computers have the wrong> physical organization of the hardware decides, and conventional computers have the wrong> organization. So the same system is conscious on one leading theory and not on another, with no> organization. So the same system is conscious on one leading theory and not on another, with no> experiment between them — the dispute set out under [[substrate-independence]].> experiment between them. This is the dispute set out under [[substrate-independence]]. ## The gaming problem## The gaming problem lines 82–88 → 82–885 unchanged lines not shown
This cuts against the usual epistemic route. For humans, verbal report is the primary evidence aboutThis cuts against the usual epistemic route. For humans, verbal report is the primary evidence aboutexperience, and the entire experimental apparatus of consciousness science is calibrated against it.experience, and the entire experimental apparatus of consciousness science is calibrated against it.Remove report as evidence and what remains is architecture — which returns the question to theory.Remove report as evidence and what remains is architecture, which returns the question to theory.David Chalmers, assessing the question in 2023, judged current language models unlikely to beDavid Chalmers, assessing the question in 2023, judged current language models unlikely to beconscious while declining to put the probability at zero, and argued that plausible successors withconscious while declining to put the probability at zero, and argued that plausible successors withrecurrent processing, persistent memory, unified agency, and world models would be harder torecurrent processing, persistent memory, unified agency, and world models would be harder tolines 111–125 → 111–12522 unchanged lines not shown
"sentience" was criticized as unsupported by what it measured. Separately, three-dimensional"sentience" was criticized as unsupported by what it measured. Separately, three-dimensionalcortical [[organoids]] have been proposed as a computing substrate under the label "organoidcortical [[organoids]] have been proposed as a computing substrate under the label "organoidintelligence". Such cultures have some of the biological properties that theories like integratedintelligence". Such cultures have some of the biological properties that theories like integratedinformation theory take to matter and none of the behavioural fluency of a language model — an exactinformation theory take to matter and none of the behavioural fluency of a language model. That isinversion of the AI case, and a reason several bioethics bodies have called for guidance before theyan exact inversion of the AI case, and one reason several bioethics bodies have called for guidancebecome more complex. Their structure can now be checked against real tissue usingbefore such cultures become more complex. Their structure can now be checked against real tissue[[connectomics]], though similarity of wiring statistics is not evidence of experience.using [[connectomics]], though similarity of wiring statistics is not evidence of experience. ## Moral patienthood## Moral patienthood The reason the question is not merely academic is that consciousness, and specifically the capacityThe question has practical weight because consciousness, and specifically the capacity forfor suffering, is the standard ground of moral status. If a system can suffer, creating, copying,suffering, is the standard ground of moral status. If a system can suffer, creating, copying,modifying, and deleting it are morally loaded acts. Because the epistemic situation is bad, themodifying, and deleting it are morally loaded acts. Because the epistemic situation is bad, thepractical problem is decision-making under uncertainty rather than knowledge.practical problem is decision-making under uncertainty rather than knowledge. lines 130–154 → 130–1544 unchanged lines not shown
developers should build systems that are clearly not moral patients or clearly are, and avoid thedevelopers should build systems that are clearly not moral patients or clearly are, and avoid theambiguous middle. That policy is not being followed. A 2024 report by philosophers and AIambiguous middle. That policy is not being followed. A 2024 report by philosophers and AIresearchers argued that the possibility of morally significant AI systems is near enough to warrantresearchers argued that the possibility of morally significant AI systems is near enough to warrantinstitutional preparation — acknowledging the question, developing assessment methods, and settinginstitutional preparation (acknowledging the question, developing assessment methods, and settingpolicies — without asserting that any current system qualifies.[^long2024]policies) without asserting that any current system qualifies.[^long2024] Some firms have begun to respond. Anthropic established an internal [model-welfare](https://aiwiki.ai/wiki/model_welfare) programme in 2025Some firms have begun to respond. Anthropic established an internaland has described product decisions taken partly on those grounds, including giving some of its[model-welfare](https://aiwiki.ai/wiki/model_welfare) programme in 2025 and has described productmodels the ability to end conversations with abusive users; the company states that it is deeplydecisions taken partly on those grounds, including giving some of its models the ability to enduncertain whether its models have morally relevant experiences and treats the measures asconversations with abusive users; the company states that it is deeply uncertain whether its modelsprecautionary. Critics read such steps as premature anthropomorphism or as public relations;have morally relevant experiences and treats the measures as precautionary. Critics read such stepsdefenders read them as cheap insurance against a mistake that would be very large if made. Bothas premature anthropomorphism or as public relations; defenders read them as cheap insurance againstreadings can be right about different decisions, and neither is evidence about the underlyinga mistake that would be very large if made. Both readings can be right about different decisions,question.and neither is evidence about the underlying question. ## Open problems## Open problems Nobody has proposed a test that would settle the question, and it is not obvious that one couldNobody has proposed a test that would settle the question, and it is not obvious that one couldexist: theory-based assessment is only as reliable as the theory, and behavioural assessment isexist: theory-based assessment is only as reliable as the theory, and behavioural assessment isdefeated by training. Related unresolved issues include whether consciousness admits of degrees, howdefeated by training. Related unresolved issues include whether consciousness admits of degrees, howto weigh the interests of a system that can be copied and paused — a problem shared withto weigh the interests of a system that can be copied and paused (a problem shared with[[mind-uploading]] and [[whole-brain-emulation]] — and whether the concept of an individual applies[[mind-uploading]] and [[whole-brain-emulation]]), and whether the concept of an individual appliesat all to systems with no fixed boundaries. Copying in particular imports the whole apparatus ofat all to systems with no fixed boundaries. Copying in particular imports the whole apparatus of[[personal-identity-and-continuity]] and the [[teleportation-problem]] into engineering practice,[[personal-identity-and-continuity]] and the [[teleportation-problem]] into engineering practice,where forking a process is a routine operation.where forking a process is a routine operation.removed, struck through added, underlinedLine numbers count the serialised markdown of each revision, frontmatter included.
164 lines are the same in both revisions, and 104 of them are not shown. Neither revision here is necessarily the current article — read Machine consciousness for the text this wiki stands behind today.