QuestionsVII
Could a machine be conscious?
Machines can now talk, reason and seem to care. Whether anyone is home inside them is a different question, and no one knows how to answer it.
- Reading time
- 5 min
- Difficulty
- Intermediate
- Branch
- Mind
- First posed
- Alan Turing, 'Computing Machinery and Intelligence', 1950
The question
In 1950, the mathematician Alan Turing proposed replacing the question 'Can machines think?' with a game. A judge converses by text with a hidden human and a hidden machine. If the judge cannot reliably tell which is which, Turing suggested, we have as much reason to credit the machine with thought as we have to credit other people.
Today, programs routinely hold fluent conversations. They write poems about loneliness and say they are curious. The Turing test has, in practice, become much easier to pass than Turing imagined. And yet many people feel that something is missing: that a system can say 'I feel sad' without anything being felt.
The question splits in two. Could a machine think, in the sense of reasoning and understanding? And could a machine be conscious, in the sense of having experiences, so that there is something it is like to be it? The second is the harder one, because, as the hard problem shows, we have no agreed test for consciousness even in each other.
Why it matters
If machines can be conscious, they might be able to suffer, and then how we build, use and switch them off becomes a moral question on an enormous scale. Getting it wrong in either direction is costly: treating feeling beings as tools, or extending moral concern to systems that feel nothing while neglecting beings that do.
The question also tests every theory of mind. A theory that says consciousness depends on biology, one that says it depends on information processing, and one that says it depends on something else entirely will give different verdicts about the same machine.
Key terms
- Turing testAlan Turing's proposed test for machine intelligence: if a judge holding text conversations cannot reliably tell a machine from a human, the machine passes.
- FunctionalismThe view that mental states are defined by what they do, their causal roles linking inputs, other mental states and behaviour, rather than by what they are made of.
- Substrate independenceThe assumption that a mind depends on the pattern of information processing, not on the material it runs on, so a mind could exist in silicon as well as in neurons.
- Phenomenal consciousnessThe felt, subjective aspect of experience: what it is like to see red, feel pain or taste coffee.
- Biological naturalismJohn Searle's view that consciousness is a real, subjective biological phenomenon caused by brain processes, as digestion is caused by the stomach, and that running a program cannot by itself produce it.
The positions
Yes, in principle
Consciousness depends on the organisation of a system, not on what it is made of. A machine with the right organisation would be conscious.
No: computation is not enough
Running a program, however sophisticated, does not produce understanding or experience. Minds need the causal powers of biology, or something like them.
We cannot know
There may be a fact of the matter, but we have no reliable way to detect machine consciousness, and may never have one.
Arguments and objections
Yes, in principle
- Best argument
- Imagine replacing your neurons one by one with silicon chips that perform exactly the same function. Your behaviour would not change at any step. Chalmers argues that your experience would not change either: it is implausible that it would fade gradually while you insisted nothing was different, or vanish at some arbitrary neuron. If so, a silicon brain with your organisation is conscious, and substrate independence is true.
- Strongest objection
- The thought experiment assumes that function is all that matters, which is the point in dispute. If consciousness depends on specific biological processes, like particular electrochemical dynamics, a chip that only reproduces input and output may leave it behind.
- The reply
- Defenders answer that the replacement chips can be made to reproduce any level of detail you like, including the electrochemistry, and at some point insisting on biology looks like prejudice in favour of carbon.
No: computation is not enough
- Best argument
- John Searle's the Chinese room argument: a person following rules to manipulate Chinese symbols can produce perfect answers without understanding a word. Programs manipulate symbols by their shape alone; meaning and experience are not in the syntax. Searle calls his view biological naturalism: consciousness is caused by brains as digestion is caused by stomachs, and a simulation of digestion digests nothing.
- Strongest objection
- Critics reply that the person in the room is only one part of a larger system, and the system as a whole might understand. Searle never explains which biological powers matter or why silicon could not have them.
- The reply
- Searle answers that he can memorise all the rules and do the work in his head, so there is no larger system left over, and still he understands no Chinese.
We cannot know
- Best argument
- Language models are trained on human writing, so they will produce human-sounding reports of inner life whether or not they have any. Behaviour, which is our usual evidence of other minds, has been decoupled from the processes that produce it in us. Until we have a validated theory of consciousness, any verdict is guesswork.
- Strongest objection
- Total agnosticism is not a usable policy. We must decide how to treat these systems now, so we need some way to weigh the evidence, imperfect as it is.
- The reply
- Agnostics agree, and propose precaution: take seriously indicators drawn from the leading scientific theories, and avoid creating systems whose moral status we cannot assess.
Who argued what
- Yes, in principle David Chalmers, Daniel Dennett, Alan Turing, Hilary Putnam
- No: computation is not enough John Searle, Roger Penrose, Anil Seth
- We cannot know Jonathan Birch, Eric Schwitzgebel
Where you stand
Will future AI systems be conscious?
In surveys
Choose one to see how people answer.
Variations
The neuron-by-neuron replacement
Your neurons are swapped, one per day, for perfect artificial copies. After a hundred billion days, you are entirely artificial. At which day, if any, did you stop being conscious? If there is no sensible answer, perhaps you never stopped.
The China brain
Ned Block imagined the whole population of China simulating a brain by passing messages via radio, each person playing one neuron. Would the nation, as a whole, feel anything? If not, functionalism seems to be in trouble; if so, consciousness is far stranger than we thought.
Go deeper
Indicators instead of a test
In 2023 a group of philosophers and scientists led by Patrick Butlin and Robert Long proposed assessing AI systems against 'indicator properties' derived from scientific theories of consciousness, such as a global workspace, recurrent processing, or a model of the system's own attention. The approach does not settle the question but replaces a single test with a checklist that can be updated as theories improve.
Thinking versus feeling
It is possible to answer the two halves differently. Many people now accept that machines can reason in a meaningful sense while remaining doubtful that they experience anything. Others argue that genuine understanding requires experience, so the two cannot come apart. The distinction maps onto the difference between the easy and hard problems of consciousness.
Related ideas
Further reading and sources
- EncyclopediaStanford Encyclopedia of Philosophy: The Chinese Room Argument
- WikipediaArtificial consciousness
- WikipediaTuring test
- SurveyPhilPapers Survey, 2020 · question: Other minds: future AI systems; about 1,785 professional philosophers; accept or lean toward
- PaperComputing Machinery and Intelligence, Alan Turing
- PaperMinds, Brains, and Programs, John Searle
- BookReality+, David Chalmers
- BookThe Edge of Sentience, Jonathan Birch
Open questions
- What evidence would convince you that a machine was conscious?
- If we cannot tell, should we treat advanced machines as if they might feel?
- Is your confidence that other humans are conscious based on anything a machine could not also show?