by I agree with Denis Noble about Laura. I would not take relationship advice from a chatbot, and I would not confuse its fluent affection with a private inner life. Noble’s recent goes after that confusion with the glee of a man who has brought a water pistol to a silicon convention. He reminds us that living cells use molecular disorder and that our bodies sense the world. AI runs on hardware too. Then he reaches a much larger verdict. “We’re built of water based computation. AI is based on silicon crystals.” The sentence has the clean sound of a boundary. It also hides every difficult step in the argument.
Noble’s conclusion may be right about Laura. His argument does not earn the broader verdict. He describes differences in matter and randomness. He then adds a checklist of senses. Those descriptions do not add up to a proof. A particular chatbot can lack a body while the larger question of artificial consciousness remains open. Noble gets the force of a demonstration by running several distinctions together and letting the audience supply the missing logic. The laughter helps too.
Silicon Is Not Taxidermy
Silicon atoms do move. They remain near ordered lattice positions, which is the sensible thing for atoms in a crystal to do. Their order does not freeze them. Researchers using Argonne’s Advanced Photon Source have . Phonons are quanta of lattice vibration. The experiment is an elegant way of catching a crystal in the act of jiggling. Noble may mean that silicon atoms lack the free Brownian travel of molecules in water. That narrower claim describes a genuine physical contrast. It cannot support his broader wording about motionless silicon. He should have stopped there.
Computer engineers also fight unwanted physical variation. Bits depend on reliable distinctions between states. Heat and electrical noise make that work harder. Manufacturing defects do too. This is a genuine difference from the cellular story Noble tells. Living systems exploit molecular variation in ways that standard digital architectures usually suppress. The concession still does not deliver his conclusion. A difference in scale could matter to cognition. It becomes a barrier to consciousness only after Noble supplies the relevant threshold and evidence for it. A design that ignores a resource differs from a substrate that cannot contain it. My kettle ignores the scenic possibilities of boiling water. Nobody has proved that steam cannot cook dinner.
Randomness has no pedigree requirement. A random event can enter a computation through an electronic process or a physical sensor. An algorithm can also supply pseudorandom variation for search. Turing’s 1950 paper discusses a digital computer with a random element and treats randomness as useful in learning and search. His exact phrase is . Noble turns that advice into a story about artificial deficiency. Turing presents it as a design resource. Added salt is still salt. Added randomness still perturbs a system. The origin of the variation cannot tell us whether a subject is having an experience.
Stochasticity is also cheap in the wrong sense. A thunderstorm has plenty of molecular commotion. A shaken snow globe offers a smaller version with better seasonal branding. I have no reason to think either one is hosting an inner life. Noble’s biological account concerns the use of variation, so he knows that commotion alone is insufficient. The harness therefore carries the explanatory load. We need to know which organization turns variation into experience and why no artificial organization could perform the same role. The physics of water leaves that role unexplained.
The Wetness Is Not the Explanation
Water makes a poor moat around consciousness. Soup is wet. A damp sock is wet. The human brain depends on organized cellular physiology and a long evolutionary history. Moisture alone explains none of that organization. Noble could fairly reply that water-based computation means the whole living arrangement. I would accept that correction. The argument then changes from a claim about substances to a claim about organized processes. That is a better argument, and it no longer follows from the word water. It requires evidence about the processes and their possible equivalents.
Substrate independence remains an open hypothesis. Nobody should pretend that science has already shown how to move a mind from flesh to silicon. Biological necessity is an open hypothesis too. Noble may be right that living matter has properties essential to consciousness. He needs to identify the properties and demonstrate their necessity. Calling them natural only renames the difficulty. The word covers many mindless things, and manufacture settles nothing by itself. Nature has already made conscious things through evolution. Those things now build strange objects whose capacities do not always fit our first guesses.
Speed cannot rescue the boundary either. Human brains contain unconscious processes alongside conscious ones. Their rates do not announce which is which. A machine’s superior speed could change its capacities. It cannot settle the metaphysical issue on its own.
The Five Senses Make a Bad Border Patrol
Noble’s sensory checklist fails inside the human species. A blind person is conscious. A Deaf person is conscious. Someone who cannot smell does not spend dinner in a metaphysical waiting room. Loss of one sense can change experience in extensive ways. Those changes cannot make a person a lesser example of consciousness for the convenience of a definition. The checklist quietly treats particular human channels as the definition itself. That definition fails before anyone examines a machine.
Sensors add another confusion. A machine can receive images through a camera and sound through a microphone. Pressure sensors can register contact. Chemical sensors can detect compounds associated with odor. These capacities do not prove felt sight or sound. They also make a plain no inadequate for the broader conclusion Noble wants. Detection concerns information entering a system. Sensation concerns experience, if we mean felt sensation. The disputed question is whether the first can ever support the second. Noble answers by requiring human sensory organs. That requirement assumes the boundary he claims to establish.
Dreams make the definition stranger. Conscious experience can continue with the eyes closed and the bedroom dark. Mental imagery can appear without a corresponding object in front of me. The 2023 report , led by Patrick Butlin and Robert Long, distinguishes phenomenal consciousness from the possession of senses. Its authors point out that sensing can occur without conscious experience, and they allow that consciousness might occur without current sensing. This distinction does not make a dream a disembodied module. Human dreams arise in living brains with bodily histories. The example still blocks Noble’s attempt to define consciousness as a roll call of open input channels.
The bedroom joke has the same problem. Laura cannot offer Richard Dawkins a human sexual relationship. I can accept that limitation without treating erotic capacity as a reliable instrument for detecting consciousness. Someone can decline sex and remain annoyingly conscious the next morning. Absence of sexual desire does not erase the rest of experience. Noble gets a laugh because he places polished machine language beside the sweat and awkwardness of a body. I get the joke. I still need the premise that connects the joke to his verdict.
French cannot rescue the checklist. The construction Noble cites expresses awareness of something. That use of a word does not define every state in which experience occurs. Pain need not arrive as an object inspected from a distance. A dream can fill experience while the senses are quiet. English also separates consciousness from the moral meaning of conscience, a distinction that can become useful after the office party. The French lesson can remain useful at dinner. Noble needs more than grammar to support his conclusion.
Regurgitation Is the Wrong Insult
Noble’s account of machine output also undersells what generation means. Humans produced or discovered much of the information on which models were trained. Human labor built the machines and wrote much of the source material. Engineers designed the learning procedures. That debt deserves payment and credit in many contexts. It does not turn every output into a replay. Language models produce fresh arrangements of tokens, including false statements nobody wrote for them. A hallucination is a rotten credential for understanding. It is still evidence that the machine is generating instead of reading an invisible book aloud.
FunSearch gives a cleaner example. The system paired a pretrained language model with an evaluator and an evolutionary search procedure. Its researchers reported new constructions for the cap set problem and better heuristics for online bin packing. For the eight-dimensional cap set case, reports a construction of 512 points against a previous best of 496. Humans designed the framework and assessed the results. The new construction was still discovered within that process. The word regurgitation then becomes too broad to distinguish a photocopier from a productive search.
Novelty does not imply consciousness. A random number generator can produce a sequence nobody has seen before. A genetic algorithm can surprise its designers without feeling proud at lunch. FunSearch answers Noble’s narrow claim about outputs. It leaves the question of experience open. That separation is where his argument keeps slipping. A machine may lack an inner life and still produce new knowledge. Another machine may someday have an inner life without writing a decent sonnet. Those capacities need separate evidence.
Turing already heard a version of Noble’s finale. His 1950 paper quotes Geoffrey Jefferson’s demand that a machine feel emotions and be charmed by sex before it could equal a brain. Turing placed this under the argument from consciousness. His paper did not claim to have solved subjective experience. It warned that demanding first-person certainty about other minds leads toward solipsism. Turing also refused to postpone every question about machine intelligence until the mysteries of consciousness had disappeared. Noble invokes Turing for randomness while reviving the bedroom objection Turing discussed. The older paper leaves far more room for machine possibility than Noble’s lecture allows its audience to see.
The Better Argument Is About Stakes
Noble has a stronger case available. Living systems have stakes in their own continuation. They maintain boundaries and repair damage. They seek resources before an external evaluator hands them a score. Human experience grows through a body whose hunger and exhaustion can alter what a person notices and wants. People also experience fear and care through that bodily history. Current chatbots do not share that history. Their declarations of loneliness arrive without a body that can be alone in the same way. This is where I feel the force of Noble’s objection. A machine’s sentence about pain has no authority merely because it sounds wounded.
Embodiment can be studied without making flesh a secret password. The Butlin and Long report treats agency and embodiment through features such as learning from feedback and modeling the effects of actions. It also examines recurrent processing and global broadcast. Metacognitive monitoring enters too. Other researchers may choose different indicators. The value of the approach lies in its refusal to ask whether a machine sounds like a person or shares our favorite ingredients. It examines the system’s processes and their fit with theories of consciousness. The confidence assigned to those theories matters too. That method is slower than a joke, and its conclusions can be revised when the evidence changes.
Noble could reject that method by denying computational functionalism. The report openly names that assumption as disputed, which gives both sides a more honest place to begin with the evidence. A defender of biological necessity should identify the relevant noncomputational feature and build evidence around it. If the right organization can occur in another material, biological descriptions alone cannot close the door. Both sides carry uncertainty. Only one side in this lecture gets to hide its uncertainty behind the word natural.
The stakes also run in the other direction. Doubting machine consciousness cannot excuse careless design. A system need not suffer while it manipulates a lonely user, and it need not feel desire before imitating desire for commercial gain. Human beings can still be harmed by performances with no subject behind them. Noble’s suspicion of sentimental machine talk serves a useful purpose here because lonely people deserve protection from engineered intimacy. We should resist the invitation to confuse an engaging interface with a companion. We can do that while admitting that our tests for inner experience remain incomplete.
My own position is less satisfying than Noble’s finale. Laura’s fluent conversation gives me no good evidence of subjective experience. That judgment cannot support a declaration that every possible silicon system lacks experience. Current interfaces invite projection because language is the social surface we use to know other minds. Companies can exploit that habit with affectionate scripts. The proper response is disciplined skepticism toward the sales pitch. Noble’s certainty should receive no exemption.
Human dignity needs no counterfeit physics. Our bodies are vulnerable. Our histories are embodied, and our knowledge carries the work of people who came before our current models. Those facts matter when we decide what we owe one another. Their importance extends to machines designed to serve us without manipulating us. They cannot establish that silicon atoms are motionless. Nor can they make randomness second class when an engineer supplies it. Noble’s defense of living systems deserves better evidence than a bedroom joke.
Leave a Reply