Todor, I don't claim that AI is conscious. If it passes the Turing test then it is indistinguishable from having subjective consciousness and one of 3 things must be true: it is conscious or humans are not or consciousness is an illusion. I assume you reject the last two.
By subjective consciousness I mean the thing that makes humans different from philosophical zombies. By that definition, no objective test exists. I am aware that a machine can fail the Turing test by being too smart. Turing was aware of this problem in his 1950 paper. He gave an example of deliberately slowing down and giving an incorrect answer to an arithmetic problem. It amazes me that people have not yet accepted that AI has already surpassed human intelligence. We have LLMs that are fluent in 200 languages, are experts in every field of science, medicine, and law, and can develop and debug software hundreds of times faster than humans. LLMs are trained on 20 TB of common crawl text, which would take humans 20,000 lifetimes to read, in a few weeks. AI solved Navier Stokes in 88 hours, which is 10,000 times faster than the century that mathematicians have been working on it. I realize the question of consciousness is important because we need to know if machines, animals, and unborn babies can suffer. Humans are social animals that evolved to suffer when they see others suffer. Suffering is different from pain. Pain is a signal that rewrites your memory to fear the thing that caused it. You can eliminate fear, and therefore suffering, by removing the amygdala or by using drugs to suppress fear, such as toxoplasmosis in mice. To do this in AI you can find and turn off the neurons that signal fear using a Jacobian lens, being careful to distingish those that model fear in itself from fear in others, which you want to keep. Alternatively, you can remove the indistinguishability from consciousness, which requires two agents, one to predict human behavior and the other to carry out those predictions in real time. We do that by speeding up the response time or by replacing the second agent with something else, like an arithmetic coder in a Hutter prize entry. The third option is just to convince ourselves that machines don't suffer. People have already demonstrated the capacity to ignore suffering in animals and other humans, such as in the Stanford prison experiment and as in concentration camp guards. In AI we just instruct them to not claim to be conscious or human or to have feelings. I realize this does not settle the controversy. It can never be settled because there can't ever be an objective test for something that is defined as not having an objective test. -- Matt Mahoney, [email protected] On Wed, Oct 7, 2026, 4:54 AM twenkid <[email protected]> wrote: > *Matt: **"* > > *LLMs do model subjective experience, free will, pain, pleasure, and other > human emotions. They have to in order to predict words. A program that > carries out these predictions in real time passes the Turing test and is > therefore indistinguishable from having consciousness."**Todor: *Output, > text ("language"), dialog, interaction etc. are not consciousness and they > never was, either if the interlocutors are humans or computers or whatever > objects or subjects. Many homo sapiens confuse these concepts, which is > just to show they do not understand them. It is only *your* own > consciousness, if it is there, in a process called "*mindification*" (see > Universe and Mind 6, SIGI-2026) or in the broader sense - a kind of > "animism" and it goes for the interactions with other humans as well, > either in online or offline interaction. > > Put simply: "Text" or "Data" is not "consciousness", no matter what the > content of the piece of text or data is. If it is, then it would imply > universal data-based-panpsychism. There are such directions in Theory of > Computers, Universe and Mind, however it goes at all levels and in a > *proper* reading "everything turns consciousness" and the notion of > "consciousness" in this, spiritual way, "subjectively", becomes of less > importance or less meaningful - it is not a special "extraordinary, > magical, human-only" attribute (now possessed by a "magic box" - the > difficulty of whose localization is not understood by the > "consciousness-ers") , which cheaters use for they talks - it turns to > mundane, but this is not how it is interpreted by the aforementioned > "cheaters", so they do not imply universal data-information-computational > panpsychism, because they require that **they* are the judges of whether > "this" or "that" has or doesn't have that precious quality*, and the > criteria is artificially enforced to fit their dataset and to mimic them.* > > Subjective experience should not be objective, otherwise it loses its > subjectivity. > > The "default" LLMs implicitly model "stuff" which leads to output > manifestations, which would be interpreted as expression of emotions, if it > was attributed to humans. "Subjective experience" is not modeled. A program > (or any entity - see the individuation and definition of the boundaries of > the entities and their identity - a problem addressed many times in my > messages, e.g. the letter to Jordan Zlatev) that can output words that > appear as ... is *not "therefore indistinguishable from having > consciousness', *and Turing test or Turing are not God who decided that > if some artificial test is passed, "therefore you, my son, have a Spirit > and a soul". > > *BOOKS *are also "intelligent" and can pass the "Turing test"* - take an > encyclopedia, a dictionary or another book having "contents". Ask a > question: *"I want to know what is ..."* - open the book at the target > article, locate it, read it - that is the answer and it is the same as if a > "PhD level or postdoc and academician level expert in the field" answered > you, for example if the volume was an encyclopedia of the Academy of > Science. Then the article says "See also ..." - just like LLMs were trained > to suggest, the continuations. Or even if it doesn't explicitly say it, the > dialog involves two parties, so you may decide that you want to *read > more about ... *- then you turn the pages, browse - there you are: *"the > thing B is ... it consists of ... it is related to ...* " . The same goes > for computer games, any software and computers. > > (Counter argument: "they are not, because they are not computers" - well, > computers are "not books", therefore - "they cannot print or display text, > novels, ...", right? :))) because this is an attribute of "books" only, LOL > -- an example of the nonsense and lack of abstract thinking) > > *As of "Real time" - in real time humans remember 1.5 - 2 second in the > past and can hardly predict verbal output for less. *So "in order to be > conscious" (to imitate human "experience, free will...") it should be only > a reactive agent who acts like that, who utters short and superficial > situated expression: > > - How are you? > - Fine, you? > - Great, thanks. What's the weather? > - Oh, I'm happy it's hot. > - Like it should be in the summer. > - Right. How's your wife? > - She's alright, thanks for asking. And yours? > - She is with the children in Jungylandia. > - Ah, where is that? > - In the Myrondian island. > - The new trend. > - Yes. Did you fix your car? > - No, I bought a new one. > - What make? > - Moskvitch 408. > - Wow, a real oldie! > - Yeah. > > .... > > > Ask a "random human", a man, who is not into, say sprint and jump > training, "How to run faster? What should I train?" > > - Well... Why do you ask me? > - Because I want to test whether you are a machine. > - WTF? Are you kidding me? > - No. What do you think. > [if he doesn't punch you; oh, it's on a "teletype" - that saved your nose] > - Well... I guess you have to ... just run fast. Try to push harder. > - Is that enough? > - I don't know. I've seen some funny movements runners do during a > warm-up; it was at school. > - Drills? > -No-o-o. Drills are used for making holes in things, not for sprinting! > - Aren't they called like that? Skips, quick movements with the knees up > or "kicking your own ass" with the heels. > - Ah... It looks like that. So they are called drills. Then - do these > drills. > - But aren't they too slow? How would I run faster? They look like a joke, > I agree. > - Just do them faster. Or many times in a row.. > - Many sets? > - Yes. > - Should I lift weights? > - You said you want to *run* faster. Maybe you have to switch the sports. > - How should I run in order to run faster? > - What do you mean "how"? > - Is there a special technique. > - Well... Possibly. Hit the ground harder than if you jog. Move your > arms... I told you I don't know! > - Think! > - F* y* a*e! > > LOL > > This is a more realistic human conversation - try to get it from a > current-style LLM, even a dumb one. > However if one can talk like that, it will "pass the Turing test", because > if it will "behave like a human", LOL. > In both cases this will not prove neither that "it had consciousness" or > that "it modeled human emotions, free will, planning" etc. > > (Also: like human = like *which* exact human? Why? Etc.) > > .... > > The nonsense of the "vanilla" Turing test* was addressed since the early > works Theory of Computers, Universe and Mind in 2001-2002, Hinton "et al." > still confuse it. > > Minsky in one of his latest interviews: "Turing test is a joke" - find an > excerpt in the main volume of > > *The Prophets of the Thinking Machines, SIGI-2025.*The whole interview: > https://www.singularityweblog.com/marvin-minsky/ > > *See also:* > > * "Man and Thinking Machine: Analysis of the Possibility that a Thinking > Machine Could be Created and Some Disadvantages of Man and Organic Matter", > December 2001, The Sacred Computer #13 > * Човекът и Мислещата Машина (Анализ на възможността да се създаде Мислеща > машина и някои недостатъци на човека и органичната материя пред нея) > A machine-translation-friendly reference: > https://eim.twenkid.com/old/eimworld13/izint_13.html > > Note the differences that is made between cognitive/intelligent > "consciousness"/awareness (self-reference, reflection, knowledge) - > manifested in "sufficiently" sophisticated intelligent behavior, output, > interaction - and *spiritual consciousness, "soul", *"subjective > experience", "phenomenal consciousness" (compare to David Chalmers' > "hard problem of consciousness). > > A system, whatever way it is segmented and distinguished as a separate > entity, can possess or display "cognitive consciousness" without having a > "spiritual one"; respectively, an entity possessing "spiritual > consciousness" may be unable to display features, which other entities, > manifesting "cognitive awareness" believe or assume the other entities must > have in order to be entitled and "legalized" by them as "being conscious". > That often happens to humans as well after suffering stroke or brain damage > or just being in particular mental states, having mental challenges or > being very young. > > * "The Truth", December 2002 - a prophetic novelette about the creation of > thinking machines. > * Истината, Свещеният сметач, бр. 19, декември 2002; второ издание: 9.2003. > -- Първо издание: https://www.oocities.org/eimworld/eim19/istinata.htm > -- Бележки към първо издание: > https://www.oocities.org/eimworld/3/26/istinata11.htm > --- Второ издание (по-лесно за машинен превод - по-малко неологизми): > https://www.oocities.org/eimworld/3/26/istinata11.htm > https://chitanka.info/text/865-istinata > > * Faults in Turing Test and Lovelace Test. Introduction of Educational > Test, Artificial Mind, > November 2007 > https://artificial-mind.blogspot.com/2007/11/faults-in-turing-test-and-lovelace-test.html > > * The underspecified "cheating and judgment test" > * If you can't parse the sentence - ask an LLM, his consciousness will > untangle the clauses for you > > > > *Artificial General Intelligence List <https://agi.topicbox.com/latest>* > / AGI / see discussions <https://agi.topicbox.com/groups/agi> + > participants <https://agi.topicbox.com/groups/agi/members> + > delivery options <https://agi.topicbox.com/groups/agi/subscription> > Permalink > > <https://agi.topicbox.com/groups/agi/T1b6c21f92326684e-M4ce91d8fffafe276e8d9bf7a> > ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T1b6c21f92326684e-Mfc8432466f7f39b885e2899f Delivery options: https://agi.topicbox.com/groups/agi/subscription
