Confusions in Artificial Intelligence Friday, Dec 22 2006
AI 9:46 am
Most everyone has heard about the UK gov’t report, sponsored by chief scientist Sir David King, that says the arrival of sentient robots by 2056 is a real possibility, and if it does happen, robots will deserve all the rights and responsibilities as human citizens. This is a wonderful report, which shows how forward-looking the UK can be. However, this isn’t the first time a report like this has come out, so for me, what is particularly interesting is not the report itself, but the myriad reactions to it.
There is a lot of confusion around AI, created by a complex mix of science fiction and folk psychology at their worst. The interesting thing is that many of the spacey questions you hear people ask in reaction to this report actually have definitive, yes-or-no answers. For example, “can a machine think?” The answer is an unqualified yes – humans are machines, and we can think, so machines can think. Philosophizing, references to computer science degrees or Asimov novels, etc., are all unnecessary – we know the answer to the question, and the answer is yes.
Let’s try another question, a little more complicated: “How will you ever know a robot is truly sentient or just programmed to mimic it?” The answer rests on the exact, physical definition of sentience, which we have not discovered yet, but if it does exist, then we will eventually discover it precisely, and be able to answer whether or not a given physical object is sentient by looking at its internal physical structure. If we discover that there are gradients of sentience, then we will make lists of physical qualities that correspond to those gradients, and be able to determine which gradient the agent belongs to by checking for those physical qualities. In the future, there may exist robots that are programmed to mimic sentience, or even spontaneously generate sentientlike behaviors (not explicitly programmed in!), but are, in fact, non-sentient. In this scenario, the simplistic Turing Test would surely not be enough, because it could theoretically fool the judge. Again, you’d have to check the physical structure of the AI, either by some sort of scanning or a printout of its source code. The nice thing about reality itself is that it has very low ambiguity on the macro-level: some things are a certain physical structure, and others are not. The mind is what the brain does, so there are certain types of minds that are sentient, and others that are not. If the first test we imagine can’t tell the difference, then we need to devise another test, and use it.
Another type of reaction to this issue has to do with motivation. Here is what a commenter on Digg had to say:
“I just can’t see robots ever being given that sort of autonomy. Artificial intelligence gets put in charge of things we can’t do, not things we can. Why would you want to put it in charge of a humanoid body? We already have plenty of those.”
This comment is neither malicious, nor truly ignorant, but is a standard example of status quo bias, which could happen to anyone. Why change things from the way they are? However, a few small points are being missed here: 1) AI does do things we can do, like chess and sorting tasks, but they quickly become things we used to do, because the tasks get outsourced to AIs, 2) it only takes 1 engineer out of millions to decide to put an AI in a humanoid, or otherwise autonomous body. For #2, there are numerous chances for the commenter to be proven right, but all it takes is one dissenter to be proven wrong. Especially if the bodies an AI were given had the ability to reproduce, this “aberration” could very rapidly magnify.
Many people think AI is not coming so soon:
“If any of you Asimov-wannabes actually reads the article, you’ll get the refreshing truth, for a change. “Anyone who expects robots to start protesting and paying taxes in their lifetimes have spent a little too much time living in a rich and detailed fantasy world.” That’s all there is to it. If any of you takes it for granted that computers will become sentient “one day” or “within 50 years”, you are a gullible moron. None of you understands computer science, nor reality, well enough to know WTF you are talking about, and you prove your ignorance with your lame and played-out predictions. You nerds just repeat what you hear from each other. And it’s all make-believe shit. Meanwhile, back in the real world, there are no fucking sentient robots. Just AMD and Intel computers. That’s all. There aren’t even any PLANS to on HOW to build sentient robots. Not even those. All you will ever find are plans to MAKE plans. Those are everywhere. And fictional stories. All progress in this field is fictional. The closest thing we have to an artificial being is the latest Tickle Me Elmo doll. Everything else is, and will always be, frantic, sweaty, nerd hand-waving. And anyone who takes the AI field seriously is in a deep, ongoing state of delusion. Wake up, geniuses. For your own mental health, if nothing else.”
- commenter on Digg“Anyone who expects robots to start protesting and paying taxes in their lifetimes have spent a little too much time living in a rich and detailed fantasy world.”
- Adam Frucci, with scifi.com
Arguing that true AI is likely to be invented within the next twenty or so years is not easy, but many intelligent people do it, despite many other intelligent people disagreeing with them. Here are a few arguments, some of which have been given in recent popular books, but I won’t go into them here.
Another comment:
“Robots will no doubt one day become intelligent and sentient, but they’ll have no feelings and not suffer, so why afford them rights?”
- Digg commenter
This is another problem in predictions of robots/AI – saying that they’ll all be a certain way. In truth, if we do create intelligent robots, some will be able to have feelings and suffer, some won’t! That’s because certain physical structures correspond to qualiabearing (feeling-having) beings, and certain ones don’t. We know for a fact that humans do, but rocks don’t, for example. We’re pretty sure that lobsters and Windows XP don’t, as well. But as our software and robotic systems get increasingly complex, that boundary will get fuzzy and eventually lines will be drawn – with some robotic systems on the side with feelings, some not.
Another:
“But of course, as we haven’t even begun to fully grasp how the brain actually works.”
- Digg commenter
This is a classic one, and quite amazing. There are thousands, if not millions, of lengthy books and scientific papers on various aspects of brain function, much of which has been experimentally confirmed. This field is called cognitive science. We know a tremendous amount on the brain, so much that no one man or woman can hope to learn more than 1% of the field. However, it’s true, despite all that we know, much still remains unknown. But will we need to understand the human brain in its entirety to make artificial intelligence? The human brain is only a particular instantiation of intelligence, like an F-18 is only a particular instantiation of flight. If the F-18 were the only functioning example of flight that we had, would we need to understand it in its entirety in order to duplicate the functionality of flight? Not at all. That people think we need to fully and completely understand the human brain in order to make anything intelligent at all is pure anthropocentrism – as if our brain is the only possible physical structure in the universe that can have true intelligence.
And finally, possibly my favorite:
“this ruins the whole point of robot slaves/servants
if a robot gets rights, why dont we basically give tvs and computers rights. Robots will never be living things with real feelings and emotions, so why do they need rights. I would care if someone beat up their robot about as much as i would care if they through a brick at their tv. i.e. they’d be stupid but i wouldn’t give a shit.”
- Digg commenter
I actually do think that the majority of all AIs/robots will indeed be mindless slaves (to other robots as well as humans), and it won’t matter, because they won’t have any feelings to hurt! Transhumanists should definitely understand this point, and shake away the feeling that making any intelligence live to be a slave is fundamentally bad. Making any conscious being be a slave may be bad – but you can have intelligence without consciousness!
Anyway, it’s funny to think of a conscious AI as “ruining the point” of the robot slave paradigm, which derives from old-school science fiction.
Also: here’s Robin Hanson on why you aren’t entitled to your opinion.
Consciousness as a subjective experience is actually an illusion.
Illusion or not, it clearly exists, and has ethical weight. In fact, ethics and morals are defined around it. An illusion as important as this is vastly more important than many things that are not illusions, remarkably!
“And it’s all make-believe shit. Meanwhile, back in the real world, there are no fucking sentient robots.”
Yup. After all, in 1880, there were no flying machines, digital computers, or plastic bags, and no serious plans to make them. Therefore, they are all clearly impossible and make-believe.
“Just AMD and Intel computers.”
In 1970, there were mainframes. And more mainframes, and that’s pretty much it. Therefore “home comptuers” are completely impossible and should be disregarded even as a future possibility.
“There aren’t even any PLANS to on HOW to build sentient robots.”
Really? Have you been living in a hole for the past forty years? It seems nowadays like every researcher and their dog has their own plan.
“All progress in this field is fictional.”
Yes, we all know that all of the progress in understanding intelligence, the advances in hardware, the thousands of papers published, new algorithms, books, ideas, and whatnot, are all completely fictional. You see, neither the Wright Brother’s or the Red Baron’s plane could carry 300 passengers, so clearly any progress in aviation must be fictional.
“The closest thing we have to an artificial being is the latest Tickle Me Elmo doll.”
Not to mention many private research systems, countless public commercial software programs, and God knows what else. Do you seriously think that the Tickle Me Elmo doll is more advanced than, say Eurisko, Deep Blue, or the automatic crash-control program in your car?
“Everything else is, and will always be, frantic, sweaty, nerd hand-waving.”
You remind me of Lord Kelvin, a late-19th century British physicist. And he actually knew stuff! Some neat quotes:
“”There is nothing new to be discovered in physics now, All that remains is more and more precise measurement.”"
“X-rays will prove to be a hoax.”
“I accept no theory of gravitation. Present science has no right to attempt to explain gravitation. We know nothing about it. We simply know NOTHING about it.”
“We know that light is propagated like sound through pressure and motion.”
“Space is continuously occupied by an incompressible frictionless fluid”
“[Is there any matter not subject to the law of gravitation?] “I think that I may say with absolute decision that there is. We are all convinced… that ether is matter.”"
“[The vector] “has never been of the slightest use to any creature.”"
“Radio has no future.”
“Wireless [telegraphy] is all very well but I’d rather send a message by a boy on a pony!”
“Writing to Niagara Falls Power Company: “Trust you will avoid the gigantic mistake of alternating current.”"
“”I can state flatly that heavier than air flying machines are impossible.”"
“Landing and moving around on the moon offer so many serious problems for human beings that it may take science another 200 years to lick them.”
“The limitations of geological periods, imposed by physical science, cannot, of course, disprove the hypothesis of transmutation of species; but it does seem sufficient to disprove the doctrine that transmutation has taken place through ‘descent with modification by natural selection.’”
“War is a relic of babarism probably destined to become as obsolete as duelling.”
“but you can have intelligence without consciousness!”
How so? Please explain.
Hot air balloon was invented in 1783.
I feel the wrong questions are being asked. AI will represent collective conciousness. Let me explain.
If a company were going to set up a public relations department the first thing it would do is to construct a website and queries would be an appropriate piece of that website. We do not very often give considered off the cuff statements, all our statements are largely records of our memory.
All competitors in Turing/Loebner have adopted this approach. There is reason to suppose that if the semantics of any question could be analyzed correctly we could pass the Turing Test. Hence my statement about the website being the essential entity rather than the company’s super intelligent computer.
My second point concerns operating systems. AI will not develop on a single machine, it will develop on one super duper computer which is essentially the Web itself. In other words we have a logical program which runs on a number of machines and is intelligent. Replacing one machine will not alter this intelligence. Ocean Store is designed to provide a framework where computers can join and leave a network and where less powerful computers can be replaced by more powerful ones seamlessly. Hence the scenario described is total nonsense. Ocean Store will store the master copy of a file in a Reed Soloman code. Storage is on 64 computers with any 16 needed to reproduce the information. Hence if you withdraw one computer the information will be in tact and will (eventually) be copied back.
Let us consider semantics. Google translated “The season of spring” as “La estacion de ressorte” Just look at the Spanish covering up the English and you have visions of elastic stations, or perhaps of foot bridges that sway like the Millennium Bridge before it was put right. The way to beat Turing in fact is to have a nonsense sentence that is grammatically – El barco attravesta una cerradura another Googleism. Think in a second language and you are certain to beat any of the current generation.
Current research in semantics addresses this. Get a Spanish text, apply Latent Semantic Analysis and you will get reasonable translations from English. A number of companies (eg. Hakia) are doing searching using semantic analysis. This is the real way forward.
What does this add up to philosophically? Perhaps we should be thinking of AI as common cultural understanding – a common mind, rather than a supercomputer with human feelings.
To be sure they are ethical issues but they relate much more to privacy and the accessability of the information held about us than “robot rights” as such.
I am a retired scientist with a strong interest in AI. If you have any question I will be pleased to expand and clarify what I written.
Ian Parker
Interesting stuff.
Indeed – and according to one expert in the field, you can have artificial consciousness without intelligence! It’s beyond my limited knowledge to decide which (if either) situation would be the more useful or desirable, but it’s certainly a rewarding line of philosophical enquiry.
It doesn’t take any finagling of definitions to show intelligence in a chess-playing program. Consciousness is typically more wishy-washy to define. Self-reference (ala Douglas Hofstadter) and deliberative thought (ala Yudkowsky’s LOGI) seem good places to start.
Also, ‘sentience’ is used where ‘sapience’ might better suit for some of the above.
By ‘feelings’, we usually don’t simply mean ‘sensory input’, but ‘consciousness’ of the same, right? Where does one start with ethics? Fairness to apparently similar minds? How to draw the line without anthropomorphic hypocrisy?
By the way, Ian, you might replace some of those ‘will’s with ‘may’s. :)
[...] 2 – Confusions in Artificial Intelligence “In the future, there may exist robots that are programmed to mimic sentience, or even spontaneously generate sentientlike behaviors (not explicitly programmed in!), but are, in fact, non-sentient.” (tags: philosophy futurism robots machines consciousness sentience intelligence artificial AI) [...]
There are two questions regarding an AI. First, is there intelligence at work. We will assume the answer is “yes.” Second, whose intelligence is it? If we assume that the intelligence in a computer is an extension of that of the programmers, then an AI is not a separate being at all.
On the other hand, I don’t think this applies to an AI created by genetic algorithms. One argument against Intelligent Design is that it implies that we are God’s robots instead of separate beings.
As usual, wonderful discussion. THANKS much to Tom (McCabe), again as usual, for his always spot-on contribution(s)!
randpost: “Consciousness is a illusion” is logically incoherent, since the *concept* “illusion” *presupposes* awareness, i.e. consciousness. Brute quale (e.g.,I see yellow now) is a brute (albeit subjective, internal-to-the-organism) FACT **for the experiencing agent**. Metaphysically (ontologically) there is simply no way around this (so-called eliminative physicalism, reductive physicalism, etc., damn well NOTwithstanding). On all this see, David Chalmers, *The Conscious Mind: In Search of a Fundamental Theory* (Oxford U Pr) as well as Gregg Rosenberg, *A Place for Consciousness: Probing the Deep Structure of the Natural World* (also Oxford U Pr). See also the splendid little book by philosopher Wallace I. Matson: *Sentience* (U. Cal. Pr). Now the latter has been thought to argue for a reductive take on sentience, but I think one can read it as making a case for a dual-aspect ontology. In any event brute quale is not something one can merely dismiss as an “illusion”.
And Chris is certainly spot-on to point out that what we really want in an AGI is not merely *sentience* but also *sapience*, i.e., the ability to reason, to think, to not only deal with concepts, but to *originate* new concepts (such as, e.g., new theoretical entities or mechanisms—on which, btw, see, btw, e.g., Rom Harre’s classic discussion in his (now unfortunately out-of-print) superb treatise *Principles of Scientific Thinking* (U. of Chicago Pr, 1970) ). While in terms of the bio-cognitive development(s) on Earth, so far, sapience has presupposed (and followed-on shortly after) (the development of) sentience, it may be possible to have robust sapience w/o having all that terribly robust a sentience in an AGI. But, if I understand it (rudimentarily, at least), it seems that both Yudkowsky’s and Goertzel’s projects will produce sentience concurrent with sapience. (This also would certainly seem to be the surmise of, e.g., Hans Moravec.) But pursuing robust sapience w/o robust sentience seems to me to be a not-outlandishly-unreasonable prospect someday.
As for the whole robot-slave thing: If a robot is a *person*—has the “property”, if you will, of “personhood”—then it *ipso facto* would have basic rights, and couldn’t morally/righteously be “enslaved”. However, not all robots, even ones with robust sentience and/or sapience need necessarily be deemed to have full personhood. There **will**, however, almost beyond any reasonable doubt, be robotic/AGI entities which WILL qualify as persons, and who thus will have to be treated as such. But if Moravec, Goertzel, Yudkowsky, et al, have their way, these robotic persons—eventually many, many orders of magnitude more intelligent/sapient than un-enhanced, garden-variety humans (“humants” as I sometime whimsically call us)—will nonetheless be, *willingly*, what amounts to both our “servants” (but note the scare quotes just then) as well as our mentors. The next 10-30 yrs will surely tell…
And Joseph is surely correct in pointing out that Intelligent Design theory ultimately implies that we are God’s robots. Whereas Gardner’s “Biocosm” meta-model implies that WE are, for all epistemic and ontological purposes (especially in light of the so-called Fermi Paradox), what the late Gerard O’Neil called the PRIMANS: We may very well be the first an so far only intelligent, sapient life in the entire cosmos (though, especially in light of recent [meta]cosmology, not the only intelligent/sapient life in the entire multiverse, much less in all of what the late Robert Wesson termed the “MetaCosmos”): That is, in the entire spacetime that we now inhabit, and way beyond its current light-boundary, we may very well be the very first in the “sapience club”. Yet on Gardner’s cosmology/ontology, we play (and indeed must play) a vital role in (meta)cosmic evolutionary/development. A big responsibility? You bet! That’s why it is indeed vital that we as a species—individually and collectively—get our act(s) together over the next 5-20 yrs. We have to radically reconsider not merely WHO we are but WHAT we are…
You are not what you’ve been taught (on this see the superb works of John Taylor Gatto, btw). You are not what you’ve been told. YOU—WE—are **COSMIC**. Wake-up and smell the coffee—RIGHT WHERE YOU ARE SITTING NOW…
Happy Holidays, kids; and Happy & Prosperous New Year!
Live Long & Prosper…Love always…
randpost :
Consciousness as a subjective experience is actually an illusion.
Oh! Yeah?
Please how do you sort out “illusion” from “non illusion” for something for which there is NO access except introspection?
I mean, for everything in the “outer world” you may devise experiments thru various devices and access paths to cross check your theories.
For anything “from inside” like any qualia, color, taste, smell, pitch, CONSCIOUSNESS, etc… you may ONLY investigate correlations between the reported “feeling” and some EXTERNAL event/object a candle, sugar, a turd, a frequency, a FMRI image etc…
There is NO way one could discover, for instance, that one’s own vision has altered “feelings” about colors, say swapping the “feel” of red and green if this does not entail distortions in the color spectrum recognition (like daltonians).
Thus, how do you know MY consciousness is an illusion (or not), or even YOUR consciousness is an illusion (or not)?
Pretty much as an useless question as “how many angels can dance on the head of a pin?”
Ian Parker : I feel the wrong questions are being asked.
Me too but probably not in the same way, rather in the way of Donald Michie
AI is a collective name for problems which we do not yet know how to solve properly by computer. Corollary: Once we solve a problem, it is no longer AI. Therefore, there are no solvable AI problems.
Still quoting you :
Current research in semantics addresses this. Get a Spanish text, apply Latent Semantic Analysis and you will get reasonable translations from English.
Yup, LSA is “interesting” but does not shed much light unto semantics and is not the currently most promising path of research, see Jaime Carbonell et als. Context-Based Machine Translation.
Also of some interest for Language, Meaning and Automatic Concept Learning are Dominic Widdows publications.
OTOH for a contrarian (but well though out) view on Natural Language Processing see Is Evidence Based Linguistics the Solution? Is Voodoo Linguistics the Problem? by Alex Gross a professional translator.
Michael,
I wanted to point out that if humans are really just machines, then there may be no difference between a human and a machine programmed to mimic a human.
The standard counterargument to this is related to the notion of qualia; however, it is hard to present any evidence of the existence of qualia other than appeals to intuition. Additionally, qualia do not seem indispensable: it would seem you can find consistent theories of consciousness without appealing to the notion of qualia. This is still a subject of much debate.
Best Wishes,
Armchair Guy.
Well, an attack using semantics is pretty weak. Consciousness is an illusion/it doesn’t exist. Pretty much the same thing. Nothing you can say can prove consciousness, becose it is possible we are just a bunch of molecules. I can imagine a world which would be exactly like ours but would only be made by atoms, and in this world these beings made by evolution would discuss consicousness. But of course they would just be “zombies”. And this world could be the one we live in.
Believing in consciousness is a religion. It requires some kind of supernatural entity, as current science can look at our brains and only see physical reactions.
Randpost:
Illusions do exist. If you hallucinate a purple frog, the frog may not exist in reality. But the mental state which resulted in the hallucination is real. Even though consciousness may be an illusion (or an emergent property of various brain processes), as you say, it still exists. Not only that, consciousness is a common phenomenon in the sense that everybody experiences it. Viewed this way, believing in it is not religion, but of course it is not quite science either.
Best Wishes!
“it is hard to present any evidence of the existence of qualia other than appeals to intuition.”
You do realize that each and every conscious sensation you experience falls under the category of “qualia”? And that these sensations can be detected directly on an MRI machine? See http://www.yudkowsky.net/bayes/technical.html for an explanation of how concepts like “qualia” and “phlogiston” are incoherent anyway.
“But the mental state which resulted in the hallucination is real.”
Classical map-and-territory distinction. The map and territory both exist (ie, are both physical objects), but they’re still very different.
“consciousness is a common phenomenon in the sense that everybody experiences it.”
Common in human society, not in the universe or even on Earth. Fly to a random spot on Earth and note how much consciousness you see (odds are, it’ll be in the middle of the ocean/desert/mountain range/jungle, so not much).
“it is not quite science either.”
Neuroscience has already gotten to the point where we can stick a wire in someone’s head and make them see blue. So yes, it is definitely science at this point.
“Nothing you can say can prove consciousness, becose it is possible we are just a bunch of molecules.”
What the hell? We know we are made out of molecules, and we know we have consciousness; therefore, I think it is safe to assume that consciousness is a property of molecules. The two are NOT mutually exclusive.
“It requires some kind of supernatural entity, as current science can look at our brains and only see physical reactions.”
As I said earlier, you can now stick a wire into someone’s head and make them see blue (or any other color; see http://www.geek.com/news/geeknews/2002Aug/gee20020815015888.htm). So we can actually see mental reactions via physical stimuli and vice versa; we must be getting close to a solution here. Damn philosophers confusing the issue.
“consistent theories of consciousness without appealing to the notion of qualia.”
Please show me; it seems like they would have to just stick in qualia under a different name.
“Please how do you sort out “illusion” from “non illusion” for something for which there is NO access except introspection?”
Easy. Wire up a ~2020-era high-resolution brain scanning device. If the relevant area of the brain lights up, it’s real; if not, it’s an illusion.
“you may ONLY investigate correlations”
That’s how all of science works. Bayes’ Theorem mathematically defines the idea of correlation, and science is a subset of the Theorem. Seriously, the only way we can describe, say, gravity scientifically is a correlation between GR and objects in the universe, a correlation between objects and our various instruments, a correlation between instruments and the photons they emit, and a correlation between the photons and the image we see in our heads. And the last two are so common we don’t usually even bother with them anymore.
Randpost: With all due respect, you’re still uttering logically incoherent GIBBERISH: As philosopher A.C. Ewing once put it: “Solipsism is **obviously false** since **I am not a solipsist**.” Similarly, “Zombie-ism” is just as OBVIOUSLY false, since **I** am NOT a “zombie”. This is LOGICALLY *AIRTIGHT* utterly **unassailable* in *incontrovertible*.
I can’t imagine how/why you don’t understand this rudimentary metaphysical point.
Tom McCabe:
“every conscious sensation you experience falls under the category of “qualia”?”
Really? Qualia refer to private, subjective, first-person sensations (to quote one of their strongest proponents, John Searle). It is not at all obvious that every sensation is a quale, or that there is anything unmeasurably, exclusively “first-person” about it. Unless, of course, you appeal to intuition and simply claim that there is.
“And that these sensations can be detected directly on an MRI machine?”
The very definition of qualia precludes their detection on any machine other than the consciousness of the “experiencer”. If you redefine qualia to include such surrogates as neural activity (which you are talking about above), then there is nothing subjective and private about them– which would defeat the purpose of their definition. As far as I know the measurable surrogates you are talking about are not called qualia in the literature.
“Common in human society, not in the universe or even on Earth.”
True… but so very irrelevant.
“Neuroscience has already gotten to the point where we can stick a wire in someone’s head and make them see blue.”
Yes. But making someone see blue is a far cry from creating consciousness (including the sensation of the ego). It is quite possible consciousness is an emergent property of our complex neural organization. There are some preliminary investigations of the sense of the ego that I have heard of, but no detailed scientific deconstruction. We can’t even agree on a scientific definition of the term.
“concepts like “qualia” and “phlogiston” are incoherent”
and
“Please show me; it seems like they would have to just stick in qualia under a different name.”
These sentences seem contradictory. Do you or don’t you think qualia exist?
Most forms of behaviourism, physicalism and functionalism don’t appeal to qualia. See also Dennett, “Consciousness Explained”. These theories seem mostly consistent, unless you claim the existence of qualia (in the private subjective first person sense, not in the measurable surrogate sense).
I don’t know whether qualia exist; it is just not obvious to me that they do.
MCP2012: You prove consciousness by claiming to be conscious? I would expect even you to see the logical fallacy in that.
“private, subjective, first-person sensations (to quote one of their strongest proponents, John Searle).”
Yes, that’s how qualia is defined. Note that I referred to the “conscious sensations” you personally experience.
“It is not at all obvious that every sensation is a quale,”
That’s how qualia is defined, as far as I am aware. Therefore “every sensation is a quale” is correct by tautology.
“The very definition of qualia precludes their detection on any machine other than the consciousness of the “experiencer”.”
Then we’re going to need a new buzzword soon, because with high-resolution brain scanning we’ll be able to detect a whole long list of neat stuff reliably and repeatably. Note that this is all a quibble over semantics.
“But making someone see blue is a far cry from creating consciousness (including the sensation of the ego).”
Consciousness is obviously not a single, unified entity; it has various bits and pieces, and we are slowly learning how to induce some of these pieces artificially. Just because we aren’t yet able to specify every single piece scientifically doesn’t mean that the whole thing is some sacred mystery forever beyond science.
“detailed scientific deconstruction.”
Okay, we may not be able to get a complete deconstruction yet, but there are many, many instances of bits and pieces being removed separately. Look at lobotomy and other neurosurgery patients, blind people, deaf people, victims of accidents like Phineas Gage, and so on.
“These sentences seem contradictory. Do you or don’t you think qualia exist?”
Okay, the first point is that things like “qualia” are ill-defined and ultimately useless in the same sense that “phlogiston” and “ether” turned out to be. The second point is that we still don’t know enough to construct a complete theory of consciousness, and so anyone proposing one will probably just have to stick in “qualia” under a different name as a blanket term for all the stuff we don’t understand yet.
randpost: Thanks, kid-o, for the “even”…love you, too. Would you not agree that the late A.C. Ewing’s (himself a sort of idealist, btw) refutation of solipsism is logically airtight: “Solipsism is obviously false, since I am not a solipsist…”? What Ewing brilliant illuminates is that solipsism, while perhaps not self-refuting, is nonetheless self-*stultifying*. No one can sincerely believe in solipsism, much less act on it as a premise. But “zombieism” is even more incoherent: A “zombie”, as used in philosophy-of-mind literature over the last 30 yrs or so, is an entity that is outwardly (and even behaviorly) indistinguishable from a normal, garden-variety human person, with the exception that this entity has no subjective mental states, subjective qualia, subjective experience(s) of any sort whatsoever. Setting aside HOW such an posited entity can successfully mimic a normal human, behaviorly, as the question is begged as to how such a zombie acquires/processes info from the outside(-of-itself) world so as to orient/direct its own such behavior, it should be obvious that any human being with subjective qualia and subjective mental processes (and I happen to be a member of that set, whatever anyone’s appraisal of my cognitive/intellectual ability[s]) is a living, embodied **instantiation** of a knock-down REFUTATION (or disproof, or falsification, if you prefer) of the “hypothesis” so “zombieism.” Now, granted, the “problem” of Other Minds remains (whether one is talking about carbon-based, wet, squishy minds, or minds on some other substrate[s]). But this is completely distinct (metaphysically, ontologically) from the “problem” of “qualia” or its obverse, the “problem” of “zombies” or the possiblity of across-the-board “zombieism” (by which I mean the thesis that “We” [humans] ARE ALL “zombies”. I’m here to tell you, randpost, completely *incorrigibly* and *incontrovertibly*, that **I** happen NOT to be a “zombie”, and thus stand as a embodied instantiation of the falsification of “zombieism”…***QED***. “Zombieism” IS indeed utterly false, precisely because **I** am indeed NOT a “zombie”. Now, due to the metaphysical and epistemological landscape (as it were) of the Other Minds “problem”, YOU, randpost, cannot be in an epistemic position to either confirm or refute that. You can only accept it. YOU, however, are in the same unique epistemic position to say something(s) about your own subjective experience(s). And if you want to say that you yourself are indeed a “zombie”—then I’ll take your word for it. But such a position on your part would, indeed, be self-*stultifying*. On self-stultifying positions in general, see Lewis White Beck’s classic, *The Actor and the Spectator* (Yale U. Pr.), which, like Matson’s *Sentience*, is a wonderful, concise little gem.
Live Long & Prosper, kid-o…(with or without consciousness…)
MCP2012: You don’t seem to understand my point. You claim to be consciousness but you offer no evidence for it. I can make a program in few lines which will print “I am consciousness”, but we can probably both agree that a simple program like that is not consciousness. For me you are just a bunch of molecules, a zombie, until proven otherwise. Debating about consciousness is very difficult and creates alot of problems precisly becose the whole premise of consciousness is proved by assumption, which is a logical fallacy.
“You claim to be consciousness but you offer no evidence for it.”
From his conversation, he’s obviously human, as there are currently no other entities besides humans capable of speaking intelligently. Therefore, all you have to do is get a textbook on human neurology and look up all the areas of the brain in which phenomena attributed to “consciousness” occur.
“For me you are just a bunch of molecules,”
You sound like the creationists- “evolution” means we are just “a bunch of molecules”, and that can’t be right, because, er, well, just because! Seriously- do you have a single scrap of evidence that consciousness and being built out of molecules are mutually exclusive? And do you deny the blatantly obvious fact that humans are made of molecules and follow the laws of QED?
I think you say it well, MCP2012. The only thing I see problematic about consciousness is that we don’t know how to reproduce it yet to do neat experiments like experiencing a central-observer jump from my original to my clone across the room back to my original, where the series of experiences is interreferential in the same way going to a dinner party and leaving a dinner party can be interreferences of the experiences, ‘expectation of leaving the dinner party later’ and ‘recalling having arrived at the dinner party’, respectively, and everything in between.
Science can only help us get to the point where we can do the engineering, even if iteratively. But a scientific paper with mere symbols will never be sufficient to know consciousness fully; it would simply be transferring the problem from eyes-closed cogitation to eyes-open cogitation. Deeply immersive VR computer models could come extremely close.
Until higher technological sophistication, I have no problem not being able to quantize my experience events and transmitting them perfectly to the brains of hard scientists who want to refer to an illusion right from “an illusion”. In information theory, there are accepted notions like ‘full description’; e.g., when you have a full description of something, “nothing is left out”. That is one of the main sources of hard scientist practice of trying to undermine the phenomenon of consciousness, I surmise. Another one of the main sources of that practice, as again I surmise, is that hardly anyone wants to state explicitly for critical evaluation, even though it seems clearly implied, “I assume determinism. I want my deterministic volition to be the farthest seeing and extend beyond, account for, and embody as many other deterministic volitions as possible. Part of a good strategy to do this is to know a lot. To reach ‘knowing a lot’, it’s easier if there’s little as possible to actually know about. To eliminate having to know about consciousness would be all the better on so many important fronts, especially the one where I’m given license to throw many more deterministic volitions into the inconsequential, quantized, categorical container labeled ‘Mystics’.”
I’m surely no enemy of science. I just don’t want to do it in certain self-stultifying manners, even if I have to trade the self-stultifying manners of ‘suicidal science’ for ‘suicidal science’.
randpost: Lamentably, dear colleague, you are confusing the “Other Minds” problem with the “Zombie” problem. Though philosophical cousins (if not, perhaps, siblings), **they are not the same**, and are, indeed, quite **distinct** (though, arguably, very, very closely-related). You’re most recent post/reply evinces that you’re hung-up on the epistemic (epistemological) problem of YOUR *knowing* that *I* (or any other entity possessing—or *claiming* to possess—consciousness or sentience) actually do have subjective experience/qualia/etc. or whether I am, in fact, a zombie. This is the Other Minds problem. With fellow-humans, I think we have to resolve the problem along the lines of those discussed/propounded in J.L. Austin’s classic essay “Other Minds” (See also Plantinga’s discussion in his book *God & Other Minds*). The “zombie” problem is usually one along the lines of “can we fully conceive [in a robust, non-question-begging way] of a humanoid (the notion of “humanoid” broadly/liberally conceived, btw) that is actually devoid of subjective experience/qualia, and yet is otherwise indistinguishable, physically/behaviorally, from a “normal” human?” Or is there something(s) left unaccounted-for, *functionally*—a *functional* “metaphysical dangler” as it used to sometimes be called? And I think both Chalmers (ref. above) and Rosenberg (ref. above), as well as others, make a good (if not very nearly unassailably airtight) case that one cannot dismiss consciousness, functionally, w/o running into some quandaries. But—nb—this so far has been primarily applied to philosophy of mind as it relates to humans and humanoids. Now, admittedly, this literature has, for about 4.5 decades now, overlapped with the whole AI/android literature, to one extent or another (See, e.g., Mike Scriven’s superb [and somewhat underappreciated] classic essay, “A Compleat Androidology”, 1961 [reprinted in various anthologies over the ensuing yrs].) But the problem of both zombies and Other Minds (in the sense of evidentiary/verificational procedure was discussed by none other than the late Paul Feyerabend in an article (whose title eludes me, forgive me) in an early volume of the series *Minnesota Studies in Philosophy of Science* in which he introduced the notion of a “cerebroscope” with which to peer into the “brain bubblings” of another human being’s cranium. (This is also discussed, btw, in Matson’s superb little book, *Sentience* ref. above). But, of course, all the observe sees is indeed brain bubblings and what-not—*sentience* is no*where* to be *found*. The metaphysical/ontological solution to this, of course, going back to Spinoza and Leibniz, and continuing into modern philosophy with Russell, Whitehead, Sellars (more Sr. Roy than Jr. Wilfred), and cointinuing up thru Gregg Rosenberg (ref. above): A dual-aspect ontology wherein sentience and qualia are an internal (internal-to-the-organism) *aspect* of physical processes (this is my take on John Searle, btw, no matter what it may himself say!).
But—and I’m running out of time here—I will agree that the Other Minds and Zombie problems more-or-less converge when talking about AGI’s on a different substrate than human beings. How will we know they’re sentient? Do they even have to **be** all that especially *sentient* to nonetheless be ultra-ultra-**SAPIENT**?! These are merely a few among the myriad questions that the current development of technology present to us. And, while philosophy can certainly help, it has yet to provide any clear “meta” level guidelines (though Ben Goertzel’s recent stuff on **patterns** may indeed be a promising avenue of investigation). We are, I concur, randpost, faced with an Other Minds problem (and a Zombie problem too) regarding supposed **minds** on different substrates. Brain interface is one possible epistemic solution. Would brain-interfacing with a sentient AGI be experienced(!!!) by the human as a “meeting of minds”? We shall probably soon see.
I *do* understand your point(s), randpost. We were, perhaps, at least in part, talking past one another.
Ciao for now…
In case attention is placed on my preceding remark, whether in some approval or full disapproval, I need to add clearly that I personally tend to posit determinism to help ensure my condition is maximally disastrous, only something to make a little better. To posit determinism, it seems, is also to help conceptualize volition in such a way as to be less distracted by issues resulting from the possibility of sentience in this or that object. We should be distracted by something else, if only in addition to fundamental self-righteous concerns.
This particular conception borrows from the previously mentioned ‘sapience’ concept.
If we have determinism, there exist deterministic volitions. Besides it being logically and physically impossible to achieve causa sui, the only other tragedy with this reality is that ‘volition’ is simply the sense of agency in the absence of agencies which are fast and smart enough to appear as precognitive, along with the absence of being able to predict in divine infinitudes. It’s a tragedy insofar as there is nothing impossible about having tea with a faster and smarter precog, expectedly even in some decades. While it’s amusing that this precog would be accurately describing to you non-you events before they happen, like which sequence of the next 20 moves the waitress will make shortly, you could never know it could do the same of you except through ex post facto reports.
That can be really good, or that can be really bad. Those who want to complain about the probable logistics when sentient robots are a part of society, or those who only like to ridicule, should have something actually worth being cynical about. Logically, sentience is only crazy complex (and I have no problem with laws against members of Kingdom Animalia/Robotica obstructing volitions of any other members of Kingdom Animalia/Robotica just to be safe: “Now just stay outside of the 10-step movement course of that damn volitional robot arm of distinct proprietorship”), but relative precognition is scary complex (where jurisprudence is now inapplicable no matter what: “We need to get recursively self-enhancing AI right the first time”). Perhaps it’s time for all parties to be more open to the project of figuring out how to coexist as undetectable singletons, which would satiate both sides of the conflict: the pitiful cogs who don’t want to be wary of the all-powerful precogs that can do almost anything, it seems, except unhide their direct programming of the pitiful cogs’ subvolitions; and the precogs whose deterministic volition sees the choice and is deciding on, in this world as it would in all possible worlds, Virtual Perfection (only right next to causa sui).
The notion of “Consciousness” is, IMO just unfalsifiable folk psycology. Also, the term carries anthropocentric and biocentric conotations, that is why people sceptical about human-equivalent AI being possible obsess over the “consciousness” question, they think humans have this special thing machines can’t have and this special thing is “consciousness.”
Nathan B. Says:
“If one examines the precapitalist paradigm of discourse, one is faced with a choice: either reject dialectic narrative or conclude that academe is intrinsically used in the service of outdated perceptions of consciousness. In a sense, the subject is contextualised into a Debordist image that includes art as a totality. The main theme of la Fournier’s analysis of the precapitalist paradigm of discourse is the genre, and some would say the failure, of prepatriarchialist class.”
Okay, that wasn’t actually what he said; I stole it from the Postmodernist Generator at http://www.elsewhere.org/pomo. What he did say was that there are “precogs”, minds that can infallibly predict the future. This is neat in that it contradicts both science and math at the same time. Science says that the universe is made up of subatomic particles which interact via the various forces, and that these interactions are probabilistic (see the math of quantum electrodynamics for the details). Therefore, it is physically impossible to predict the exact outcome of any given event. Also, due to the sheer number of interacting particles, it is computationally infeasible to simulate a human brain with the degree of fidelity required; a very small group of atoms would be enough to shift a synapse from “on” to “off”, and this could ripple through the rest of the brain.
But even if we’re omniscient, have infinite computing power, and live in Newton’s clockwork universe, there’s a mathematical theorem called Rice’s Theorem which says that it is fundamentally impossible to determine whether a given Turing Machine (aka, us and every other computer) implements a given algorithm. Such a precognition machine could predict a Machine’s behavior, allowing it to derive whether or not it implemented the algorithm and breaking the Theorem. There is also the Halting Theorem, stating that it is impossible to predict whether any given Turing Machine would halt; such a prediction machine would obviously violate it by prophesizing whether the machine halted.
And before you postulate magical supercomputers that can get around all this, note that such a supercomputer would have to be built out of atoms. Atoms follow the laws of quantum electrodynamics, which can be put into a modern computer. Therefore, the supercomputer could be put into a modern computer, and therefore the modern computer can do anything the supercomputer can do given enough time and memory. QED.
I didn’t say that either, Tom.
“I didn’t say that either, Tom.”
Really? So I guess I must have been delusional again. Obviously your numerous references to them didn’t imply in any way, shape, or form that they were actually possible. Sorry, my mistake.
Counter-productive sarcasm – you’re talking about an abstractly intellectual topic – no need to get testy.
If Nathan didn’t say they were precogs, then you can’t confront him over that.
However, Tom has a point – at times, Nathan, your language sounds like the output of the Postmodernism Generator – you have to radically simplify… for example, your extremely long sentence:
…could be less opaquely phrased as, “Because we can’t yet create AI, and don’t have advanced nanotechnology, we can’t test the properties of consciousness thoroughly. This makes its processes less intuitive than the everyday processes of natural causality.”
You habitually refer to things in a roundabout way. This is just a matter of communication mismatch. Everyone has their own personal style of communication, and some ways need to be reworked more or less to fit in with the consensus method, which is necessary to transmit information to others.
I said they were precogs but referred to them differently than apparently interpreted. For instance, nothing about my reference entailed infallibility.
Well I do appreciate the constructive criticism, Michael, but you might use a somewhat different perspective if I explain. First thing to explain, I don’t want to explain, not even this. This is a locality where I perceive peers to interact with, not to dominate. Some of my language on this site may seem domineering, but it should be usually clear that I narrate from various points, sometimes across the same sentence, sometimes directly against myself at another point. Second thing. If I use a concept and relate it to another concept, it may sometimes apply to another concept I instantiate but to which I didn’t make a direct relation. If it applies, by all means apply it, and please don’t digress into a strawman about how I need to care about something miniscule compared to what I obviously care about. For instance, and third thing, a little while ago, I explicitly suggested (keyword there) that I don’t have a sacred relationship with natural language, or with natural language combined with pubescent variables. If someone else does, that’s their problem. It’s my problem to the extent that I will attempt passing grammar, spelling, and using relatively simple vocabulary, with occasional compounding but which is almost always modular. To the extent that I make it my problem – mainly with the desire to use very basic tools to acknowledge others – I happen also to instantiate relatively bigger and better problems which I don’t necessarily have to share with my natural language-speaking peers, who for some reason want to view conversation as laborious or a channel for their entire goddamn lifeline. Ignore me, block me, tell me to mind my own business if I can’t properly mind yours, but don’t try to correct the maximally inconsequential aspects of my behavior in such a roundabout way. This, not as a hostile threat, of course, but because now we together should see the net futility, especially if futility was already perceived as my starting position.
“If Nathan didn’t say they were precogs, then you can’t confront him over that.”
That is certainly true, but Nathan has clarified that he does in fact believe in de-facto precogs, just not “infallible” ones.
“I said they were precogs but referred to them differently than apparently interpreted. For instance, nothing about my reference entailed infallibility.”
Okay, so the precogs may not be mathematically infallible, but you’re calling for a level of accuracy which is still ridiculous to expect out of any actual computing machine- things like “which sequence of the next 20 moves the waitress will make shortly,”, as I said, rely on keeping track of extremely small groups of atoms moving about.
To do some simple math, suppose that a person has 10^24 atoms in their brain. These atoms are in a liquid or solid, so they are continuously interacting with other atoms. Suppose that we can safely keep track of just the interactions of each atom with 1000 others (that’s only around five atomic radii out, assuming the atoms are packed cubically). Atoms at room temperature move at around ~500 m/s (for a neat derivation, look at http://www.chem.arizona.edu/~salzmanr/480a/480ants/kmtmod/kmtmod.html). To simulate these reactions with a high enough degree of fidelity, say to 1/100th of an average atomic radius (~10^-12 m), you would therefore have to take a simulation frame every ~10^-15 sec. Assuming each interaction can be computed using one FLOP, that’s a rough estimate of 10^42 FLOPS. Even assuming you had rooms full of ridiculously powerful computronium to do all this calculation, conventional computing (Landauer’s principle) would require you to provide ~3*10^21 W of power, equivalent to the peak power flux in the atomic bomb that leveled Hiroshima.
Oh, and there’s still the problem of getting the initial location of all of these atoms to begin with. Even nanoscale scanning devices are very limited at the atomic level due to Heisenberg’s uncertainty principle. Oh, and there’s still the fundamental uncertainty of quantum reactions as explained by Feynman and others. And even if you could do all this, you forgot about the rest of the room that’s continuously interacting with his brain, etc., etc….
If you’ve found some way to accurately simulate someone’s brain over a wide range of circumstances without going through atom-by-atom simulation, you could win yourself a free trip into suborbital space at https://www.space-shot.com/. Collecting the prize requires you to predict the weather sixteen times in a row better than some other guy, and whatever solution you have will probably apply just as well to weather systems. The fact that nobody has claimed the prize yet means this must be difficult, but hey, important problems have been solved by amateurs before. And if that isn’t enough motivation, the same problems of chaotic systems apply to things like protein folding, so I’m sure you’d find a company willing to invest in you if you published a paper or demonstrated a working system. I would, if only I had large piles of money laying around; but plenty of venture capital firms have exactly that problem: large piles of money and nowhere to put it all.
“and some ways need to be reworked more or less to fit in with the consensus method, which is necessary to transmit information to others.”
Speaking of that, do you know where I can find a decent guide to learning how to write?
Jean-Luc Delatre Says:
Current research in semantics addresses this. Get a Spanish text, apply Latent Semantic Analysis and you will get reasonable translations from English.
Yup, LSA is “interesting” but does not shed much light unto semantics and is not the currently most promising path of research, see Jaime Carbonell et als. Context-Based Machine Translation.
Also of some interest for Language, Meaning and Automatic Concept Learning are Dominic Widdows publications.
OTOH for a contrarian (but well though out) view on Natural Language Processing see Is Evidence Based Linguistics the Solution? Is Voodoo Linguistics the Problem? by Alex Gross a professional translator.
I can reply in two ways.
1) I think your comments reinforce the view that progress is being made in language translation. Language translation is a central problem in a number of areas. VoIP is basically a translation problem as the problem is sorting words out according to context. Phonemes are recognized as accurately, if not more so, than a human.
2) Specifically on LSA. LSA has the advantage that is not dependent on human input. It is purely a stistical process. The paper you quote talks about pairs of words. LSA does this in essence. The vectors essentially associate allowed combinations together. DARPAs GALE project (Arabic, Chinese, English) is investigating all possible techniques, statistical and manual.
“VoIP is basically a translation problem as the problem is sorting words out according to context.”
Huh? VoIP stands for Voice over Internet Protocol, a system used to transmit real-time audio messages over the Internet (similar to the telephone system). This has nothing whatsoever to do with translation, as the audio is simply recorded with a microphone, digitalized, and shipped over the network. Translation programs can be added to VoIP systems, but VoIP is not, in and of itself, a translation problem.
I was referring to the conversion of voice into text. Once something is in text form it can be mined.
“I was referring to the conversion of voice into text. Once something is in text form it can be mined. ”
Ah. I believe Ray Kurzweil has developed a similar product. See http://www.kurzweiltech.com/raybio.html.