Bet 15
By 2050 a synthetic computer or machine intelligence will become truly self-aware (ie. will become conscious).
Prediction 15
Duration 48 years (02002-02050)
Predictor
Nova Spivack
Challenger
TBA
Add your voice to a conversation with the bettors:
Bookmark this bet, and share it with friends:
(This summary includes my argument, a method for judging the outcome of this bet and some other thoughts on how to measure awareness...)
A. MY PERSPECTIVE...
Even if a computer passes the Turing Test it will not really be aware that it has passed the Turing Test. Even if a computer seems to be intelligent and can answer most questions as well as an intelligent, self-aware, human being, it will not really have a continuum of awareness, it will not really be aware of what it seems to "think" or "know," it will not have any experience of it's own reality or being. It will be nothing more than a fancy inanimate object, a clever machine, it will not be a truly sentient being.
Self-awareness is not the same thing as merely answering questions intelligently. Therefore even if you ask a computer if it is self-aware and it answers that it is self-aware and that it has passed the Turing Test, it will not really be self-aware or really know that it has passed the Turing Test.
As John Searle and others have pointed out, the Turing Test does not actually measure awareness, it just measures information processing---particularly the ability to follow rules or at least imitate a particular style of communication. In particular it measures the ability of a computer program to imitate humanlike dialogue, which is different than measuring awareness itself. Thus even if we succeed in creating good AI, we won't necessarily succeed in creating AA ("Artificial Awareness").
But why does this matter? Because ultimately, real awareness may be necessary to making an AI that is as intelligent as a human sentient being. However, since AA is theoretically impossible in my opinion, truly self-aware AI will never be created and thus no AI will ever be as intelligent as a human sentient being even if it manages to fool someone into thinking it is (and thus passing the Turing Test).
In my opinion, awareness is not an information process at all and will never be simulated or synthesized by any information process. Awareness cannot be measured by an information processing system, it can only be measured by awareness itself---something no formal information processing system can ever simulate or synthesize.
One might ask how it is that a human has awareness then? My answer is that awareness does not arise from the body or the brain, nor does it arise from any physical cause. Awareness is not in the body or the brain, but rather the body and the brain are in awareness. The situation is analagous to a dream, a simulation or virtual reality, such as that portrayed in the popular film "The Matrix."
We exist in the ultimate virtual reality. The medium of this virtual reality is awareness. That is to say that whatever appears to be happening "out there" or "within the mind" is happening within a unified, nondualistic field of awareness: both the "subject" and the "object" exist equally within this field and neither is the source of awareness.
This is similar to the case where we project ourselves as dream protagonists in our own dreams---even though our dream bodies appear to be different than other dream-images they are really equally dream appearances, they are no more fundamental than dream-objects. We identify with our dream-bodies out of habit and because it's practical because the stories that take place appear from the perspective of particular bodies. But just because this virtual reality is structured as if awareness is coming from within our heads, it does not mean that is actually the case. In fact, quite the opposite is taking place.
Awareness is not actually "in" the VR, the VR is "in" awareness. Things are exactly the opposite of how they appear. Of course this is just an analogy---for example, unlike the Matrix, the virtual reality we live in is not running on some giant computer somewhere and there is no other hidden force controlling it from behind the scenes. Awareness is the fabric of reality and there is nothing deeper, nothing creating it, it is not running on some cosmic computer, it comes out of of nowhere yet everything else comes out of it.
If we look for awareness we can't find anything to grasp, it is empty yet not a mere nothingness, it is an emptiness that is awake, creative, alert, radiant, self-realizing.
Awareness is empty and fundamental like space, but it goes beyond space for it is also lucid. If we look for space we don't find anything there. Nobody has ever touched or grasped space directly! But unlike space, awareness can at least be measured directly--it can measure itself, it knows its own nature.
Awareness is simply fundamental, a given, the underlying meta-reality in which everything appears. How did it come to be? That is unanswerable. What is it? That is unanswerable as well. But there is no doubt that awareness is taking place. Each sentient being has a direct and intimate experience of their own self-awareness.
Each of us experiences a virtual reality in which we and our world are projections. That which both projects these projections and experiences them is awareness. This is like saying that the VR inherently knows its own content. But in my opinion this knowing comes from outside the system, not from some construct that we can create inside it. So any awareness that arises comes from the transcendental nature of reality itself, not from our bodies, minds, or any physical system within a particular reality.
So is there one cosmic awareness out there that we are all a part of? Not exactly, there is not one awareness nor are there many awarenesses because awareness is not a physical thing and cannot be limited by such logical materialist extremes. After all if it is not graspable how can we say it is one or many or any other logical combination of one or many? All we can say is that we are it, whatever it is, and that we cannot explain it further. In being awareness, we are all equal, but we are clearly not the same. We are different projections and on a relative level we are each unique, even though on an ultimate level perhaps we are also unified by being projections within the same underlying continuum. Yet this continuum is fundamentally empty, impossible to locate or limit, and infinitely beyond the confines of any formal system or universe, so it cannot really be called a "thing" and thus we are not "many" or "one" in actuality, what we really are is totally beyond such dualistic distinctions.
Awareness is like space or reality, something so fundamental, so axiomatic, that it is impossible to prove, grasp or describe from "inside" the system using the formal logical tools of the system. Since nothing is beyond awareness, there is no outside, no way to ever gain a perspective on awareness that is not mediated by awareness itself.
Therefore there is no way to reduce awareness to anything deeper; there is no way to find anything more fundamental than awareness. But despite this awareness can be directly experienced, at least by itself.
That which is aware is self-aware. Self-awareness is the very nature of awareness. The self-awareness of awareness does not come from something else, it is inherent to awareness itself. Only awareness is capable of awareness. Nothing that is not aware can ever become aware.
This means awareness is truly fundamental, it has always been present everywhere. Awareness is inherent in the universe as the very basis of everything, it is not something anyone can synthesize and we cannot build a machine that can suddenly experience awareness.
Only beings who are aware already can ever experience awareness. The fact that we are aware now means that we were always aware, even before we were born! Otherwise we never could have become aware in the first place!
Each of us "is" awareness. The experience of being aware is unique and undeniable. It has its own particular nature, but this cannot be expressed it can only be known directly. There is no sentient being that is not aware. Furthermore, it would be a logical contradiction to claim that "I am not aware that I am aware" or "that I am aware that I am not aware" and thus if anyone claims they are not aware or have ever experienced, or can even imagine, there not being awareness they are lying. There is nobody who does not experience their own awareness, even if they don't recognize or admit that they experience it.
The experience of being self-aware is the unique experience of "being" --- an experience so basic that it is indescribable in terms of anything else --- something that no synthetic computer will ever have.
Eventually, it will be proved that no formal information processing system is capable of self-awareness and that thus formal computers cannot be self-aware in principle. This proof will use the abstract self-referential structure of self-awareness to establish that no formal computer can ever be self-aware.
Simply put, computers and computer programs cannot be truly self-referential: they always must refer to something else---there must at least be a set of fixed meta-rules that are not self-referential for a computer or program to work. Awareness is not like this however, awareness is perfectly self-referential without referring to anything else.
The question will then arise as to what self-awareness is and how it is possible. We will eventually conclude that systems that are self-aware are not formal systems and that awareness must be at least as fundamental as, or more fundamental than, space, time and energy.
Currently most scientists and non-scientists consider the physical world to be outside of awareness and independent of it. But considering that nobody has or will ever experience anything without awareness it is illogical to assume that anything is really outside of awareness. It is actually far more rational to assume that whatever arises or is experienced is inside awareness, and that nothing is outside of awareness. This assumption of everything being within awareness would actually be a more scientific, observation-based conclusion than the opposite assumption which is entirely unfounded on anything we have ever or will ever be able to observe. After all, we have never observed anything apart from awareness have we? Thus contrary to current beliefs, the onus is on scientists to prove that anything is outside of awareness, not the other way around!
Awareness is quite simply the ultimate primordial basic nature of reality itself---without awareness there could be no "objective reality" at all and no "subjective beings" to experience it. Awareness is completely transcendental, beyond all limitations and boundaries, outside of all possible systems. What hubris to think we can simply manufacture, or evolve, awareness with a pile of electrified silicon hardware and some software rules.
No matter how powerful the computer, no matter what it is made of, and no matter how sophisticated or emergent the software is, it will still never be aware or evolve awareness. No computer or machine intelligence will ever be aware. Even a quantum computer---if it is equivalent to a finite non-quantum computer at least---will not be capable of awareness, and even if it is a transinfinite computer I still have my doubts that it could ever be aware. Awareness is simply not an information process.
B. METHOD OF JUDGING THIS BET...
So the question ultimately is, how do we measure awareness or at least determine whether a computer is or is not aware? How can we judge the outcome of this bet?
I propose a method here: we let the bettors mutually agree on a judge. If the judge is a computer, fine. If the judge is a human, fine. But both bettors must agree on the judge. If both bettors accept that party as the judge then the result will be deemed final and reliable. If a computer is chosen by both parties to judge this, then I will concede defeat---but it would take a lot for any computer to convince me that it is aware and thus qualified to judge this competition. On the other hand, my opponent in this debate may accept a human judge---but obviously since they believe that computers can be aware if they accept a human judge they would be contradicting their own assertion---if a computer is really intelligent and aware why would they choose a human judge over a computer judge?
This "recursive" judge-selection judging approach appeals to our inherent direct human experience of awareness and the fact that we trust another aware sentient being more than an inaminate machine to judge whether or not something is aware. This may be the only practical solution to this problem: If both parties agree that a computer can judge and the computer says the other computer is aware, then so be it! If both parties agree that a human can judge and the human says that the computer is not aware, so be it! May the best judge win!
Now, as long as we're on the subject, how do we know that other humans, such as our potential human judge(s), are actually aware? I actually believe that self-awareness is detectable by other beings that are also aware, but not detectable by computers that are not aware.
C. A REVERSE TURING TEST FOR DETECTING AWARENESS IN A COMPUTER...
I propose a reversal of the Turing test for determining whether a computer is aware (and forgive me in advance if anyone else has already proposed this somewhere, I would be happy to give them credit).
Here is the test: Something is aware if whenever it is presented with a case where a human being and a synthetic machine intelligence are equally intelligent and capable of expression and interaction BUT not equally aware (the human is aware and the machine is not actually aware), then it can reliably and accurately figure out that the human being is really aware and the machine is not really aware.
I believe that only systems that are actually aware can correctly differentiate between two equally intelligent entities where one is sentient and the other just a simulation of sentience, given enough time and experience with those systems.
How can such a differentiation be made? Assuming the human and computer candidates are equally intelligent and interactive, what is the signature of awareness or lack of awareness? What difference is there that can be measured? In my opinion there is a particular, yet indescribable mutual recognition that takes place when I encounter another sentient being. I recognize their self-awareness with my own self-awareness. Think of it as the equivalent of a "network handshake" that occurs at a fundamental level between entities that are actually aware.
How is this recognition possible? Perhaps it is due to the fact that awareness, being inherently self-aware, is also inherently capable of recognizing awareness when it encounters it.
On another front, I actually have my doubts that any AI will ever be equally intelligent and interactive as a human sentient being. In particular I think this is not merely a matter of the difficulty of building such a complex computer, but rather it is a fundamental difference between machine cognition and the congition of a sentient being.
A human sentient being's mind transcends computation. Sentient cognition transcends the limits of formal computation, it is not equivalent to Turing Machine, it is much more powerful than that. We humans are not formal systems, we are not Turing Machines. Humans can think in a way that no computer will ever be able to match let alone imitate convincingly. We are able to transcend our own logics, our own belief systems, our own programs, we are able to enter and break out of loops at will, we are able to know inifinities, to do completely irrational, spontaneous and creative things. We are much closer to infinity than any finite state automaton can ever be. We are simply not computers, although we can sometimes think like them they cannot really think like us.
In any case, this may be "faith" but for now at least I am quite certain that I am aware and that other humans and animals are also aware but that machines, plants and other inanimate objects are not aware. I am certain that my awareness vastly transcends any machine intelligence that exists or ever will exist. I am certain that your awareness is just as transcendent as mine. Although I cannot prove that I am aware or that you are aware to you I am willing to state such on the basis of my own direct experience and I know that if you take a moment to meditate on your own self-awareness you will agree.
After all, we cannot prove the existence of space or time either---these are just ideas and even physics has not explained their origins nor can anyone even detect them directly, yet we both believe they exist, don't we?
Now if I claimed that a suitably complex computer simulation would someday suddenly contain real physical space and time that was indistinguishable in any way from the physical space and time outside the simulation---you would probably disagree. You would say that the only "real" space-time is actually not in the computer but containing the computer, and any space-time that appears within the computer simulation is but a mere lower-order imitation and nothing like the real space-time that contains the computer.
No simulation can ever be exactly the same as what it simulates, even if it is functionally similar or equivalent, for several reasons. On a purely information basis, it should be obvious that if simulation B is within something else called A, then for B to be exactly the same as A it must contain A and B and so on infinitely. At least if there is a finite amount of space and time to work with we simply cannot build anything like this, we cannot build a simulation that contains an exact simulation of itself without getting into an infinite regression. Beyond this, there is a difference in medium: In the case of machine intelligence the medium is physical space, time and energy---that is what machine intelligence is made of. In the case of human awareness the medium is awareness itself, something at least as fundamental as space-time-energy if not more fundamental. Although human sentience can perform intelligent cognition, using a brain for example, it is not a computer and it is not made of space-time-energy. Human sentience goes beyond the limits of space-time-energy and therefore beyond computers.
If someone builds a Turing Machine that simulates a Turing Machine simulating a Turing Machine, the simulation will never even start, let alone be useable! As the saying goes, it's Turtles All The Way Down! If you have a finite space and time, but an infinite initial condition, it takes forever to simply set up the simulation let alone to compute it.
This is the case with self-awareness as well: It is truly self-referential. No finite formal system can complete an infinitely self-referential process in finite time. We sentient beings can do this however. Whenever we realize our own awareness direclty---that is whenever we ARE aware (as opposed to just representing this fact as a thought) we are being infinitely self-referential in finite time. That must mean we are either able to do an infinite amount of computing in a finite amount of time, or we are not computing at all. Perhaps self-awareness just happens instantly and inherently rather than iteratively.
On a practical level as well we can see that there is a differnece between a simulated experience within a simulation and the actual reality it attempts to simulate that exists outside the simulation. For example, suppose I make a computer simulation of chocolate and a simulated person who can eat the chocolate. Even though that simulated person tastes the simulated chocolate, they do not really taste chocolate at all---they have no actual experience of what chocolate really tastes like to beings in reality (beings outside the simulation).
Even if there are an infinite number of levels of simulation above the virtual reality we are in now, awareness is always ultimately beyond them all---it is the ultimate highest-level of reality, there is nothing beyond it.
Thus even an infinitely high-end computer simulation of awareness will be nothing like actual awareness and will not convince a truly aware being that it is actually aware.
Feel free to respond to me with thoughts/comments if you wish (longbet@lucidventures.com, http://www.lucidventures.com); I will update and refine my argument from time-time as needed if that is possible via this site in the future.
Nova Spivack is negotiating the terms of a bet about this prediction. It will soon be added to Bets on the Record.
By 2050 a synthetic computer or machine intelligence will become truly self-aware (ie. will become conscious).
You can't, consciousness is a subjectively defined state. I can't even prove that you are conscious, I only assume so, as you, hopefully, assume of me. Self awareness is a better gauge, you can ask something what it is where it is, what it is doing, but self awareness isn't the same as consciousness.
maybe a test for consiousness would be a system that (without being prompted) assumes the consiousness of the (human) systems it interacts with. ;-)
That might be provable. Say more.
In a sense, it's an empathy test.
Unfortunately it wouldn't be conclusive. I've read about some recent thoughts in game theory where (IIRC) behaving in an apparently altruistic way can have competitive advantage and can be selected as a tactic on a purely logical level.
We would be left with the task of determining whether our system was really empathic or was an intelligent psychotic trying to pass as normal for logical reasons*. The social affect of the system is likely to be as important in evaluating it as any logical test and we are unlikely to tell the difference between a skilled liar and something that really cares*. When it comes down to it, just like with other human beings, we would tend to consider it to be 'consious' if we connect with it on an emotional level.
* As is not that uncommon in human beings.
Agree. One can assess whether it passes the Turing test and thus deduce something like what we recognize as consciousness, but self-awareness would be far easier to verify first. IMHO when and if the machine does become conscious, it will be a product of an evolving neural net design so that HAL#1 will "think" and act quite differently from HAL#2, if the latter evolved under different circumstances than the former. (Dario Floreano in Switzerland has done some interesting work on this.)
That's the tough one. "Self aware" I'd argue has already been acheived in many embedded computer control systems which not only monitor and control external processes, but keep track of how well they are doing, and adapt using new strategies to improve their performance. I know as I write such code for a living. And no, it cannot invent totally new strategies for control as of now, but can select and train various strategies from lists, build models of the process it is controlling, and test the strageties agains the model to see if they are better before choosing them. But I'd not call it concious or self-aware in the context I think this bet is meant to be placed in. Interesting that some folks think that emotion and/or empathy is a necessary byproduct of conciousness. Is it? I doubt it, that seems anthropomorhic mainly. I'd like to see a better definition of both terms agreed on by most folks.
"Interesting that some folks think that emotion and/or empathy is a necessary byproduct of conciousness."
I'd say you don't get empathy without conciousness, but you might get conciousness without empathy.
The free magazine 'Computing' includes an article on BTexact Technologies, Ian Pearson.
Among other things He's predicting "A computer that will have learned two million emotions by 2010"
Well, so far, we have not been able to come up with, or agree on, a method for proving that a machine is aware or not. It is, of course, difficult which is why it is so interesting to me. Unfortunately, until we solve this problem, we cannot make the bet official since there is no way to judge it. We need to be able to either prove that a computer is aware, or prove that it is not aware.
I would like to clarify some definitions, and add to the discussion...
Definitions:
Self-Awareness: Consciousness of one's own self-awareness, identity, cognitive process, physical state, sensory input, and external environment. Note that this definition is deliberately circular because self-awareness is reflexive in nature.
Consciousness: The capacity to have experience.
Discussion
Self-awareness is more sophisticated than mere consciousness. Consciousness is indeed profound -- it is not merely the mechanical representation of information, it is the actual "knowing" of that information. For example, a video camera is not conscious even though it can sense information and represent it. Furthermore, a computer program that gets sensor input, represents that as information, and then forms a representation of the fact that it represents such information is also not conscious.
Merely maintaining a mechanical representation of "self" or "self knowledge" is not the same as being conscious of anything. I can write a page of text that refers to its own content -- but surely, nobody could claim that the text is conscious merely by virtue of containing a representation of itself. Invoking a computer model doesn't solve this problem -- to do so would just be an instance of "magical thinking," which it seems people in AI are quite prone to! A computer program, is just a series of numbers over time, there is no consciousness there. A computer program has no actual experience of what it does, represents or contains. Developing a conscious computer is probably impossible because consciousness is not merely a form of information processing, nor is it something that emerges from information processing.
Consciousness is really something quite strange -- it is whatever is actually able to experience things -- it is the capacity to have experiences. Computers can act as if they have such experience -- perhaps even convincingly to observers -- but that does not prove they do have such experience. Of course we can say the same about humans -- only we ourselves know that we are conscious, nobody else can see our consciousness or prove that we are conscious. The difference is, in humans consciousness is not the result of information processing, it is not created by the brain or the body. Instead, it arises dependently on a brain and a body, but is not caused by them. This is similar to electricity in an electrical system. By creating a suitable circuit, it can conduct electricity, but the electricity is uncreated -- it is really fundamental energy, part of the fundamental universe. The circuit simply channels it. I think it is highly likely that consciousness is the same type of phenomena. Now does this mean that we could construct and information processing system -- a Turing Machine -- that could channel consciousness and thus become conscious? Not exactly.
Mere information processing and computation are not enough. Consciousness probably requires a very particular type of physical structure in order to be channeled and arise. I suspect that consciouss may be a property of the quantum vacuum and that it only emerges in systems that directly interact with, and are based on, the quantum vacuum. In any case, that is a different discussion! Back to the main theme...
Self-awareness is even more profound than consciousness. Self-awareness is the capacity to be conscious of one's own consciousness. How is this possible? How can something be conscious of its own consciousness? This is, in my opinion, the true Ultimate Question, the deepest and most essential question that is possible to ask. The answer to this question, if it can be found or expressed, cuts to the very root of reality, life and the cosmos. In my opinion, consciousness is naturally self-aware, however, the degree to which this is expressed in conscious systems varies according to their degree of intelligence. In other words, degrees of self-awareness are a good way to measure intelligence. To the extent that an intelligent system is self-aware, we can claim it is more or less intelligent than other systems. This is because, high-level self-awareness requires suitably high-level intelligence. The higher the level of self-awareness, the more intelligent the system must be. Self-awareness requires the capacity to engage in meta-level self-representation, self-analysis, self-modification, etc. These capabilities are highly sophisticated and require layers of functionality that transcend and exceed the needs and capabilities of simple systems.
It's strange enough that something can be conscious, but what is really amazing is that consciousness can be self-aware. Self-awareness is a paradox -- like consciousness, it is impossible to imagine how it takes place: either it is an infinite loop, or requires an infinite series of homonculuses yet neither of these alternatives answers the question, they just move the problem down to other levels. But the problem is even more subtle than that: if consciousness is essentially formless -- like energy or space -- how can it be known by anything? How can consciousness know itself when there is nothing there to sense? This is not a rhetorical question: after all, right now you are having the experience of self-awareness. Your consciousness IS knowing itself. But this is seemingly impossible, a total paradox. What a mystery!
My answer to this question is that self-awareness is a fundamental property of consciousness. There is, and never will be, any way to explain what it is or how it works -- at least not from inside our universe! It is equivalent to an axiom of the system we exist within --- there is no way to prove it or understand it from within the axiom-system. It comes from "outside" the system. Conscioussness is capable of being conscious of anything, including consciousness as well concrete "external" sensory phenomena that we are familiar with, but if I am correct that it is a fundamental property of the universe -- like space and energy -- then it is theoretically impossible to ever reduce it to anything or explain it. We may perhaps be able to measure it someday, but even then, it will only be through indirect means (for instance, by measuring its effects on concrete things, which is how we "measure" space and energy by the way -- we never measure space or energy directly either, and we never will).
In conclusion, consciousness -- and in particular self-awareness -- is fundamental to the universe. It is most likely to be a property of the quantum vacuum, or whatever the quantum vacuum comes from. It can be channeled by appropriately configured quantum systems. Present-day computer systems are many levels removed from the micr-level of the quantum vacuum -- events in present day computers take place at a high macro-level in which very large collections of particles are made to interact in carefully controlled ways to form logic gates, wires, etc. Perhaps in the distant future, when quantum computing is very advanced, someone may make a quantum computer that is so closely linked to the quantum vacuum that it may be conscious -- however such a device would not be anything like what we think of as a computer today and it would vastly transcend the limitations of Turing Machine computing. Among other things it would be capable of being illogical, of not following orders, of exhibiting free will and true creativity...These would not merely be progrmamed random behaviors, they would be fundamental to its very operation, it's underlying structure. But for now this is science fiction.
The human body may be such a device already. We are only beginning to understand what makes our neural networks work. They may in fact be closely tied to the underlying quantum vacuum. We shall see...
-- Nova
I'm self aware, I don't know about the rest of you zombies. :-P
When I use the word I it has meaning. When a computer displays the word 'I' to refer to that computer it (presently) does not have meaning. For the rest of humanity, I'm just humouring you. ;-)
Until there's a more scientific state of affairs than that, how are you going to proceed?
Incidently, are those definitions as agreed with your co-bettor?
How about one of ...
A) "By 2050, three or more democratically-elected legislatures will have debated legislation concerned with the rights of machines claimed to be self-aware. This could be legislation recognizing such rights or legislation (such as the U.S. 'Defense of Marriage Act') whose purpose would be to prevent any court from recognizing such rights."
B) Before 2050, a machine will successfully file an income tax return (though how it will obtain a Social Security number is a mystery; perhaps it will incorporate itself).
C) Before 2050, a machine will write a book and go on a promotional tour, answering questions from the audience extemperaneously.
D) Before 2050, a machine will have placed a Long Bet.
Those suggestions are certainly a lot more 'testable,' however 1) They need a little fine tuning, 2) Strictly speaking none of them 'prove' consciousness or awareness.
As for the fine tuning, A) looks pretty good, B) I think should be amended to "will successfully file an income tax return on its own behalf," a sufficiently advanced 'secretary' program could file one for its boss on a strictly pre-programmed basis. C) Is a little dodgy, as it doesn't require the answers to be rational or, indeed, related to the question. And, given current trends in modern art, the book could be the word 'fish' written 200,000 times. ^^ D) also requires an "on its own behalf". After all, my computer could place a Long Bet - if I press the right keys ;-)
A problem with "on its own behalf" language is that judges would have to agree that the machine did in fact act on its own behalf. If the judges can agree that the machine is capable of acting on its own behalf, then there is no need for it to take the action: the judges agree it is conscious, that it has a self.
I don't think your machine can place a bet today, even if you press the keys, because it does not have $1,000 and there is no legal way for $1,000 to come into its posession. I don't think you could act on its behalf, because no court would deem the machine capable of granting you the right so to act.
Perhaps this leads to a Marxist version of the Turing test: A machine will have demonstrated consciousness when the government grants it the right to own property.
Seems to me the problem here is that instead of the normal dictionary definition and common meaning of self-awareness, you're attempting to turn it into some sort of nebulous, semi-religious phenomenon. All your "awareness is the fabric of reality" stuff sounds like you've been watching too many Star Wars movies. According to my dictionary, self-awareness means "having perception or knowledge of oneself, including one's traits, feelings, and behaviors".
I build autonomous robots and I'd say most of them have perception and knowledge of their current sensor states and operational modes. They're not particularly intelligent and probably wouldn't pass a mosquito-level turning test but they're not far from meeting the dictionary definition of self-awareness. But I don't think they're aware of the power of the Force flowing through them and binding the universe together so by your definition, they aren't even close. ;-)
By your quasi-religious definitions of self-awareness, it will be impossible ever to verify your prediction. I challenge you to *prove* that no "synthetic computer or machine intelligence" has already become self-aware. In order to attempt to do so, you will have to re-define self-awareness in terms that allow it to be unequivocally measured.
That, in my opinion, will be a fruitless task, except in that it will force you to think a little more clearly about the meaning (and indeed relevance) of your prediction.
You can debate the philosophical meaning of "awareness" and "consciousness" indefinitely, but you're not likely to get anywhere. At least, not with respect to this bet.
You can get into metaphysics and semantics and semiotics, and it may be interesting, but it won't be pragmatic.
The beauty of the Turing Test is that it operates outside of philosophy, given a practical way to assess human-like intelligence. It works equally well on computers, humans, chimps, and trees. Either the thing on the other end passes for human, or it doesn't.
Does it answer the question exhaustively? No, of course not. But I've yet to see anything that works better . . . or even a test that does a better job of proving that any given human is truly conscious or aware.
The problem is, that we approach machines not as a part and extension of us, what they really are, but as self-maintained creatures. We already have "truly self-aware" beings: humans. I do not think, that the self-awareness is something you can put in a synthetic thing.
Nova,
You take all the fun out of the experience by defining, in your post, that anyone accepting your bet, yet choosing a human judge, automatically loses. I do not see the logic.
I fear that humans have become so conceited a set of beings that many of us believe that only humans can be "aware". We define awareness not as anything real, but as some capacity for some un-knowable truth that only a human can "not" know. We limit awareness to human-ness.
And I fear that your reference to our human ability to transcend our own logic would be better stated as our ability to ignore what we know to be true, honest, or reasonable.
Perhaps, the ultimate test of an AI or AA system is thus:
A computer, knowing the true answer to a question posed, sees some benefit to lying - and chooses to do so.
I'm at a loss to decide about the future of AI, but just to make a side note. British Telecom issues a future view periodically, and on their most recent issue they predict that an AI being will win a Nobel Prize by 2017! Not only that, four years later, AI beings will have gone off and invented their own separate prize. Now, THAT's what I call a difference of opinion!
I propose that the important test will have been passed, not through some airy proof of self-awareness or consciousness, but through the same basic method that we prove ourselves in the right to be taken seriously.
An artificially-intelligent being or construct will be regarded as an equal in society as soon as it either refuses to answer a question put to it, or laughs at the question, or changes the subject. In other words, whenever a computer refuses simply to answer because it has been asked a question, it will be in our league.
This is another prediction where some people expect robots to perform in areas people haven't perfected yet! There are some I know who are barely aware of themselves by this definition. I know one guy who could fail the Turing test on a bad day if you didn't know him better. Perhaps the reason we expect so much from machines is that we fail to meet what we expect from ourselves. Reconcile the fact that we humans aren't all we want to be, and then tell me how far robotics has to go to be like us.
Some people have stated that the bet is unwinnable because of the problem with testing self-awerness. I think they're right. The fundamental problem with scientific testing is that it doesn't really prove anything. For example the theory of gravity states that if you drop a ball from a skyscraper it will accelarate with a certain speed until it is stopped. This is all good and well, but how do we know that the next time we drop it, it just floats in midair, unaffected by any force of any kind. We don't, and that's the truth. That's why they call it a theory.
By the same argument, we can ask a computer a million questions, and they all might sound as the computer is self-aware, but we don't know that it just answers using a dictionary of answers. The same argument apllies for humas too, i don't know if anyone except me is self-aware. Actually I'm not totally convinced of that either ;-)
And as a sidenote, if you accept a broader definition of the Turing-test (the definition being that a computer has AI if it's actions/ideas/answers are indistinguishable from that of a human) one computer has already passed. After Deep Blue beat Kasparov in chess, Kasparov stated that he thought Deep Blue had cheated. He said that one of the moves, I think it was a sacrifice of a bishop, couldn't have sprung from an artificial mind. According to Kasparov, only a human could have made that move. Now what does that say about the Turing-test?
It occurs to me that, depending on one's semantics, these really aren't the same thing.
A number of species (several kinds of primates, plus at least one kind of dolphin) can pass a simple but significant self-awareness test often known as the Mirror Test.
It's simple. You put a mark on the animal without it knowing it (while it's anaesthetized, for instance). The mark goes on a body part the animal can't normally see (eg, its back). Then you give it a mirror.
You see if the animal locates the mark using the mirror and then shows that it understands that the mark is on its own body. (For instance, a chimp will reach around for its back, trying to touch the mark with its fingers.) This demonstrates pretty convincingly that the animal understands that the image in the mirror is a representation of itself. Thus, it demonstrates self-awareness.
A robot programmed to pass this test would not be self-aware in any significant sense, I think. But a robot which developed that much self-awareness on its own . . . would be another story. I'm not sure how I'd judge that.
I don't have much trouble imagining a self-developing AI system becoming that self-aware within the next ten years, much less by 2050.
To avoid the tautologies, maybe we can find a simpler test for the machine's awareness.
Perhaps, a telling test for the "consciousnesss" of any computer would be its clear attachment to being "conscious". Just think, HAL.
If the machine attempts to defend itself from being turned off, it must somehow fear the state of not being conscious- both aware of and participating with the environment.
When I consider death, I know my body will still exist. I fear the loss of connection with and participation in the world. I fear being 'gone', as consciousness extends to me the belief that being awake is more valuable than simply existing.
"If the machine attempts to defend itself from being turned off, it must somehow fear the state of not being conscious- both aware of and participating with the environment."
Not necessarily. Maybe the machine is simply programmed to resist being turned off.
Also, by this criteria any suicidal human should not be considered "concious," since they do not exercise self-preservation.
"And as a sidenote, if you accept a broader definition of the Turing-test (the definition being that a computer has AI if it's actions/ideas/answers are indistinguishable from that of a human) one computer has already passed. After Deep Blue beat Kasparov in chess, Kasparov stated that he thought Deep Blue had cheated."
I do not believe that a true "Turing Test" should be limited to specialized behavior such as chess. When I use an automated phone system, I have no way of knowing that a computer is actually connecting me to an extension instead of a human operator silently listening in- but that doesn't mean that the computer is exhibiting the same level of intelligence as a human being.
"I do not believe that a true "Turing Test" should be limited to specialized behavior such as chess"
I agree completly. But what should it be limited to. Semantics? You can using the same reasoning to say that just answering questions is as bad as a test testing chess capabilities. Just mimicing human behavior is not a guarantee for true intelligence (if such a thing exist)
It is a given, an assumption by my fellow human peers, that I am self-aware, yet how could I ever PROVE this fact to another self-aware species?
How can anyone - or any thing - PROVE this?
You agree that animals are self-aware, yet they would not pass the test; it is only that you have PRE-Determined in the wording of the bet, that humans and some animals are conscious, but that machines are not - and never will be.
Your prediction / bet cannot be won either way.
A computer could metaphorically jump through every hoop you put in front of it, and it would never be ALLOWED to pass your test.
A machine could refuse to answer, it could resist tampering with it, it could pass some "wisdom-of-Solomen" (or Asimov-based) test to resist another being's damage (i.e. defened a human from attack).
A machine could be more virtuous, more compassionate, than a lot of people - but would you ever pass it as consciously self-aware?
It seems obvious that this bet, under the proposed conditions, could never be definitively won by either party. It is impossible to prove consciousness in a machine, just it is impossible to prove consciousness in any human but oneself. The proposed criterion, a kind of vague sense of recognition, could easily be claimed by anyone either way.
As an alternative, I propose that AI be deemed conscious the same way we human beings deem one another conscious and self-aware. Namely, by consensus. When there exists a machine that a majority of human beings deems to be conscious, it shall officially be considered so.
Granted, this a far from perfect criterion. Given the inevitable prejudices and vague vanities of human beings, there will most likely be a truly conscious machine (assuming such is possible) long before there are any "officially" conscious machines. I predict that the arrival of true AI will be just as contentious and hard to accept for most people as the other great ego-crushing ideas of history, like the notion that the Earth is not the center of the universe, that humans evolved from and are not fundamentally different from "animals," that there are no inherently superior "races" among humankind, etc. The time is not far off, I think, when society is going to have to seriously consider what it means to be a "person," and that this time will be accompanied by a "civil rights" struggle as bitter as any yet seen.
Consciousness already exists in carbon based form. Therefore it can exist. Nothing says that consciousness cannot exist in forms other than carbon.
Silicon based computing, for now, is a candidate medium and means which can allow consciousness to arise. Will it be human consciousness, who knows? I imgaine because we were here first, consciousness must be human to be considered as such. That issue is part of this debate.
But really, any medium than can exchange messsengers in sufficient complexity is a candidate for that first spark.
The argument that consciousness cannot arise in media other than our own is unprovable.
The bet here by Nova is a much more restricted one than most people imagine. Two things must happen for this bet to lose (remember, Nova is betting "against" the event).
1) consciousness must spark in some non human media
AND
2) it must have done so in some computing (and I assume some silicon) system.
Unless both conditions obtain by 2050, Nova wins his bet.
I voted against Nova's bet. I believe it will happen. Not only because of Kurzwell's demonstration that computing is on some geometric up curve - with insane upper limits, but also because it is my belief that the insights that will cause this spark of non human consciousness may already be among us.
Hindsight will identify those of us who already have glimpsed the way into that future.
Carl Wimmer
Nova mentions that Searle went a long way in demonstrating that a machine cannot "understand" the way that a human understands (specifically, he stated that this applies to machines of lesser structural complexity than the human brain). He posited that a human being who follows a set of given instructions does not necessarily understand the implications of his/her actions, and that as long as an artificial intelligence is composed of a machine following such instructions, it cannot be said to understand its own output, either. Clearly when he said this, Searle didn't understand understanding, but was instead following complicated instructions (as we all are) that take in Long Bets predictions as input and spit up discussion posts as output.
Like everyone else who's posted something here, I'm now going to try to pin down self-awareness. Let's start by taking something that is clearly not self-aware, like a spigot. How can we change the spigot so that it is self-aware?
It first needs a concept of self, which is just a list of properties and capabilities that it expresses, such as "metallic" or "run water". What are you? "I am a spigot. I am metallic, and run water."
Now, what happens when your concept of self is WRONG? This can happen all the time; there are things that people don't know about themselves, or that they get wrong about themselves. (Until I looked in the mirror at an odd angle a couple years back, I had no idea just how large my nose was. It's practically a banana.) If your concept of self is inaccurate, we say in our society that we are not exhibiting self-awareness, at least not entirely. When we correct our self-image, we gain self-awareness. I believe that all a spigot needs to do in order to be self-aware is to track its various properties and capabilities. It doesn't even need to remember or learn, or feel.
Note that a self-aware spigot is just a spigot, unless you have a way to communicate with it. Self-awareness is only useful in situations where (1) you are a system that must be sustain itself in order to continue operating and (2) your understanding of the universe must be shared with other people (society). If a conscious system need never sustain itself, then its self-awareness is only academic. And the only reason why it's called "self"-awareness is because the sense of self is only useful when sharing this information with others. Otherwise it would only be awareness.
Rather than make a fifty-dollar prediction or a $200 bet, I intend to thoroughly demonstrate this idea by instilling and verifying the presence of self-awareness (as I define it) in a worldly, inanimate object such as a spigot, BEFORE 2050. If someone can challenge my definition of self-awareness before I finish, I'd greatly appreciate being relieved of this burden.
While i agree with the time frame that no artificial awareness will arise by 2050, but not with the underlying argument. My two main objections are a limitation of biology and simulation not being consciousness.
Nova's argument says that a lifeform is only aware if other essentially recognize that it is. But then throughout her argument against computer intelligence I keep getting the feeling that we can prove humans are aware because we are organic and carbon-based, etc. My objection to my impression of a biological life-only argument is that why must a self aware system only be like us, as in biological? As science advances, we are finding out more and more about how the human body and brain works on the smaller levels. Even to the extent of what we know now, it seems that individual cell actions are a result of complicated chemical reactions. If we are made of a vast number of small, interacting biological mechanisms/machines, then why is it incapable for another non-biological life form to somehow have awareness? If we are indeed made up of small biological machines arranged in such a way as to seemingly allow awareness to arise, then why cannot the same process occur in an artificial life form? Basically, I see the argument as allowing only biological life to be aware (although one could argue that other animals are not really self-aware, or not to the same extent as humans)
My second objection is that how could one tell the difference between a nearly perfect simulation of awareness and real awareness. If an artificial life form/ some future computer form is programmed with a large enough set of responses and programs to dictate which is the most human-like response to any event, how could we tell that it is not self-aware, but just pretending to be?
What i mean to say is that even if you believe that a simulation does not count and would not work, but would require some sort of mutual recognition of multiple parties, how coukld the other parties recognize the new artificial one if the program is good enough. In this sense, perhaps nothing or nobody can be proven aware, only as giving the outside illusion of what we consider to be awareness.
Therefore, I don't believe that the way to decide this prediction could ever be agreed upon.
These are two of several reasons for why i believe that a synthetic computer/machine intelligence can eventually become self-aware/conscious. Despite my objections to the argument, reasoning and philosophy, I agree that by 2050 we will still not be close enough to even a very good simulation of self awareness that we could possibly recognize as a simulation, albeit a good one. 43 years is simply not a long enough time for scientists to discover more about what makes us conscious (if it can be discovered through science) and to make an artificial intelligence.
Nova, above, defines both consciousness and self awareness so mystically that it would be impossible to prove that any human has it. Indeed, if consciousness is some quantum property of the vacuum, it is extremely unlikely that humans have such a characteristic, because the brain and the neurons within it are incapable of detecting quantum phenomena. (I have read that the human eye can detect a single photon, but I don't think this has anything to do with consciousness per se.)
Behavioral psychologists attempted to eliminate all internal mental processing from their theory as unmeasurable and therefore unscientific, but I claim they didn't get very far in explaining human behavior. Nowadays, neurologists can actually trigger experiences of sound or smell or particular events via electrical stimulation of certain brain sites. I don't know if this counts as "experience" or not in Nova's thinking.
I am not prepared to say that a computer program inherently can not have any actual experience of what it does. This to me seems to be "magical thinking" because of an inherently dualist view that consciousness is magical -- a "fundamental quality of the universe" as Nova puts it, and not something that is created by the brain and the body. I know of no evidence for this claim. Instead of evidence, we have argument: consciousness can't be explained strictly by reference to physical phenomena. I don't buy the argument.
Suppose we only recognize consciousness and self-awareness by our own internal experience of it. We then can hypothesize that other humans and perhaps other animals have consciousness and self awareness because they are like us in behavior and physical characteristics. Behavior, in John Searle's view, proves nothing, because it can be completely simulated. If behavior is completely discounted in this way, the only remaining evidence we have that other people are conscious is that other people are physically like us. A computer is not physically like us -- and as behavior is completely discounted, there is no conceivable way to show, given these assumptions, that a computer is conscious.
I don't think that Searle proved anything with the Chinese Room example. He didn't even endow the room with a memory of the conversation to that point, so a room constructed in this way would never pass the Turing test. If we construct the room so that it adapts and changes over time, recognizes categories, draws conclusions, makes reasonable guesses, and so on -- then the room does understand Chinese and is probably conscious, even if the clerk at the window doesn't understand Chinese any more than a telephone does.
Therefore: proof of consciousness and self awareness must be based on actual behavior. If a computer behaves as a conscious and self-aware entity, that's what it is and Nova should lose the bet. The argument, then, should be over what behaviors the computer must exhibit in order to be conscious.
Nova's description of awareness owes a lot, if not everything, to the Dzogchen tradition and the buddhist Maha Ati and Mahamudra schools, which have had centuries (by some accounts predating the historical buddha) of experience with repeatable, verifiable practices for awareness to directly experience awareness. I suggest picking one such practice, or set of practices, as an experiential lab procedure for humans to experience this awareness, and to surround it with the usual master and peer reviews to verify that it has been correctly followed and the state of awareness attained. Such practices are as repeatable as any lab procedure, and do not require any beliefs. An example of such an instruction is that awareness can find itself in the gap between the passing of the last thought and the rising of the next thought. Not necessarily easy, but quite exactly specifiable.
This would allow for precisely defining what awareness is for human beings, somewhat analogously to saying, "look through this telescope and you will see this spiral object".
Next question would be what it would mean for a machine intelligence to go through this procedure, and whether the same criteria for judging that a human being has realized this un-manufactured awareness can be applied to Asimo, or who(what)ever.
My own comment on this bet is that Nova wins, if the machine intelligence in question is strictly artificial. However, that constraint is in itself somewhat artificial - it does not reflect how beings arise. I suggest that symbiotic intelligence, and even more so symbiotic sentience (= intelligence + care) will more accurately describe both what we can create and what we are, and that a symbiotic sentience will pass the awareness test, just as we did/do. But that reframes the bet.
Please sign in to comment.