Abstracted from ‘Qualitative experience in machines,’ The Digital Phoenix: How computers are changing philosophy.
1. Many people, perhaps most people, have the idea that, however problematic qualitative experience is for the case of human beings, it is a lot more so for that of machines constructed by human beings. Few philosophers doubt that human beings’ experiences have qualitative characters, but many doubt or disbelieve outright that robots and computers (much less backhoes and can openers) could ever have qualitative experiences at all. Often the latter denial is just evinced, as an “intuition,” though occasionally it has been argued. There are even some philosophers who think that the big problems have been pretty well solved for human beings or can be solved without much further effort, but who also think that machines simply could not be conscious, have qualitative or subjective experiences, etc.; that is the most extreme version of the idea I am considering.
My purpose in this paper is to defend the goose-gander thesis that the disparity here is specious: There is no problem for or objection to qualitative experience in machines that is not equally a quandary for such experience in humans. It is, I contend, mere human chauvinism or at best fallacy to suppose otherwise.
Just for the record, here are the leading problems regarding the phenomenal character of human experience: Leibniz’-Law objections; the immediacy of our access to qualia; essentialistic and other Kripkean (alleged) modal features of qualia; “zombie”- and “absent-qualia”-type puzzle cases; first-person/third-person asymmetries of several kinds and the perspectivalness of the mental; putative funny facts as claimed by Thomas Nagel and Frank Jackson; qualia in the strict sense, the introspectible monadic properties of apparent phenomenal individuals; the grainlessness or homogeneity of qualitative features, emphasized by Sellars; and Joseph Levine’s now celebrated “explanatory gap.” That is an impressive array of difficulties for the materialist (1). It is so impressive, in fact, that it immediately lends support to my goose-gander claim. For if there is a problem about qualitative experience in machines that is not equally an objection to a materialist view of people, that problem must be additional even to the many and wide-ranging ones I have listed. It must also be grounded in some substantive difference between machines and human beings.
2. For present purposes, then, we must mean by “machine” something that contrasts interestingly with “human being.” (In one sense, uncontroversially, human beings are machines.) Let us mean a kind of artifact, an information-processing device manufactured by people in a laboratory or workshop, out of physical materials that have been obtained without mystery or magic. A paradigm case would be a robot driven by a present-day supercomputer. But I want to allow technologically imaginable extensions of that paradigm; a machine need not have von Neumann architecture, or even digital architecture (whatever that means) at all. And let us idealize a bit: I shall assume that problems of information storage and retrieval, such as the notorious frame problem, are solved. (A fairly outrageous assumption, true. The reason I get to make it is that my chauvinist opponents do not think that their objection could be overcome even if the frame problem and its ilk could be; they think their obstacle arises no matter how good our machine might be at mere information storage and retrieval.)
What, then, are the most obvious differences between machines in the foregoing sense and natural-born human beings, that might support the chauvinist position? Let us begin by abstracting away from the most obvious deficiency of actual, 1990s machines: that no such thing has a humanoid behavioral repertoire or anything remotely approaching it, because no present-day machine is anywhere nearly as complex as a human being or gifted with a biologic brain’s almost unthinkably vast information-processing capacity. Here again, my opponents deny that more information processing (per se) would help; no further amount of the same, no matter how large, would convert a mere machine into a sentient creature capable of subjective, qualitative experience.
So let us help ourselves to some futuristic, science-fiction technology, and suppose that such resources have afforded us an expert human simulator. Elsewhere I have introduced a character called Harry(2), who through amazing miniaturization and cosmetic art is an entirely lifelike android. He is also a triumphant success of AI: his range of behavior-in-circumstances is equal to that of a fully acculturated and rather talented late 20th-century American adult. No one would ever guess that he is not an ordinary person. (Let us further suppose that his internal functional organization is very like ours; his total pattern of information flow is parallel to ours, even though it runs on considerably different hardware.)
But our question is, in the relevant sense, is Harry a person at all? He is, remember, only a computer with limbs; his humanoid looks are only a brilliant makeup job. Some philosophers will readily grant that he has beliefs or belief-like states; after all, he stores and deploys information about his environment and about the rest of the world. But desires are a bit harder; hopes, embarrassments and other conative attitudes still harder. Yet even those who would award Harry a full range of propositional attitudes might still balk at qualitative experience. Even if in some sense he thinks, he does not feel in the most immediate sense in which we doso says the chauvinist.
3. Before we go on to look at some further differences between Harry and the rest of us, let us note that there is a heavy presumption in favor of my egalitarian goose-gander claim (3). First, how do we now tell that any familiar humanoid being is conscious? Normatively pursued, this is just the Problem of Other Minds. But we need not take a stand on the best solution to that problem in order to note its origin. The problem begins with the fact that the ordinary person’s evidence for ascribing mental states, including qualitative states, to another human being is the latter’s behavior, broadly construed, in the circumstances, broadly construed. How we justify the epistemic move from that behavior to the mental ascription is a topic of notorious controversy, but unless we succumb to global skepticism about other minds, we do not doubt that the mental ascription is justified by our observation of the behavior. (Of course the justification is defeasible.)
Few readers will have failed to foresee my next move: By hypothesis, Harry is a flawless human simulator and behaves, in any circumstance, just as a human being might. So, over time, he provides his viewers with just the same sorts of behavioral evidence for mental ascriptions that you or I doincluding ascriptions of qualitative experience. So far as we have evidence for ascribing qualitative phenomenal states to each other, we have just as strong prima facie evidence for ascribing them to Harry. And common sense, at least, counts that evidence as very strong, so strong that we rarely even entertain potential defeaters.
Notice further that in the case of human beings, such behavioral evidence does not require assumptions about the subject’s innards (4). We mature and educated people do know that other human beings are biologic organisms and we presume that the others’ biology is like theirs, but the standard tacit behavioral reasoning does not depend on that presumption. 1) A child or naïf who did not know those things would be just as well justified in her/his mental ascriptions, or at least very nearly as well justified, as we. And 2) if we were watching a videotape of humanoid creatures which might be from another planet and might have a biology quite different from ours, then if those beings behaved just like humans, we would still be justified in imputing human mental states to them–indeed, I submit we would not even think about it, unless our philosophical guard were up.
The foregoing points, especially subargument 2), might be thought to beg the question against the chauvinist. But they do not. I have granted (and would insist) that the justification conferred by the behavioral evidence is in every case defeasible. That leaves open the possibility that for machines, or even for aliens, the class of potential defeaters is wider than that which attends mental ascription to human beings, and I have not assumed otherwise. My present point is only that powerful defeat is required in Harry’s case; the chauvinist is already one–a big one–down.
Here is my second argument for the same conclusion. (Science fiction again:) Suppose that Henrietta, a normal human being, requires neurosurgery; indeed her entire CNS is under attack by a virus that will gradually destroy it. The surgeons start replacing it (and if you like, much of the rest of Henrietta) with prostheses. First a few neurons are replaced by tiny electronic devices. These micromachines so sucessfully duplicate the functions of the neurons they replace that Henrietta’s performance is entirely unimpaired. Then a few more neurons are removed and substituted for; complete success again.
And so on until there is no wetware lefteventually, Henrietta’s behavior is controlled entirely by (micro)machinery, yet her intelligence, personality, poetic abilities, etc., and most importantly her perceptual acuity, sensory judgments and phenomenological reports remain just as always. Now, a chauvinist must maintain that at some point during the sequence of operations, Henrietta ceased to have qualitative experiences; she has become cold and dead inside and is now no more sentient than a pocket calculator. One can imagine a particularly boorish chauvinist asserting this to her face. She would protest, of course, and tell him that her inner life is as rich and vivid as ever, describing it as lyrically as time and his rudeness allow. It is hard to imagine how the boor, or any other chauvinist, would be able to draw a line and state with assurance that after the nth operation, Henrietta ceased to have a phenomenology (whatever she may think to the contrary). It is a hard position to defend.
Here again, I do not want to beg the question against the chauvinistor to commit a slippery slope fallacy, either. For there may be a defeater that cuts in at some point and does override the behavioral evidence; and the “point” may be a vague one to boot (5). As before, I am not asserting that no such defeater exists, but only emphasizing that the chauvinist bears the burden of coming up with one and that it is a considerably heavier burden than one might think.
4. What, then, are the defeaters specific to machinekind? I can think of three possibilities. First: There is Harry’s origin. He is an artifact; he was not of woman born, but was cobbled together on a workbench by a group of human beings for purposes of their own. Perhaps a workshop is not a proper mother (imagine Dame Edith Evans enunciating, “A workshop?”).
I do not think any sound chauvinist argument can be based on that difference. For suppose we were to synthesize billions of human cells and stick them together, making a biologic humanoid organism. (We could either make a mature adult straightway or, what is technologically easier, make a fetus and nurture it.) We might further suppose that the resulting pseudo-humanlet us call him Hubertis a molecular duplicate of a preëxisting human being. There is little doubt that such a creature would have qualitative experience; at least, if he did not, that would probably not be simply because of his early history (6). So artifactuality per se seems not to count against having phenomenal states. Our first difference is no defeater (7).
Second: It may be said that Harry is not a living organism. (Paul Ziff made such an appeal in his well-known article, “The Feelings of Robots” .) If something is not an organism at all, properly speaking, then there does seem to be something odd about ascribing sensations and feelings to it.
Much depends on what is considered criterial for “living organism.” We have already failed to find reason to think that artifactuality per se precludes qualitative experience. Parallel reasoning would show that artifactuality per se does not preclude something’s being a living organism either, for surely our synthesized pseudo-human would count as a living organism. Putting artifactuality aside, then, what constitutes living? Automotion? Autonomous growth and regulation of functions? Reproduction or self-replication? Metabolism? Being made of protein?
Whatever. Some of these things–the first three, at least–could be done by a machine, in which case the machine would be “alive” in the relevant sense and the objection’s minor premise goes false. Others, very likely the last two, could not be done by machines; but in that case we should ask pointedly why they should be thought germane to consciousness, qualia and the rest. E.g., why should a thing’s metabolizing or not bear on its psychological faculties in so basic a way as to decide the possibility of qualitative experience? It is hard to see what the one has to do with the other, or to imagine a plausible argument leading from “no metabolism” to “no qualitative experience.” And likewise for being made of protein.
Also, remember Henrietta. She started out as a normal human being but was gradually turned into a machine. Did she go from being a living organism to being non-living, inanimate? In that caseif she had been alive and then ceased to liveshe died, and obsequies are in order. It would be both hard and easy to make her funeral arrangements: Hard, because we would first have to persuade her that she was dead and that services should be held at all; she might resist that suggestion fairly indignantly, especially when we got around to the question of burial vs. cremation. But then easier, because we would not have to guess posthumously at her wisheswe could just ask her what hymns she wanted, whether there should be a eulogy or a general sermon, and so forth, right up till the last minute. I must say I think I would enjoy attending that funeral; I am not so sure that Henrietta would, herself.
1 As is perhaps surprising and certainly far from well enough known, every one of those problems is resolved in my books Consciousness (Cambridge, MA: Bradford Books / MIT Press, 1987) and Consciousness and Experience (Cambridge, MA: Bradford Books / MIT Press, 1996).
Incidentally, in this paper I shall concentrate only on “feels” in the sense of qualia. But for explicit defense of the thesis that machines can have feelings in the sense of emotions, see (e.g.), A. Sloman and M. Croucher, “Why Robots will Have Emotions,” in the Proceedings of the 7th International Joint Conference on Artificial Intelligence (Vancouver, B.C., 1981), and N.H. Frijda, “Emotions in Robots,” in H.L. Roitblat and J.-A. Meyer (eds.), Comparative Approaches to Cognitive Science (Cambridge, MA: Bradford Books / MIT Press, 1995).
2 “Abortion and the Civil Rights of Machines,” in N. Potter and M. Timmons (eds.), Morality and Universality (Dordrecht: D. Reidel, 1985), pp. 141ff.; and the Appendix to Consciousness, loc. cit.
3 The following two arguments are reprised from the Appendix to Consciousness, loc. cit., pp. 125-26. (Hereafter I am going to spare myself typing “loc. cit.” in references to my own works.)
4 This claim is contested by Christopher Hill, in Ch. 9 of Sensations (Cambridge University Press, 1991), and by Andrew Melnyk in “Inference to the Best Explanation and Other Minds,” Australasian Journal of Philosophy 72 (1994): 482-91. What follows in this paragraph is in part a reply to their objections.
5 Ch. 2 of my Consciousness and Experience defends the claim that the notion of conscious awareness is vague and comes in degrees of richness or fullness.
6 My suggestion about molecular twinning is not meant to suggest that qualiaphenomenal propertiesare “narrow” in the sense of supervening upon molecular constitution. In Ch. 6 of Consciousness and Experience I argue that they are “wide” and do not. But there is no reason to think that the external factors needed to determine qualitative character include the circumstances of one’s coming into existence.
7 In fact, I think that discrimination against Harry on the basis of his birthplace and/or his genesis would be almost literally a case of racism.
8 Analysis 19 (1959): 64-68. In reply, see also J.J.C. Smart, “Professor Ziff on Robots,” Analysis 19 (1959): 117-18, and Hilary Putnam, “Robots: Machines or Artificially Created Life?,” Journal of Philosophy 61 (1964): 668-91. Interestingly, I have found that young children uniformly resist the anthropomorphizing of computers on the grounds that computers are not alive.