In light of the very interesting interview with Dave Chalmers in the Opinionator I thought I would revisit some of my objections to the notion of artificial consciousness (AC). I am somewhat of a skeptic about artificial consciousness in a way that I am not about AGI (artificial general intelligence). My suspicion is that intelligence and consciousness are indeed dissociable and so we could end up with artificial systems that were intelligent in the sense of being able to solve problems, but which would not be conscious (nothing that it is like for them to solve problems or ‘think’ about them). In the past I have called the view I am interested in Biologism about consciousness but then I became aware of the problematic history of that word and I have not come up with something better…Biologicalism maybe? Or just go with Searle’s Biological Naturalism? Here I’ll just call it the Biological view. Below is a draft I wrote a while ago exploring the case against the biological view…I still think it is pretty good and so would appreciate any feedback. (by the way, just to be clear I am open to being convinced that the biological view is wrong, it may just be my last common sense prejudice about consciousness refusing to succumb to good arguments, if so what are these good arguments?)

Whatever you call it, the issue is whether consciousness is fundamentally a biological phenomenon or whether it is an ‘organizational invariant,’ (Chalmers 1995; Chalmers 1996), or is ‘substrate independent’ (Bostrom 2003). On one side of the debate are those who affirm that this is so and thus hold that a properly organized computer program could in fact be conscious in just the way that you or I ordinarily are. Because of this I will call this view computationalisma bout consciousness. On the other side of the debate are those that hold that consciousness is a distinctively biological phenomenon and that a maximally specific computer simulation of those biological processes would not result in consciousness.

The debate over computationalism is independent of the debate between the physicalist and non-physicalist. A non-physicalist may hold that a suitably programmed computer would come to instantiate non-physical qualia (as David Chalmers (1996) has suggested). All that would be required is that there be fundamental laws of nature that relate the computational states to non-physical qualia. While a physicalist may hold that the implemented program itself is conscious. Physicalists who endorse computationalism usually do so by endorsing some kind of (computational) functionalism whereas physicalists who endorse biological view usually do so by endorsing some kind of type identity theory.

Though there are those that defend the biological view (Searle 2004, Block 2009) the dominant view is computationalism. This is perhaps in part due to the popularity of functionalist views. However it is very rarely argued for. It is much more common to hear remarks to the effect that ‘neurons aren’t magic’. That is not an argument. It is also not enough to cite arguments for functionalism either. For instance the argument from multiple realizability at most suggests that there may be other biological configurations that may be conscious.  It is quite another thing to suggest that non-biological systems could be conscious. 

Of course one might invoke the conceivability of non-biological creatures that are conscious. But doing so begs the question. It may be the case that when we imagine that there is a silicon system that is conscious, a commander Data type robot, we imagine something that is the equivalent of xyz. Given that we know that water is essentially H2O we know that the xyz is not a candidate to be our world. Our world could not have been the xyz world. If the biological view is true then the same is true for the commander Data world. It is conceivable but it is not a way our world could have been. Given this what we need is an independent argument for computationalism. 

Perhaps the best (only?) argument for computationalism is David Chalmers’ Dancing Qualia and Fading Qualia arguments. Recently Chalmers (2010) has come to change his mind on the strength of the dancing qualia argument. His change of heart was motivated by empirical findings from cognitive neuroscience, and in particular change blindness. Here I will argue that we have further empirical findings that motivate a questioning of the strength of the fading qualia argument. What this shows is that one can reasonably think that the biological view is true. And absent any reasons to think that it isn’t true it should probably be the default view.  Before introducing the empirical results I first turn to looking at the original arguments and the reasons that Chalmers has given for his change of heart on their relative strength. In section three I introduce the empirical results from the partial report experiments. The upshot is that the original dancing and fading qualia arguments look less plausible and so the biological view of consciousness is more than likely true after all. 

II. Flip-Flopping on Dancing and Fading Qualia

These arguments both have roughly the same form. We start with a fully biological creature that is fully conscious. For vividness we can imagine that they are having an intense headache, a migraine say, while watching a movie and eating a box of sour patch kids. We then replace one of their neurons with a functionally identical computer chip. We can even imagine that we are able to do so in such a way that the subject is unaware that it is happening (the wonders of nanotechnology being what there are and all). We then imagine a series of replacements like this, with the second in the line having two neurons replaced, the third 3 neurons, etc. until we reach the other end where we have completely replaced the brain with computer chips. At each stage in the series the subject we end up with is functionally identical to me, or you. Given this there will be no way that the subject’s behavior can change in any way. As the neurons are being replaced the subject continues to complain of the headache, asking for the volume to be tuned down and remarking that the taste of the sour patch kids helps to distract from the migraine, etc. If we assume that computationalism is false then in the fading qualia case we would have to imagine that as we replace the neurons with silicon our conscious experience fades as a light on a dimmer switch would even though our behavior continues to be the same through out. In the dancing qualia case we imagine that we have a switch that activates and deactivates the group of silicon chips that has replaced the neural circuitry that is responsible for a certain conscious experience. As we flip the switch our conscious experience is blinking in and out of existence yet, again, your behavior continues to be the same through out. 

So, if we assume that computationalism is false, then in each case we end up with subjects that are radically out of touch with their own conscious experiences. They say that they have an intense headache but that isn’t true. In one case the headache is dim and in the other it is blinking in and out of existence. In our world we typically do not find ourselves in this kind of position. It seems plausible that we are not radically out of touch with our own conscious experience. So it is much more plausible to think that, while these scenarios are possible in some sense, they do not describe the actual world with its actual laws. In other words, it is safe to assume that in our actual world the complete silicon brain would be conscious in just the way that the biological creature originally was. 

These arguments are not presented as strict reductios but are rather offered to show how strange the biological view is and the consequences of the view. If the biological view is true then we can have systems that are radically out of touch with their own conscious experience and, for all we know, it could be happening right now! Since this seems prima facie implausible we have a prima facie case against the biological view and for computationalism. 

Originally Chalmers argued that the dancing qualia argument was stronger than the fading qualia argument. But in what sense is the one argument supposed to be stronger than the other? In his original paper on this (Chalmers 1995) he seems to suggest that the absent qualia argument is stronger because it has a stronger conclusion. The fading qualia argument, if successful, establishes only that the property of being conscious is an organizational invariant. So if it works it shows that there will be something that it is like for my silicon isomorph, but it does not show that our conscious experiences will be the same. For all the fading qualia argument shows when I and my silicon isomorph are in the same computational state I may be consciously experiencing red while the isomorph experiences blue, or even some color that is completely alien to me. The dancing qualia argument, on the other hand, is supposed to suggest the stronger conclusion that the silicon isomorph is not only conscious but that their experience is exactly the same as my conscious experience (given that we are in computationally identical states). 

In his 1996 book The Conscious Mindhe indicates another sense of strength. In this second sense the dancing qualia argument is thought to lead to an intuitively more bizarre outcome and so is much harder to accept. The subject is having a very vivid conscious experience blink in and out of existence; how could they fail to notice that? He suggests that one might be able to bite the bullet on fading qualia but dancing qualia are just too strange to be real (Chalmers 1996 p 270).

Chalmers has since come to reverse this decision based on considerations of change blindness. Here is what he says in a footnote in his recent book The Character of Consciousness(Chalmers 2010), which I reproduce in its entirety,

I still find this [dancing qualia] hypothesis very odd, but I am now inclined to think it is something less than a reductio. Work on change blindness has gotten us used to the idea that large changes in consciousness can go unnoticed. Admittedly, those changes are made outside of attention, and unnoticed changes in the contents of attention would be much stranger, but it is perhaps not so strange as to be ruled out in all circumstances. Russellian monism…also provides a natural model in which such changes could occur. In The Conscious MindI suggested that this “dancing qualia” argument was somewhat stronger than the “fading qualia” argument given there; I would now reverse that judgment (page 24 note 7)

In what sense, then, are we to take the reversal of strength that Chalmers indicates in the above footnote? It seems implausible that it should be the first notion. It can’t be the case that now the fading qualia argument establishes that computational isomorphs have the same kind of conscious experience as I do. So it must be the case that Chalmers now finds the fading qualia scenario to be more intuitively bizarre than the dancing qualia scenario. 

This, in turn, suggests that it is not such a high cost for those attracted to the biological view to bite this bullet. If this is right then Chalmers will have to back off of the claim that the computational isomorph’s experience is just like mine but he can fall back on the fading qualia argument and insist that the cost is too high to bite the bullet on that argument. If so then consciousness itself –that is, the property of their being something that it is like for the system— may still be an organizational invariant and so a more modest form of computationalism may still be true. 

It is striking that empirical results have such a dramatic effect on our intuitions about strangeness. What can seem intuitively bizarre from the armchair can turn out to empirically be verified. One might think that the mere fact that change blindness shows us how wrong we can be in our intuitive assessment of these kinds of thought experiments would give us pause in endorsing the strength of the fading qualia case. I think that this all by itself is a prima facie reason to doubt the fading qualia argument but even if one resists this somewhat empirically jaded suspicion there are further empirical results that should cause us to re-assess the fading qualia argument as well. 

III. Partial Report and Fading Qualia

In this section I will present a brief sketch of the partial report paradigm and the inattentional inflation results. The interpretation of the results is something that is currently hotly debated in the literature (Kourider 2010, Block 2012, Lau & Rosenthal 2012, Brown 2012, Lau & Brown 2019). The argument I will present does not rely on any one specific interpretation turning out to be correct. As I will show, the argument relies on only the most general interpretation of the experimental results. In fact the interpretation is so general that all of the relevant parties agree on this. 

In these experiments subjects are presented with an array of letters or objects that arranged in some particular fashion (e.g. in a city-block grid, or in clock face circular arrangement). After a brief presentation subjects are asked to freely report as to the identity of the objects they saw. Subjects, overall, cannot report the identities of all of the objects or letters.  However, if a subject is cued, by a tone say, to recall a specific row or item they do very well. The debate has centered on whether subjects consciously experience all of the letters or objects or if they instead represent them in some sparse or generic way. Once side of the debate holds that consciousness is rich and that there is more in our conscious experience than we able to report, while the other side holds that consciousness is sparse and that we experience less than we think we do (at least consciously). This debate, while interesting, does not concern us here. 

All parties to this debate agree that there are these generic representations involved. Experimental results have shown that subjects can correctly report only 3-4 out of 12 objects. Further calculations suggest that at least 10.4 of the letters must be represented in sufficient detail so as to allow identification. This suggests that subjects may have some partial or degraded conscious representations, as that would explain why they are not able to name those letters. Of course it may also be the case that they do consciously experience them but just forget their identities. To test this experimenters have replaced some of the letters in the display with things that are not letters (an upside down ‘r’ or an ‘@’ sign) and subjects failed to report anything abnormal, and when asked, said that there were only letters present. 

So while all parties agree that some of the objects are represented in a partial or indeterminate way, the question has been whether only the 3-4 the subjects get right are represented in full detail (consciously but not accessed) with all of the others being partial or degraded representations or whether instead it is only a few of the representations that are degraded with the majority being represented in full detail. Even those who believe that phenomenal consciousness overflows access are committed to there being at least some degraded or partial representations in these situations. But these subjects believe that they see all of the letters or objects. 

Thus we seem to have ended up showing that normal subjects can in fact have partially degraded experiences and yet be unaware of it. This provides us with motivation to question the fading qualia argument. Now, it is true that in that kind of case we are considering something which is much, much more radical than what is happening in the partial report cases. In that case the subject has some partially degraded experience and thinks it is non-degraded. In the fading qualia case the subject thinks it is having an intense headache even though the actual intensity of the headache consists in just a few bits. But what these kinds of considerations suggest is that it is not such a drastic scenario after al. 

Chalmers does consider one kind of empirical evidence in his original article. He considers the case of someone who denies that they are blind. In that kind of case, he argues, it is plausible that the subject is rationally defective. The have no visual information and yet they believe that they do. Because of this non-standard way that their experience and their beliefs are connected this doesn’t give us any reason to change our minds. However in the case that we have been considering here we do not have this problem. Subjects are not suffering from any kind of irrationality in these paradigms. They are simply asked to look at a bunch of objects that are briefly presented on the screen and then to report on what they had seen. It is true that the objects in question are presented relatively briefly and subjects do not get to look at the stimuli for an extended time, as we would in ordinary life. 

To be sure, this is not as radical as the kind of case imagined in the fading qualia scenario, where one’s conscious experience is severely degraded, but it does suggest that it is not as strange as one might have thought. It is basically a very severe case of what is going on in our everyday conscious lives, exactly parallel to the change blindness case for dancing qualia.  The partial report paradigm is just one of several new empirical findings that seem to vindicate fading qualia. For reasons of space I cannot go into other finding from so-called ‘inattentional inflation’ where subjects overestimate the visibility of stimuli outside of where they are attending.

IV. Conclusion

The fading and dancing qualia thought experiments were never offered as strict reductios of the biological view but were instead aimed at showing that it was unlikely to be true because it entailed some unlikely or implausible consequences about the relationship between our conscious experience and our knowledge of that conscious experience. However, as we have seen, we have good empirical reason to think that these results are a good deal less implausible than they appear to be from the armchair. Of course this doesn’t show that the biological view is true. Rather, it shows that there is not much cost in biting the bullet in both cases and so one can reasonably hold that it is true and admit that there might be cases of dancing and fading qualia. Science has shown us that the world is far stranger than any of us could have ever imagined.