So in a recent post I pointed out how unreasonable it was to assume that aliens advanced enough to transfer their consciousness into computers would have motives or behaviors anything like what the current human species does. Of course there is an implicit assumption here that a simulation of our brain process on a computer would be just as conscious as we are (the strong AI hypothesis). Here I argue that this isn’t really true. Of course I don’t doubt that artificial conscious beings can be constructed. There is nothing magical about conception, if we manufactured nerve cells in the lab and put them together in a brain it wouldn’t be any less conscious than you or I. However, this doesn’t mean that the particular means by which our brain performs it’s calculation is irrelevant to consciousness. As I shall argue here we actually have pretty good reason to believe that simply simulating what the brain does on a microchip as we know them ¹ is unlikely to give rise to any experiences no matter how faithfully it might reproduce the behavior of that brain². This is a pretty long post so I continue below the break.

First though let me reiterate the fact that there is a genuine scientific problem of consciousness (aka experience). Unfortunately attempts to ’save’ spiritual beliefs about souls by pushing this notion into the gap left between experience and physics has encouraged most hard headed scientist types to dismiss all talk of experiences as unfounded mystical crap. However, there is nothing contradictory about believing that experiences are real (I’m directly observing this right now), governed by a scientific law just like charge or mass but not (ontologically³) reducible to mere facts about position (or wavefunction) of particles and the like. In other words it’s true that what goes on in the brain completely determines what we experience but we have to go out and do science to figure out how. I can figure out that 2+2=4 without doing experiments but figuring out what configurations of particles feel pain requires I go out and see how things work in the real world⁴.

Hopefully continued neuroscience research will eventually yield a theory telling us that if a system of particles has n units of property A and m units of property B then it will be experiencing a mixture of 25% pain and 75% boredom⁵. Even though we don’t yet know what form this theory will take the fact we are trying to explain a fundamental natural property of the universe gives us some idea of what form we should expect the theory to take (Occam’s razor applies). For instance even though back in the 16th century Gallileo couldn’t possibly dream of the form quantum mechanics would eventually take he would have been justified in assuming that the ultimate physical laws wouldn’t mention St. Bernards, i.e., it would be very odd if the TOE was of the form particles behave like X unless they are part of a St. Bernard in which case they do Y.

Therefore since feeling like something is a fundamental natural property⁶ we can assume that there will be some simple property of a system which determines whether it gives rise to an experience or not. So far no problem, we just need to find some simple property that all the things we take to have experiences (people, dogs, etc..) satisfy when having experiences but things we assume aren’t experiencing (tables, anesthetized people) don’t. The difficulty is that such a property must allow a sensical evolutionary explanation of why we are conscious and experience the world in a unified, understandable fashion.

After all it could have been (if natural law was different) that our experiential lives weren’t hooked up to our behavior/environment in any coherent fashion. The efficient design of a visual system might have been such that seeing a particular sort of diagonal line created the experience of being in a small igloo while a slightly different line made one experience being burned alive. Moreover it could have been that rather than what seems a unified conscious experience each person would give rise to a myriad unintegrated experiences some occurring at a very low level (an experience arising from the module that does first pass filtering on our vision) and others at a very high level. Short of assuming miraculous good luck (evolution just happened to hit on the design that gave us unified, coherent experiences) this means that whatever simple physical property predicts the existence of experience must be (mostly) coextensive with efficient implementation of animal executive functions (so for example it doesn’t occur in my car radio).

This leaves two options for a plausible theory of experience. Either there is some simple computational property that characterizes the processing done in (higher?) animals and any device implementing that computation has experiences or there is some simple kind of (uncommon) physical interaction that is particularly well suited to implementing animal executive function (so evolution will likely select for it in just this case). I call the first theory the “magic algorithm” theory of consciousness. Presumably there is some magical couple of pages of lisp code so that only those objects implementing this code have experiences. I find this sort of theory pretty absurd on it’s face (not to mention the empirical evidence a century of AI failure gives us) but maybe it could be swallowed if this was the only problem. The bigger problem with the magic algorithm theory is that it asks us to accept that there is a well defined notion of implementing a particular computation. Also it seems apparent that any plausible characterization will require realism about causation, i.e, in order to distinguish genuine implementation of an algorithm from happenstantial agreement we must distinguish real causation from mere accidental constant conjunction. So not only does it seem doubtful that there is a well defined notion of implementing a computation but even if there was such a definition it would require metaphysical extravagance to work and is thus disfavored.

In summary this leaves us with the conclusion that there is probably some special sort of physical interaction/state (simply defined in terms of fundamental physical laws) particularly well suited to animal executive function and it is the existence of this physical interaction which gives rise to experience. Therefore merely simulating this process on a general purpose processing unit would not produce experiences. Note I’m not arguing that there is new fundamental physics in the brain like Penrose and others would have us believe. Certainly that is one thing that would fit my criteria but there is no reason that the physical correlate of consciousness couldn’t simply defined in terms of already identified types of basic physical interaction.

Then again on some days this argument just convinces me that induction is a load of crap and the only reason our physical theories have been so successful is that we are very well evolved to predict physical facts about our environment. If we give up on the idea that experience must obey simple natural laws then everything is up in the air (literally everything about the world, as we would no longer have a good reason to believe rotting tomatoes weren’t the most likely physical substrate of our experienced lives).