Platonic Functionalism in the Game of Life


If "Strong AI" is true and digital computers can in principle be conscious like humans, there may be some unexpected ontological consequences.

(Background: John Conway is coming to Berkeley to speak on the "Free Will Theorem". I don't expect to agree with him on the details -- I'm pretty much a compatibilist with regard to free will -- but I've always been a fan of his Game of Life, so what better time to rehash a 5 year old post of mine..now in blog form! Adapted from the PSYCHE-D list, October 2000).

Let's assume that robots can be built from conventional microprocessors that are functional replicas of humans, or at least good enough to pass Turing tests regularly. Under functionalism (or "Strong AI"), these robots are conscious the same way you or I are. Let's say these robots have cameras that take pictures of their surroundings, microphones to pick up sounds, and in general have sensors equivalent to those you or I have.

Let's say two such robots are communicating to each other in English over a cable. They could just send each other streams of plain ASCII text, but for now they are talking in a more intimate way: they are streaming spoken English sounds and various images (they either previously stored them, or generate them in real time) directly into each other's sensory registers. We can't tell by a cursory examination that they're engaged in communication, but they are - and we could peek into their registers and confirm this if we wanted to. So far, ex hypothesi, they're two conscious observers talking to each other and showing each other pictures. So there's no difference between what's going on in these robots and what would be going on in you, if you were for instance put into a sensory deprivation tank with earphones and virtual reality goggles connected to some rich outside world.

We're going to put these robots through some transformations, and it's the point of this thought experiment to see if any transformation robs the robots of their "observerhood" or consciousness.

The first one has already happened; shutting off contact with the real world and communicating only with each other.

Next, we play around at gracefully cycling their power on and off. Their program states are saved before power is lost, and reloaded when power is regained. They do not notice that this is happening.

Next, at each power cycling, we replace one of their conventional microprocessors with a "Game of Life" processor (GLP). I'm assuming familiarity with this cellular automaton; more information can be found on the web (see wikipedia or mathworld for details). A GLP is a hardware implementation of Life set up as a universal Turing machine (say, a G5 processor). (It has been proved that this can be done in principle). A real GLP as complicated as a modern processor, with cells a micron wide, would probably be the size of a large city, but as usual with these thought experiments we will gloss over the impracticalities. The GLPs are put in one by one, and kept synchronized with the conventional chips and each other.

Next, we replace *all* conventional chips and wiring of each robot with one giant GLP. Leads connecting chips have been replaced with glider streams, etc. They're still connected with the high-speed cable, though. But in the next transformation, the cable is also replaced and both robots are embedded on a single GLP (probably several astronomical units across).

Taking a step back now: we now have one GLP running forward in synchronous time steps, but it's isomorphic to the original setup of two free-standing robots. We still can't tell by looking from the outside, but if we examined the proper registers (glider streams) we could, with a lot of work, eavesdrop on the robots' conversation.

In the next transformation, the GLP is replaced by a giant grid of pennies, face-up for "alive" and face-down for "dead". The pennies are manipulated by hand (by someone with a lot of free time) to run the generations forward.

Next, instead of flipping pennies over to run it forward, new pennies are simply placed on top of the old ones. This is done for several trillion generations, enough for, say, 10 minutes of conversation by the robots.

Now we have one big stack of pennies. It seems as if we could now point to each layer in succession, saying 'this one is the current step', 'now this one is the current step', and so on up, thereby making the robots relive their experience over and over. But this pointing seems to be a "pointless ceremony": the robots are embedded in the structure, the same way we could be seen to be embedded in four-dimensional spacetime. So subjectively the robots cannot tell the difference; that ten-minute stretch exists to them in the same way it existed previously, and still what it's like to be them is essentially the same as what it's like to be you.

At this point, do we need to keep the stack of pennies intact? It seems we really don't have to; after all, the individual pennies are not interacting with each other in any way. So let's melt them all down (actually, its great mass would probably have done this for us long ago). Now we have a molten copper sphere in space. Something is different now - we can't peek in and check that the robots are still conversing happily - but we don't need to, since we're not running any programs anymore. And this fact shouldn't affect the robots, anyway. We may even destroy the sphere (throw it into a black hole, or let it evaporate away in the universe's heat death).

Note, that the same final product (a molten copper sphere in space) would result from any GLP implementation of any self-aware being put through the same steps. So it would appear that functionalism entails a kind of vacuity, namely, that all possible experiences of all possible simulatable experiencers exist timelessly without need for a particular (or any?) physical embodiment. What "mattered" from the beginning was not the matter the robots were made of, but the succession of states. If you can get these states in a digital computer, you can get them in the game of life, and then on a grid of pennies, between levels of a stack of pennies, and in nothing at all - since what are the pennies doing but acting as a visual reminder of what the states were?

The totality of Games of Life can be numbered. The Vast majority of these do not involve conscious robot observers talking back and forth, but a Vanishingly small number of them do. These numbers contain patterns that can be interpreted from the Intentional Stance, just like you or I can. Are these numbers also conscious?
--
For more along these lines, see the everything-list, and the writings of Max Tegmark, Bruno Marchal, and Jürgen Schmidhuber...

Posted: Wed - March 2, 2005 at 03:41 PM | | | |

©