The paper addresses the question of whether a computational process could support consciousness. Muhlestein begins by briefly recounting some thought-experiments proposed by Maudlin and others. Suppose we run a computation which instantiates consciousness: then we run the same process but remove all the unused conditional branches, so that now the course of the process is fixed. In the second case the computer goes through the same set of states (does it, actually?), but we’d be disinclined to think it could be conscious; it’s just a replay. In fact, it’s hard to see why ‘replaying’ it makes any difference since the relevant states all exist in the record, so that inert record itself probably ought to be conscious. Worse than that, since we could, given a sufficiently weird and arbitrary decoding, read random patterns in rocks as recording the same states, an indefinite number of copies of that same consciousness would be occurring constantly pretty much everywhere.
That’s more or less how the argument runs, anyway, and what’s proposed to fix the problem is the idea that for consciousness to occur, the computation must have the property of counterfactual sensitivity. It must, in other words, have had the capacity to go another way. Without that property, no consciousness. The notion has a certain intuitive plausibility, I think: we can accept that in order for me to have the experience of seeing something, the fact that it was a red square and not a blue circle must be relevant; and that consciousness perhaps must be part of a stream which is, in some unobjectionable sense, free-flowing.
Muhlestein proposes a new thought experiment with a truly formidable apparatus. He set up a physical implementation of Conway’s Game of Life and uses that in turn to implement a computer on which his processes can be run. Because his implementation of Life uses cellular automata which display and detect states using light, he can now intervene anywhere he likes by simply shining an external light into the process.
If that last paragraph is unintelligible, don’t worry too much: all you need to know for the purposes of the argument is that we have a computer set up so that we have a special power to intervene arbitrarily from outside and constrain or alter the process which is running as it progresses. Muhlestein now takes a conscious computational process (in fact he proposes to scan one on a whole-brain basis from his friend Woody – don’t try this at home, kids!) and runs it on his set-up; but here’s the cunning part: he uses his light controls not to interfere, but to project the same process back onto itself. In effect, he’s over-determining the process: it runs normally by itself, but his lights are enforcing a recorded version of the same process at the same time.
Now, the computation is taken to be conscious when running normally. It runs in exactly the same way when the lights are on; it simply loses the counterfactual sensitivity: it could no longer have gone another way. But why would extra lights on the process deprive it of consciousness? The outputs will be exactly the same, any behaviour of the conscious entity will be the same, and so on. Nothing about the fluidity or coherence of the process changes. Yet if we say it remains conscious, we have to give up the idea that counterfactual sensitivity makes a difference, and then we’re back in difficulty.
What do we say to that? Muhlestein ultimately concludes that there is no satisfactory way out short of concluding that in fact the assumptions are wrong and that computational processes are not sufficient for consciousness.
Old Peter (the version of myself that existed a couple of years ago) thought the problem was really about appropriate causality, though I don’t think he explained himself very cogently. He would rightly have said that bringing counterfactuals into it is dangerous stuff because they are a metaphysical and logical minefield. For him the question would be: do the successive states of the computation cause each other in appropriate ways or do they not? If they do, we may have a conscious process; if the causal relations are disrupted or indirect, we’re merely waving flags or playing with puppets. So in his eyes the absence of counterfactual sensitivity in Muhlestein’s example is not decisive and his lights do not remove the consciousness unless they disrupt the underlying causality of the computation: unless they make a difference, in short. Causal over-determination of the process is irrelevant. The problem for Old Peter is in explaining what kinds of causal relations between states have to obtain for the conscious computation to work. His intuition told him that in real processes, including those in the brain, each event causes the next directly, whereas in simulations the causal path is through a model or a script. Unfortunately we could well argue that this kind of indirect causal path is also typical of computation, leading us back to Muhlestein’s conclusion by a different route. Myself, I’m no longer completely sure either that it is a matter of indirect causality or that computational processes are necessarily of that kind.
For myself I should still be inclined towards suspicion of counterfactual sensitivity, but I would be more inclined to say that what we’re missing is the element of intentionality; before we can complete the analysis of the problem we need to know what it is that endows a brain process with meaning and aboutness. The snag with that strategy is that we don’t know.
All in all, I found it interesting to look back a bit. It seems clear to me that over the years this site has been up I have made some distinct progress with consciousness: every year I know a little less about it.