Commentary: Week 9 Mike LeBeau

SSP 205

Spring 2005

Many of the claims that Bostrom declares more or less as fact in his “simulation argument” seem somewhat unsubstantiated, and at the very least could be objected to on many grounds. His claiming to know, for example, a “rough estimate … of the [computational] cost of a realistic simulation of human history” as being “~10³²-10³⁵ operations”, or similarly, his claiming to know an “estimate of … how computationally expensive it is to replicate the functionality of … the entire human brain”, seems rather absurd. In large part these claims seem absurd because they are based upon higher-level claims that are far more conjecture than fact. One such claim is that it will one day be possible (1) to simulate precisely the functioning of neuronal circuitry and (2) to replicate the brain completely by creating such a simulation which replicates entirely the physical mechanisms of the brain, “down to the neuronal or sub-neuronal level”.

The idea that we could eventually understand the mechanics of anything and thus simulate it perfectly on a computer is quite compelling. However, until now humans have not had much luck simulating anything ‘perfectly’ by creating simulation models. A perfect simulation would not have to be one that resulted in the same outcome as reality every time, but would have to produce results in a range identical to the range of results produced by the simulated artifact. Note that in actuality, it may be impossible to ever know completely whether a simulation is totally faithful, since we cannot rewind and replay time to see what the range of possible results produced by, say, the human brain might be, in a particular circumstance. We may be very convinced by a simulation, but this is by no means a guarantee that it is completely accurate in every detail.

Our current inability to create completely faithful simulations, Bostrom would likely argue, is simply due to the fact that we are not yet sufficiently technologically advanced to create such simulations. What he would likely not address is the issue that perhaps a particular concept is more complex than we will be able to understand. Even assuming that we will one day become so intelligent that we can understand the inner workings of anything in existence, Bostrom does not take into account potential computability concerns when claiming that we will be able to accurately simulate the human brain. Since we do not know exactly how the brain works, there may well be something more to it than a computer is able to compute; there may be more than just the interaction of neuronal tissues. To touch on a more spiritual point, Bostrom ignores the potential discrepancy between mind and brain, assuming the functionalist perspective that the mind is merely a product of the inner workings of the brain, and while this seems plausible, it is by no means certain.

It seems clear without even getting into Bostrom’s explicit mathematical claims that there are fundamental assumptions which he chooses to gloss over for the sake of his argument. These assumptions are significant and should not be overlooked, and Bostrom’s unwillingness to address these issues make his rather bold claim even less compelling.