If the simulation trilemma is correct, it is also trivial
Reading through the Bostrom paper the first time, I missed his definition of "posthuman". Thankfully, the term is defined:
The simulation argument works equally well for those who think that it will take hundreds of thousands of years to reach a “posthuman” stage of civilization, where humankind has acquired most of the technological capabilities that one can currently show to be consistent with physical laws and with material and energy constraints.
For the purposes of the paper, the critical stage is the moment when humanity is able to harness the essentially unlimited computational power that Bostrom assumes is physically possible. So the first proposition asserts that this will likely never happen either because of some unforeseen limitation or because civilizations such as our own will become extinct. (Or, really, for both reasons at the same time.) The second proposition asserts that posthumanity will be willing to conduct the sorts of simulations that are contemplated in the third proposition. If we eliminate two of the possibilities, according to Bostrom, the third must be true.
To rephrase the trilemma:
A. If a civilization can obtain unlimited computational power,
B. And there are no self-imposed restrictions to using it,
C. Then it will be used.
Bostrom further argues from C that statistically speaking, we must assume that we are part of a simulation. In order for the simulation argument to work, you must push the number of simulations as if there are essentially no resource limits. But the sticking point isn't whether any civilization will achieve this particular posthuman state, but whether such a state is possible. Once I swallow the elephant of limitless power, the gnat of this particular science fiction scenario should go down well enough.
In his FAQ, Bostrom mentions that the argument may be generalized. So, if you plug in unlimited time travel for unlimited computational power, you could formulate an argument that you are most likely the decedent of your ancestor. At the moment, unlimited time travel lacks empirical evidence, however. Or you could plug in unlimited capacity for causing suffering and arrive at the conclusion that our ancestors are certain to become sadists. If you begin with the assumption that X is potentially unlimited, then, given enough time and an agent who willing to activate it, X is certain to have extreme influence.
Our universe need not provide us the relevant evidence
Much of the paper is dedicated to describing research that considers the limits of computational power. While that research is fascinating, it's also entirely irrelevant to the argument. The easiest way to verify this is to consider the relationship between our computers and the simulations we run on them. An obvious observation is that we simulate a restricted portions of the universe. Professionally, I deal with weather simulations and while there are some that consider the entire atmosphere, most consider only a portion of the atmosphere (say the troposphere over North America) and have limited duration.
While it's true limited computational power plays a role in limiting our simulations, a bigger problem is simulation drift. Edward Norton Lorenz discovered small deviations in initial conditions will result in radically different results over time. He expressed it with the fanciful idea that a butterfly could cause a tornado. In practice, because our measurement of atmospheric properties is inexact, simulations must be reset with new input parameters on a regular basis. Increasing computational power cannot mitigate against this problem. Due to the nature of measurement, it's also impossible for us to improve our measurements to the point that we can run perpetually accurate simulations; the act of measurement produces tiny, unpredictable changes on the properties we are trying to measure.
More troubling, a case can be made that the vast majority of our simulations are of fictional scenarios, i.e. video games. The evidence of our current activities would suggest that simulations with important variables changed for no other reason than to provide entertainment far outweigh "realist" simulations. There's no reason to not conclude that the fickle tastes of the civilization that is simulating us are stimulated by giving us the illusion that computing power could be virtually unlimited. There's no reason to assume that the most prolific posthuman simulations authors aren't their generation's Jules Verne, H. G. Wells, and Issac Asimov.
The simulation argument ignores epistemological problems
From Bostrom's FAQ:
The purpose of the simulation argument is different: not to set up a skeptical problem as a challenge to epistemological theories and common sense, but rather to argue that we have interesting empirical reasons to believe that a certain disjunctive claim about the world is true (that is, (1)v(2)v(3)). The simulation argument relies crucially on non-obvious empirical premises about future technological abilities. And the conclusion of the simulation argument is not simply that we cannot be certain that we are not living in a simulation. If we knew that fSIM (the faction of all human-like beings who are simulated) was very small but non-zero, we might not be able to be completely certain that we are not in a simulation, but this would not be a very interesting contention.
In other words, among many other assumptions, the argument assumes that we can trust our empirical observations. If we take the Aristotelian assumption that the world around us is the only reality that exists, then we have no choice in the matter. But a simulation inhabited by human-like beings must contain a reality greater than its own. If we are to know anything about that other reality, we must develop an epistemology of transcendence. Our model ought to be Plato, not Aristotle. Our a priori assumptions must include some way of relating knowledge obtained empirically to knowledge of the transcendent reality. Bostrom, in his reply to Brian Weatherson [PDF], sweeps the issue under the rug; he argues that the simulated human is only likely to be wrong when it comes to the question of whether or not they are simulated and in all other respects are epistemologically competent.
But the critical question is not whether the simulated human can have transcendental knowledge, but whether observations about computational power, human psychology, and statistical certainty are transferable from this reality to any potential transcendent reality. I would argue that if they are transferable, they must endure some sort of transformation themselves. If as a species we create a certain ratio of descendants that happen to reside in computer simulations to descendants that live in physical space, it tells us little to nothing about the ratio of descendants our creators have spawned of each type. Any conclusions we draw from this ratio must be tempered by humility.
Conclusion
As I read the paper, I kept waiting for the punchline. But it never came. Nick Bostrom seems sincere in his belief that the castle in the air that he has constructed is supported by empirical evidence even though our current observations show that no human-like being has ever been simulated on a computer and 100% of the people known to exist are part of the same reality that we inhabit. Therefore, if we are simulated, we have been deceived by someone very powerful and of questionable morality.
