simulations

In an article on his minimum viable homepage Keunwoo Lee argues against the simulation argument. In short, "future superintelligences" would be too busy with other things and "once you have the resources to simulate many beings of a given type with high fidelity, doing so becomes uninteresting".
I do not disagree with this argument, but I disagree that "superintelligence" is required to simulate us.

Every simulation or simulation program would be equivalent to a long string of 0s and 1s; in other words it would be equivalent to a very large natural number. It can indeed require intelligence to calculate a specific large number, e.g. a large prime number or a particular simulation.
But it does not require "superintelligence" to calculate all numbers, it only requires a counter with vast resources, i.e. a simple Turing machine.

One could argue that the simulation programs need to be somehow 'executed', but again a simple process with vast resources could do that:
If Sij stands for the i-th program step of the j-th simulation, then we can arrange S11, S12, S13, ... S21, S22, ... in an infinite square and use a variant of the proof that rational numbers are countable to show that i) a simple Turing machine can generate all possible simulations and ii) another simple machine can actually 'execute' all possible simulations. The combination of i) + ii) would basically be a giant clock, no "superintelligence" would be necessary, just a (quasi)infinite amount of resources.
However, we do not know what resources are available in the "really real world" and therefore we cannot estimate how likely it is that we are the result of such a process.

But we could be the result of a giant clock counting time ... tik, tok ...

1 comment:

Blog Archive