Hacker News Reader

Comment by bcherny | Hacker News Reader

It would have been interesting to hear Wolfram's take on the debate. The debate revolved around "why would future intelligent beings want to simulate lesser intelligent beings from the past?". But the broader picture is that given enough time, "intelligence", which really means "computation", is an emergent phenomena of the universe at large. Given enough time, the universe will simulate itself.

We perform computations in the same sense that complex weather patterns do - there's nothing intrinsically special about our ability to compute. So it's naive to think that we're the ones being simulated - the universe may be simulated, and we are but one emergent property of that universe.

View on HN

A universe can't really "simulate itself". There is inherent overhead to simulation, so a perfect simulation of a physical system would have to be larger and slower than the system it's simulating.

Any simulation worth running would be an approximation of the real thing, with many corners cut. Because of butterfly effects, any approximative simulation will diverge from the original, and the more corners you cut, the faster that will happen (so ancestor simulations would require near perfect accuracy, so they would probably be prohibitively expensive).

If we are a simulation, our parent universe would be several orders of magnitude more complex than this one. It's not clear to me why such a simulation would be made (what purpose it would serve to the parent universe).

"larger and slower"... That must be why they say "it's turtles all the way down"!

(For the uninitiated: https://en.wikipedia.org/wiki/Turtles_all_the_way_down )

That's not necessarily true in discrete systems. For example we can cache results of common interactions for reuse. See HashLife and relatives, for example: a vastly more efficient way to compute game of Life without considering one cell at a time, but giving identical results.

I'm aware, but I think the efficiency of these methods is rather circumstantial: they work well because of the way our computers are organized. What you are comparing are two simulation methods, with are both very far removed from what they simulate.

Instead, imagine implementing HashLife inside a Game of Life universe. It is possible, because the system is Turing complete, but the overhead would be horrific: probably thousands of cells for each cell you try to simulate. You would have to duplicate the cache all over the place to minimize roundtrip costs, each cell would be a whole machinery to allow signals to be routed there, plus logic to handle synchronization issues between each "unit", and so on.

It's a very neat idea that works very well with our computer systems, but if you were to implement it using the resources of the simulated system, it would be a catastrophe. And similarly, I don't think you could build a physical "cache" that can simulate physical interactions faster than they happen, except in a few contrived cases.

> It's not clear to me why such a simulation would be made (what purpose it would serve to the parent universe).

Imagine the simulating species is running something akin to a weather simulation, and you just happened to evolve - along with the weather patterns of interest - in the simulation.

That is, it's not about you, it's about the phenomenon the simulators want to study. Maybe it's galactic evolution?

If we are in a simulation, this is probably the case.

It's sad to think that we're just a side effect and being left unnoticed.

In that case, are we special (i.e. unexpected piece of code questioning about its existence)?

We could possibly test that.

If we could prove we were running in a simulation, perhaps we could guess the purpose of the simulation, and interfere with the results in a manner that might catch the attention of the simulators?

"This is the third time in a row that the weather simulation has produced rain only on prime numbered days."

Better hope the simulators don't fix that problem by re-imaging the universe.

"There is inherent overhead to simulation"

This is true of simulations in our universe. There's no way to know how external universes might work.

I had this same thought a while ago. If our universe is just the Nth universe in a chain of simulated universes, and the number_of_bits(N) < number_of_bits(N-1), then each simulation gets more and more "compressed" and approximate the deeper you go.

On a similar basis I take some exception to this statement, regarding individual immortality:

> If we’re programs in the computer, then as long as I have a computer that’s not damaged, I can always re-run the program.

You would quite possibly need to dump the entire universe state and re-run it from that point to get anything predictably "you" back out of it - probably not what most people have in mind when they think of a brain in a vat.

Down that road lies Greg Egan's Dust hypothesis; it is a slippery slope.

http://schwitzsplinters.blogspot.com/2009/01/dust-hypothesis...

http://gregegan.customer.netspace.net.au/PERMUTATION/FAQ/FAQ...

I actually thought that was the general assumption, rather than we, humans, being anything special in the ridiculously vast universe.

Reply on news.ycombinator.com