The Simulation Hypothesis is simple: Imagine we can create an AI that’s conscious. We’re not there yet: Current AIs, despite their assurances to the contrary, aren’t conscious. Even so, it’s not hard to imagine that we’ll get there soon.
Once we do, we’ll probably not just make one, but billions and billions of copies of conscious AIs and put them in various scenarios. In many cases, we’ll withhold the knowledge that they’re AIs from them. When those simulated consciousnesses outnumber human consciousnesses, it starts to become likely that any conscious being is simulated. If so, how do we know that we’re not ourselves being simulated by someone else? Actually, is it not much more likely that we’re being simulated? And that’s the Simulation Hypothesis. For an in-depth treatment, read Nick Bostrom’s essay.
There are two different flavors of the Simulation Hypothesis. One is that only I am being simulated, and that only what I perceive needs to be simulated. The rest of the universe, when I don’t pay attention to it, is not part of the simulation. Let’s call this the Only-Me flavor. The other flavor is that the entire universe is being simulated. Let’s call this the Everything flavor. There could of course also be some intermediate flavors, like that only the minds of people are being simulated in a persistent manner, but not the rest of the universe. There are also gradations between the two extreme flavors.
There are two empirical observations about the world as I observe it that both make the Only-Me flavor more likely.
First, the universe appears to be unnecessarily vast and intricate. It seems unlikely that anyone running a simulation would spend the resources that it takes to simulate an entire universe, with all its stars, and the billions of species of living beings on Earth alone.
Second, the beams of our conscious attention are narrow. We can only consciously perceive very little at any given time, even though our senses collect a lot more data. What reaches us is the highly condensed result of many layers of filtering and processing. There is no obvious reason why our consciousnesses wouldn’t be able to perceive more. Why is that? What’s the problem exactly? Our minds being simulated, and the Simulators wanting to not expend the compute resources it’d take to simulate a broader bandwidth of perception would be an explanation.
An unrelated but equally important question regarding the Simulation Hypothesis that the late Jim Rutt liked to ask is, Who cares? Maybe it doesn’t much matter if we’re being simulated. Nothing actionable about how we live our lives or how we do science changes with the Simulation Hypothesis. I think Rutt was wrong about that: There are important real-life implications if the Simulation Hypothesis is true. Nick Bostrom discusses them in depth here. For example: Does investigating the Simulation Hypothesis, and discovering evidence for it being true, increase the chances that the Simulators would consider their experiment a failure and switch off the simulation, killing us in the process?
