Andrew Jacob Sauer

Posts

Sorted by New

Comments

The Darwin Game

Is everybody's code going to be in Python?

The Darwin Game

What are the rules about program runtime?

Brainstorming positive visions of AI

A common concern around here seems to be that, without massive and delicate breakthroughs in our understanding of human values, any superintelligence will destroy all value by becoming some sort of paperclip optimizer. This is what Eliezer claims in Value is Fragile. Therefore, any vision of the future that manages to do better than this without requiring huge philosophical breakthroughs (in particular, a future that doesn’t know how to implement CEV before the Singularity happens) is encouraging to me as a proof of concept for how the future might be more likely to go well.

In a future where uploading minds into virtual worlds becomes possible before an AI takeover, there might well be a way to salvage quite a lot of human value with a very comparatively simple utility function: simply create a big virtual world and upload lots of people into it, then have the AI’s whole goal be to run this simulation for as long as possible.

This idea of “just run this program” seems a lot more robust and more likely to work and less likely to be exploited than attempting to maximize some utility function meant to represent human values, and the result would probably be better than what would happen if the latter went wrong. I suspect it would be well within the capability of a society which can upload minds to create a virtual world for these minds where the only scarce resource is computation cycles and there is no way to forcibly detain someone, so this virtual world would not have many of the problems our current world has.

This is far from a perfect outcome, of course. The AI would likely destroy everything it touches for resources, killing everyone not fortunate enough to get uploaded. And there are certainly other problems with any idea of “virtual utopia” we could come up with. But this idea gives me hope because it might be improved upon, and because it is a way that we don’t lose everything even if CEV proves too hard of a problem to solve before Singularity.

Troll Bridge

Thanks for the link, I will check it out!

War and/or Peace (2/8)
As for cannibalism, it seems to me that its role in Eliezer's story is to trigger a purely illogical revulsion in the humans who antropomorphise the aliens.

I dunno about you but my problem with the aliens isn't that it is cannibalism but that the vast majority of them die slow and horribly painful deaths

No cannibalism takes place, but the same amount of death and suffering is present as in Eliezer's scenario. Should we be less or more revolted at this?

The same.

Which scenario has the greater moral weight?

Neither. They are both horrible.

Should we say the two-species configuration is morally superior because they've developed a peaceful, stable society with two intelligent species coexisting instead of warring and hunting each other?

Not really because most of them still die slow and horribly painful deaths.

Troll Bridge

Sorry to necro this here, but I find this topic extremely interesting and I keep coming back to this page to stare at it and tie my brain in knots. Thanks for your notes on how it works in the logically uncertain case. I found a different objection based on the assumption of logical omniscience:

Regarding this you say:

Perhaps you think that the problem with the above version is that I assumed logical omniscience. It is unrealistic to suppose that agents have beliefs which perfectly respect logic. (Un)Fortunately, the argument doesn't really depend on this; it only requires that the agent respects proofs which it can see, and eventually sees the Löbian proof referenced.

However, this assumes that the Löbian proof exists. We show that the Löbian proof of A=cross→U=−10 exists by showing that the agent can prove □(A=cross→U=−10)→(A=cross→U=−10), and the agent's proof seems to assume logical omniscience:

Examining the agent, either crossing had higher expected utility, or P(cross)=0. But we assumed □(A=cross→U=−10), so it must be the latter. So the bridge gets blown up.

If here means "provable in PA", the logic does not follow through if the agent is not logically omniscient: the agent might find crossing to have a higher expected utility regardless, because it may not have seen the proof. If here instead means "discoverable by the agent's proof search" or something to that effect, then the logic here seems to follow through (making the reasonable assumption that if the agent can discover a proof for A=cross->U=-10, then it will set its expected value for crossing to -10). However, that would mean we are talking about provability in a system which can only prove finitely many things, which in particular cannot contain PA and so Löb's theorem does not apply.

I am still trying to wrap my head around exactly what this means, since your logic seems unassailable in the logically omniscient case. It is counterintuitive to me that the logically omniscient agent would be susceptible to trolling but the more limited one would not. Perhaps there is a clever way for the troll to get around this issue? I dunno. I certainly have no proof that such an agent cannot be trolled in such a way.

The Strangest Thing An AI Could Tell You

That's what I was thinking. Garbage in, garbage out.

Is this viable physics?

This seems equivalent to Tegmark Level IV Multiverse to me. Very simple, and probably our universe is somewhere in there, but doesn't have enough explanatory power to be considered a Theory of Everything in the physical sense.

Two Alternatives to Logical Counterfactuals

From an omniscient point of view, yes. From my point of view, probably not, but there are still problems that arise relating to this, that can cause logic-based agents to get very confused.

Let A be an agent, considering options X and not-X. Suppose A |- Action=not-X -> Utility=0. The naive approach to this would be to say: if A |- Action=X -> Utility<0, A will do not-X, and if A |- Action=X -> Utility>0, A will do X. Suppose further that A knows its source code, so it knows this is the case.
Consider the statement G=(A |- G) -> (Action=X -> Utility<0). It can be constructed by using Godel-numbering and quines. Present A with the following argument:

Suppose for the sake of argument that A |- G. Then A |- (A |- G), since A knows its source code. Also, by definition of G, A |- (A |- G) -> (Action=X -> Utility<0). By modus ponens, A |- (Action=X -> Utility<0). Therefore, by our assumption about A, A will do not-X: Action!=X. But, vacuously, this means that (Action=X -> Utility<0). Since we have proved this by assuming A |- G, we know that (A |- G) -> (Action=X -> Utility<0), in other words, we know G.

The argument then goes, similarly to above:
A |- G
A |- (A |- G)
A |- (A |- G) -> (Action=X -> Utility<0)
A |- (Action=X -> Utility<0)
Action=Not-X

We proved this without knowing anything about X. This shows that naive logical implication can easily lead one astray. The standard solution to this problem is the chicken rule, making it so that if A ever proves which action it will take, it will immediately take the opposite action, which avoids the argument presented above, but is defeated by Troll Bridge, even when the agent has good logical uncertainty.

These problems seem to me to show that logical uncertainty about the action one will take, paired with logical implications about what the result will be if you take a particular action, are insufficient to describe a good decision theory.

Load More