Roughly, pain is a sensation typically associated with damage to the body, suffering is an experience of stimuli as intrinsically unpleasant.
I do not suffer if my room is painted a color I do not like, but I still may care about the color my room is painted.
It means “being able to feel pain but not suffering from it.”
Suppose an AI were to design and implement more efficient algorithms for processing sensory stimuli? Or add a “face recognition” module when it determines that this would be useful for interacting with humans?
The ancient Greeks have developed methods of improved memorization. It has been shown that human-trained dogs and chimps are more capable of human-face recognition than others of their kind. None of them were artificial (discounting selective breeding in dogs and Greeks).
It seems that you should be able to write a simple program that overwrites its own code with an arbitrary value. Wouldn’t that be a counterexample?
Would you consider such a machine an artificial intelligent agent? Isn’t it just a glorified printing press?
I’m not saying that some configurations of memory are physically impossible. I’m saying that intelligent agency entails typicality, and therefore, for any intelligent agent, there are some things it is extremely unlikely to do, to the point of practical impossibility.
Certainly that doesn’t count as an intelligent agent—but a GAI with that as its only goal, for example, why would that be impossible? An AI doesn’t need to value survival.
I’d be interested in the conclusions derived about “typical” intelligences and the “forbidden actions”, but I don’t see how you have derived them.
Do we agree, then, that humans and artificial agents are both subject to laws forbidding logical contradictions and the like, but that artificial agents are not in principle necessarily bound by the same additional restrictions as humans?
I would actually argue the opposite.
Are you familiar with the claim that people are getting less intelligent since modern technology allows less intelligent people and their children to survive? (I never saw this claim discussed seriously, so I don’t know how factual it is; but the logic of it is what I’m getting at.) The idea is that people today are less constrained in their required intelligence, and therefore the typical human is becoming less intelligent.
Other claims are that activities such as browsing the internet and video gaming are changing the set of mental skills which humans are good at. We improve in tasks which we need to be good at, and give up skills which are less useful. You gave yet another example in your comment regarding face recognition.
The elasticity of biological agents is (quantitatively) limited, and improvement by evolution takes time. This is where artificial agents step in. They can be better than humans, but the typical agent will only actually be better if it has to. Generally, more intelligent agents are those which are forced to comply to tighter constraints, not looser ones.
I think we have our quantifiers mixed up? I’m saying an AI is not in principle bound by these restrictions—that is, it’s not true that all AIs must necessarily have the same restrictions on their behavior as a human. This seems fairly uncontroversial to me. I suppose the disconnect, then, is that you expect a GAI will be of a type bound by these same restrictions. But then I thought the restrictions you were talking about were “laws forbidding logical contradictions and the like”? I’m a little confused—could you clarify your position, please?
Could you rephrase this somehow? I’m not understanding it. If you actually won the bet and got the extra utility, your median expected utility would be higher, but you wouldn’t take the bet, because your median expected utility is lower if you do.
“Enough times” to make it >50% likely that you will win, yes? Why is this the correct cutoff point?
This all seems very sensible and plausible!
Thanks for challenging my position. This discussion is very stimulating for me!
It’s a pleasure!
Sure, but we could imagine an AI deciding something like “I do not want to enjoy frozen yogurt”, and then altering its code in such a way that it is no longer appropriate to describe it as enjoying frozen yogurt, yeah?
I’m actually having trouble imagining this without anthropomorphizing (or at least zoomorphizing) the agent. When is it appropriate to describe an artificial agent as enjoying something? Surely not when it secretes serotonin into its bloodstream and synapses?
Yeah, that was sloppy of me. Leaving aside the question of when something is enjoying something, let’s take a more straightforward example: Suppose an AI were to design and implement more efficient algorithms for processing sensory stimuli? Or add a “face recognition” module when it determines that this would be useful for interacting with humans?
This seems trivially false—if an AI is instantiated as a bunch of zeros and ones in some substrate, how could Godel or similar concerns stop it from altering any subset of those bits?
It’s not a question of stopping it. Gödel is not giving it a stern look, saying: “you can’t alter your own code until you’ve done your homework”. It’s more that these considerations prevent the agent from being in a state where it will, in fact, alter its own code in certain ways. This claim can and should be proved mathematically, but I don’t have the resources to do that at the moment. In the meanwhile, I’d agree if you wanted to disagree.
Hm. It seems that you should be able to write a simple program that overwrites its own code with an arbitrary value. Wouldn’t that be a counterexample?
You see reasons to believe that any artificial intelligence is limited to altering its motivations and desires in a way that is qualitatively similar to humans? This seems like a pretty extreme claim—what are the salient features of human self-rewriting that you think must be preserved?
I believe that this is likely, yes. The “salient feature” is being subject to the laws of nature, which in turn seem to be consistent with particular theories of logic and probability. The problem with such a claim is that these theories are still not fully understood.
This sounds unjustifiably broad. Certainly, human behavior is subject to these restrictions, but it is also subject to much more stringent ones—we are not able to do everything that is logically possible. Do we agree, then, that humans and artificial agents are both subject to laws forbidding logical contradictions and the like, but that artificial agents are not in principle necessarily bound by the same additional restrictions as humans?
knowing the value of Current Observation gives you information about Future Decision.
Here I’d just like to note that one must not assume all subsystems of Current Brain remain constant over time. And what if the brain is partly a chaotic system? (AND new information flows in all the time… Sorry, I cannot condone this model as presented.)
Well… okay, but the point I was making was milder and pretty uncontroversial. Are you familiar with bayesian networks?
Perhaps it can observe your neurochemistry in detail and in real time.
I already mentioned this possibility. Fallible models make the situation gameable. I’d get together with my friends, try to figure out when the model predicts correctly, calculate its accuracy, work out a plan for who picks what, and split the profits between ourselves. How’s that for rationality? To get around this, the alien needs to predict our plan and—do what? Our plan treats his mission like total garbage. Should he try to make us collectively lose out? But that would hamper his initial design.
(Whether it cares about such games or not, what input the alien takes, when, how, and what exactly it does with said input—everything counts in charting an optimal solution. You can’t just say it uses Method A and then replace it with Method B when convenient. THAT is the point: Predictive methods are NOT interchangeable in this context. (Reminder: Reading my brain AS I make the decision violates the original conditions.))
I never said it used method A? And what is all this about games? It predicts your choice.
You’re not engaging with the thought experiment. How about this—how would you change the thought experiment to make it work properly, in your estimation?
Perhaps land-ape psychology turns out to be really simple if you’re an omnipotent thought-experiment enthusiast.
We’re veering into uncertain territory again… (Which would be fine if it weren’t for the vagueness of mechanism inherent in magical algorithms.)
Well, yeah. We’re in uncertain territory as a premise.
The reasoning wouldn’t be “this person is a one-boxer” but rather “this person will pick one box in this particular situation”.
Second note: An entity, alien or not, offering me a million dollars, or anything remotely analogous to this, would be a unique event in my life with no precedent whatever. My last post was written entirely under the assumption that the alien would be using simple heuristics based on similar decisions in the past. So yeah, if you’re tweaking the alien’s method, then disregard all that.
I’m not tweaking the method. There is no given method. The closest to a canonical method that I’m aware of is simulation, which you elided in your reply.
It’s very difficult to be the sort of person who would pick one box in the situation you are in without actually picking one box in the situation you are in.
From the alien’s point of view, this is epistemologically non-trivial if my box-picking nature is more complicated than a yes-no switch. Even if the final output must take the form of a yes or a no, the decision tree that generated that result can be as endlessly complex as I want, every step of which the alien must predict correctly (or be a Luck Elemental) to maintain its reputation of infallibility.
What makes you think you’re so special—compared to the people who’ve been predicted ahead of you?
If it’s worse, just do the other thing—isn’t that more “rational”?
As long as I know nothing about the alien’s method, the choice is arbitrary. See my second note. This is why the alien’s ultimate goals, algorithms, etc, MATTER.
If you know nothing about the alien’s methods, there still is a better choice. You do not have the same expected value for each choice.
(If the alien reads my brain chemistry five minutes before The Task, his past history is one of infallibility, and no especially cunning plan comes to mind, then my bet regarding the nature of brain chemistry would be that not going with one box is silly if I want the million dollars. I mean, he’ll read my intentions and place the money (or not) like five minutes before… (At least that’s what I’ll determine to do before the event. Who knows what I’ll end up doing once I actually get there. (Since even I am unsure as to the strength of my determination to keep to this course of action once I’ve been scanned, the conscious minds of me and the alien are freed from culpability. Whatever happens next, only the physical stance is appropriate for the emergent scenario. ((“At what point then, does decision theory apply here?” is what I was getting at.) Anyway, enough navel-gazing and back to Timeless Decision Theory.))))
Ah! Sorry for the mixed-up identities. Likewise, I didn’t come up with that “51% chance to lose $5, 49% chance to win $10000” example.
But, ah, are you retracting your prior claim about a variance of greater than 5? Clearly this system doesn’t work on its own, though it still looks like we don’t know A) how decisions are made using it or B) under what conditions it works. Or in fact C) why this is a good idea.
Certainly for some distributions of utility, if the agent knows the distribution of utility across many agents, it won’t make the wrong decision on that particular example by following this algorithm. I need more than that to be convinced!
For instance, it looks like it’ll make the wrong decision on questions like “I can choose to 1) die here quietly, or 2) go get help, which has a 1⁄3 chance of saving my life but will be a little uncomfortable.” The utility of surviving presumably swamps the rest of the utility function, right?
Googol. Likewise, googolplex.
But the median outcome is losing 5 utils?
Edit: Oh, wait! You mean the median total utility after some other stuff happens (with a variance of more than 5 utils)?
Suppose we have 200 agents, 100 of which start with 10 utils, the rest with 0. After taking this offer, we have 51 with −5, 51 with 5, 49 with 10000, and 49 with 10010. The median outcome would be a loss of −5 for half the agents, a gain of 5 for half, but only the half that would lose could actually get that outcome…
And what do you mean by “the possibility of getting tortured will manifest itself only very slightly at the 50th percentile”? I thought you were restricting yourself to median outcomes, not distributions? How do you determine the median distribution?
You are saying that a GAI being able to alter its own “code” on the actual code-level does not imply that it is able to alter in a deliberate and conscious fashion its “code” in the human sense you describe above?
I am saying pretty much exactly that. To clarify further, the words “deliberate”, “conscious” and “wants” again belong to the level of emergent behavior: they can be used to describe the agent, not to explain it (what could not be explained by “the agent did X because it wanted to”?).
Sure, but we could imagine an AI deciding something like “I do not want to enjoy frozen yogurt”, and then altering its code in such a way that it is no longer appropriate to describe it as enjoying frozen yogurt, yeah?
Let’s instead make an attempt to explain. A complete control of an agent’s own code, in the strict sense, is in contradiction of Gödel’s incompleteness theorem. Furthermore, information-theoretic considerations significantly limit the degree to which an agent can control its own code (I’m wondering if anyone has ever done the math. I expect not. I intend to look further into this). In information-theoretic terminology, the agent will be limited to typical manipulations of its own code, which will be a strict (and presumably very small) subset of all possible manipulations.
This seems trivially false—if an AI is instantiated as a bunch of zeros and ones in some substrate, how could Godel or similar concerns stop it from altering any subset of those bits?
Can an agent be made more effective than humans in manipulating its own code? I have very little doubt that it can. Can it lead to agents qualitatively more intelligent than humans? Again, I believe so. But I don’t see a reason to believe that the code-rewriting ability itself can be qualitatively different than a human’s, only quantitatively so (although of course the engineering details can be much different; I’m referring to the algorithmic level here).
You see reasons to believe that any artificial intelligence is limited to altering its motivations and desires in a way that is qualitatively similar to humans? This seems like a pretty extreme claim—what are the salient features of human self-rewriting that you think must be preserved?
Generally GAIs are ascribed extreme powers around here
As you’ve probably figured out, I’m new here. I encountered this post while reading the sequences. Although I’m somewhat learned on the subject, I haven’t yet reached the part (which I trust exists) where GAI is discussed here.
On my path there, I’m actively trying to avoid a certain degree of group thinking which I detect in some of the comments here. Please take no offense, but it’s phrases like the above quote which worry me: is there really a consensus around here about such profound questions? Hopefully it’s only the terminology which is agreed upon, in which case I will learn it in time. But please, let’s make our terminology “pay rent”.
I don’t think it’s a “consensus” so much as an assumed consensus for the sake of argument. Some do believe that any hypothetical AI’s influence is practically unlimited, some agree to assume that because it’s not ruled out and is a worst-case scenario or an interesting case (see wedrifid’s comment on the grandparent (aside: not sure how unusual or nonobvious this is, but we often use familial relationships to describe the relative positions of comments, e.g. the comment I am responding to is the “parent” of this comment, the one you were responding to when you wrote it is the “grandparent”. I think that’s about as far as most users take the metaphor, though.)).
Unlike my (present) traits, my future decisions don’t yet exist, and hence cannot leak anything or become entangled with anyone.
Your future decisions are entangled with your present traits, and thus can leak. If you picture a Bayesian network with the nodes “Current Brain”, “Future Decision”, and “Current Observation”, with arrows from Current Brain to the two other nodes, then knowing the value of Current Observation gives you information about Future Decision.
Obviously the alien is better than a human at running this game (though, note that a human would only have to be right a little more than 50% of the time to make one-boxing have the higher expected value—in fact, that could be an interesting test to run!). Perhaps it can observe your neurochemistry in detail and in real time. Perhaps it simulates you in this precise situation, and just sees whether you pick one or both boxes. Perhaps land-ape psychology turns out to be really simple if you’re an omnipotent thought-experiment enthusiast.
The reasoning wouldn’t be “this person is a one-boxer” but rather “this person will pick one box in this particular situation”. It’s very difficult to be the sort of person who would pick one box in the situation you are in without actually picking one box in the situation you are in.
One use of the thought experiment, other than the “non-causal effects” thing, is getting at this notion that the “rational” thing to do (as you suggest two-boxing is) might not be the best thing. If it’s worse, just do the other thing—isn’t that more “rational”?
Having asserted that your claim is, in fact, new information
I wouldn’t assert that. I thought I was stating the obvious.
Yes, I think I misspoke earlier, sorry. It was only “new information” in the sense that it wasn’t in that particular sentence of Eliezer’s—to anyone familiar with discussions of GAI, your assertion certainly should be obvious.
You are saying that a GAI being able to alter its own “code” on the actual code-level does not imply that it is able to alter in a deliberate and conscious fashion its “code” in the human sense you describe above?
Generally GAIs are ascribed extreme powers around here—if it has low-level access to its code, then it will be able to determine how its “desires” derive from this code, and will be able to produced whatever changes it wants. Similarly, it will be able to hack human brains with equal finesse.
An advanced AI could reasonably be expected to be able to explicitly edit any part of its code however it desires. Humans are unable to do this.
Not meant as an attack. I’m saying, “to be fair it didn’t actually say that in the original text, so this is new information, and the response is thus a reasonable one”. Your comment could easily be read as implying that this is not new information (and that the response is therefore mistaken), so I wanted to add a clarification.
But ‘value is fragile’ teaches us that it can’t be a 1-dimensional number like the reals.
This is not in fact what “value is fragile” teaches us, and it is false. Without intending offense, I recommend you read about utility a bit more before presenting any arguments about it here, as it is in fact a 1-dimensional value.
What you might reasonably conclude, though, is that utility is a poor way to model human values, which, most of the time, it is. Still, that does not invalidate the results of properly-formed thought experiments.
To be fair, when structured as
Sadly, we humans can’t rewrite our own code, the way a properly designed AI could.
then the claim is in fact “we humans can’t rewrite our own code (but a properly designed AI could)”. If you remove a comma:
Sadly, we humans can’t rewrite our own code the way a properly designed AI could.
only then is the sentence interpreted as you describe.
Yeah, that’s certainly a fair clarification. It’d probably take a lot more space to give a really robust definition of “suffering”, but that’s close enough for gummint work.