Personally, if I were to take the approach of a preprogrammed ethics, I would define good in pseudo-evolutionary terms: a pattern/entity is good if it has high survival value in the long term. Patterns that are self-sustaining on their own are thus considered good, but patterns that help sustain other patterns would be too, because they are a high-utility part of a larger whole.
Actually, I *do* define good and ethics not only in evolutionary terms but as being driven by evolution. Unlike most people, I believe that ethics is *entirely* driven by what is best evolutionarily while not believing at all in "red in tooth and claw". I can give you a reading list that shows that the latter view is horribly outdated among people who keep up with the research rather than just rehashing tired old ideas.
Actually, that idea is what made me assert that any goal produces normalizing subgoals. Survivability helps achieve any goal, as long as it isn't a time-bounded goal (finishing a set task).
Ah, I'm starting to get an idea of what you mean behind normalizing subgoals . . . . Yes, absolutely except that I contend that there is exactly one normalizing subgoal (though some might phrase it as two) that is normally common to virtually every goal (except in very extreme/unusual circumstances).
----- Original Message ----- From: "Abram Demski" <[EMAIL PROTECTED]>
To: <agi@v2.listbox.com> Sent: Thursday, August 28, 2008 4:04 PMSubject: **SPAM** Re: AGI goals (was Re: Information theoretic approaches to AGI (was Re: [agi] The Necessity of Embodiment))
Mark, I still think your definitions still sound difficult to implement, although not nearly as hard as "make humans happy without modifying them". How would you define "consent"? You'd need a definition of decision-making entity, right? Personally, if I were to take the approach of a preprogrammed ethics, I would define good in pseudo-evolutionary terms: a pattern/entity is good if it has high survival value in the long term. Patterns that are self-sustaining on their own are thus considered good, but patterns that help sustain other patterns would be too, because they are a high-utility part of a larger whole. Actually, that idea is what made me assert that any goal produces normalizing subgoals. Survivability helps achieve any goal, as long as it isn't a time-bounded goal (finishing a set task). --Abram On Thu, Aug 28, 2008 at 2:52 PM, Mark Waser <[EMAIL PROTECTED]> wrote:However, it doesn't seem right to me to preprogram an AGI with a set ethical theory; the theory could be wrong, no matter how good it sounds.Why not wait until a theory is derived before making this decision? Wouldn't such a theory be a good starting point, at least?better to put such ideas in only as probabilistic correlations (or "virtual evidence"), and let the system change its beliefs based on accumulated evidence. I do not think this is overly risky, because whatever the system comes to believe, its high-level goal will tend to create normalizing subgoals that will regularize its behavior.You're getting into implementation here but I will make a couple of personalbelief statements:1. Probabilistic correlations are much, *much* more problematical than mostpeople are event willing to think about. They work well with very simple examples but they do not scale well at all. Particularly problematic for such correlations is the fact that ethical concepts are generally made up *many* interwoven parts and are very fuzzy. The church of Bayes does not cut it for any work where the language/terms/concepts are not perfectly crisp, clear, and logically correct. 2. Statements like "its high-level goal will tend to create normalizing subgoals that will regularize its behavior" sweep *a lot* of detail under the rug. It's possible that it is true. I think that it is much more probable that it is very frequently not true. Unless you do *a lot* of specification, I'm afraid that expecting this to be true is *very* risky.I'll stick to my point about defining "make humans happy" being hard, though. Especially with the restriction "without modifying them" that you used.I think that defining "make humans happy" is impossible -- but that's OK because I think that it's a really bad goal to try to implement.All I need to do is to define learn, harm, and help. Help could be defined as anything which is agreed to with informed consent by the affected subjectboth before and after the fact. Yes, that doesn't cover all actions but that just means that the AI doesn't necessarily have a strong inclinationtowards those actions. Harm could be defined as anything which is disagreedwith (or is expected to be disagreed with) by the affected subject either before or after the fact. Friendliness then turns into something like asking permission. Yes, the Friendly entity won't save you in many circumstances, but it's not likely to kill you either. << Of course, I could also come up with the counter-argument to my own thesis that the AI will never do anything because there will always be someone who objects to the AI doing *anything* to change the world.-- but that's just the absurdity and self-defeating arguments that I expect from many of the list denizens that can't be defended against except by allocating far more time than it's worth.>> ----- Original Message ----- From: "Abram Demski" <[EMAIL PROTECTED]> To: <agi@v2.listbox.com> Sent: Thursday, August 28, 2008 1:59 PMSubject: **SPAM** Re: AGI goals (was Re: Information theoretic approaches toAGI (was Re: [agi] The Necessity of Embodiment))Mark, Actually I am sympathetic with this idea. I do think good can be defined. And, I think it can be a simple definition. However, it doesn't seem right to me to preprogram an AGI with a set ethical theory; the theory could be wrong, no matter how good it sounds. So, better to put such ideas in only as probabilistic correlations (or "virtual evidence"), and let the system change its beliefs based on accumulated evidence. I do not think this is overly risky, because whatever the system comes to believe, its high-level goal will tend to create normalizing subgoals that will regularize its behavior. I'll stick to my point about defining "make humans happy" being hard, though. Especially with the restriction "without modifying them" that you used.On Thu, Aug 28, 2008 at 12:38 PM, Mark Waser <[EMAIL PROTECTED]> wrote:Also, I should mention that the whole construction becomes irrelevant if we can logically describe the goal ahead of time. With the "make humans happy" example, something like my construction would be useful if we need to AI to *learn* what a human is and what happy is. (We then set up the pleasure in a way that would help the AI attach "goodness" to the right things.) If we are able to write out the definitions ahead of time, we can directly specify what goodness is instead. But, I think it is unrealistic to take that approach, since the definitions would be large and difficult....:-) I strongly disagree with you. Why do you believe that having a newAI learn large and difficult definitions is going to be easier and safer thanspecifying them (assuming that the specifications can be grounded in theAI's terms)? I also disagree that the definitions are going to be as large as people believe them to be . . . .Let's take the Mandelbroit set as an example. It is perfectly specifiedbyone *very* small formula. Yet, if you don't know that formula, you could spend many lifetimes characterizing it (particularly if you're trying todoing it from multiple blurred and shifted images :-).The true problem is that humans can't (yet) agree on what goodness is --andthen they get lost arguing over detailed cases instead of focusing on thecore. The core definition of goodness/morality and developing a system to determine what actions are good and what actions are not is a project thatI've been working on for quite some time and I *think* I'm making rathergood headway.----- Original Message ----- From: "Abram Demski" <[EMAIL PROTECTED]>To: <agi@v2.listbox.com> Sent: Thursday, August 28, 2008 9:57 AMSubject: **SPAM** Re: AGI goals (was Re: Information theoretic approachesto AGI (was Re: [agi] The Necessity of Embodiment))Hi mark, I think the miscommunication is relatively simple... On Wed, Aug 27, 2008 at 10:14 PM, Mark Waser <[EMAIL PROTECTED]> wrote:Hi, I think that I'm missing some of your points . . . .Whatever good is, it cannot be something directly observable, or the AI will just wirehead itself (assuming it gets intelligent enough to do so, of course).I don't understand this unless you mean by "directly observable" that the definition is observable and changeable. If I define good as making allhumans happy without modifying them, how would the AI wirehead itself?What am I missing here?When I say "directly observable", I mean observable-by-sensation. "Making all humans happy" could not be directly observed unless the AI had sensors in the pleasure centers of all humans (in which case it would want to wirehead us). "Without modifying them" couldn't be directly observed even then. So, realistically, such a goal needs to be inferred from sensory data. Also, I should mention that the whole construction becomes irrelevant if we can logically describe the goal ahead of time. With the "make humans happy" example, something like my construction would be useful if we need to AI to *learn* what a human is and what happy is. (We then set up the pleasure in a way that would help the AI attach "goodness" to the right things.) If we are able to write out the definitions ahead of time, we can directly specify what goodness is instead. But, I think it is unrealistic to take that approach, since the definitions would be large and difficult....So, the AI needs to have a concept of external goodness, with a weak probabilistic correlation to its directly observable pleasure.I agree with the concept of external goodness but why does the correlationbetween external goodness and it's pleasure have to be low? Why can'texternal goodness directly cause pleasure? Clearly, it shouldn't believe that it's pleasure causes external goodness (that would be reversing cause and effect and an obvious logic error).The correlation needs to be fairly low to allow the concept of good to eventually split off of the concept of pleasure in the AI mind. The external goodness can't directly cause pleasure because it isn't directly detectable. Detection of goodness *through* inference *could* be taken to cause pleasure; but this wouldn't be much use, because the AI is already supposed to be maximizing goodness, not pleasure. Pleasure merely plays the role of offering "hints" about what things in the world might be good. Actually, I think the proper probabilistic construction might be a bit different than simply a "weak correlation"... for one thing, the probability that goodness causes pleasure shouldn't be set ahead of time. I'm thinking that likelihood would be more appropriate than probability... so that it is as if the AI is born with some evidence for the correlation that it cannot remember, but uses in reasoning (if you are familiar with the idea of "virtual evidence" that is what I am talking about).Mark P.S. I notice that several others answered your wirehead query so I won't belabor the point. :-) ----- Original Message ----- From: "Abram Demski" <[EMAIL PROTECTED]> To: <agi@v2.listbox.com> Sent: Wednesday, August 27, 2008 3:43 PM Subject: **SPAM** Re: AGI goals (was Re: Information theoretic approaches to AGI (was Re: [agi] The Necessity of Embodiment))Mark, The main motivation behind my setup was to avoid the wirehead scenario. That is why I make the explicit goodness/pleasure distinction. Whatever good is, it cannot be something directly observable, or the AI will just wirehead itself (assuming it gets intelligent enough to do so, of course). But, goodness cannot be completely unobservable, or the AI will have no idea what it should do. So, the AI needs to have a concept of external goodness, with a weak probabilistic correlation to its directly observable pleasure. That way, the system will go after pleasant things, but won't be able to fool itself with things that are maximally pleasant. For example, if it were to consider rewiring its visual circuits to see only skin-color, it would not like the idea, because it would know that such a move would make it less able to maximize goodness in general. (It would know that seeing only tan does not mean that the entire world is made of pure goodness.) An AI that was trying to maximize pleasure would see nothing wrong with self-stimulation of this sort. So, I think that pushing the problem of goal-setting back to pleasure-setting is very useful for avoiding certain types of undesirable behavior. By the way, where does this term "wireheading" come from? I assume from context that it simply means self-stimulation. -Abram Demski On Wed, Aug 27, 2008 at 2:58 PM, Mark Waser <[EMAIL PROTECTED]> wrote:Hi, A number of problems unfortunately . . . .-Learning is pleasurable.. . . . for humans. We can choose whether to make it so for machinesor not. Doing so would be equivalent to setting a goal of learning.-Other things may be pleasurable depending on what we initially wantthe AI to enjoy doing.See . . . all you've done here is pushed goal-setting to pleasure-setting . . . . = = = = = Further, if you judge goodness by pleasure, you'll probably create an AGI whose shortest path-to-goal is to wirehead the universe (which I consider to be a seriously suboptimal situation - YMMV). ----- Original Message ----- From: "Abram Demski" <[EMAIL PROTECTED]> To: <agi@v2.listbox.com> Sent: Wednesday, August 27, 2008 2:25 PM Subject: **SPAM** Re: AGI goals (was Re: Information theoretic approaches to AGI (was Re: [agi] The Necessity of Embodiment))Mark,OK, I take up the challenge. Here is a different set of goal-axioms:-"Good" is a property of some entities. -Maximize good in the world.-A more-good entity is usually more likely to cause goodness than aless-good entity. -A more-good entity is often more likely to cause pleasure than a less-good entity. -"Self" is the entity that causes my actions. -An entity with properties similar to "self" is more likely to be good. Pleasure, unlike goodness, is directly observable. It comes from many sources. For example: -Learning is pleasurable. -A full battery is pleasurable (if relevant). -Perhaps the color of human skin is pleasurable in and of itself. (More specifically, all skin colors of any existing race.) -Perhaps also the sound of a human voice is pleasurable.-Other things may be pleasurable depending on what we initially wantthe AI to enjoy doing. So, the definition if "good" is highly probabilistic, and the system's inferences about goodness will depend on its experiences; but pleasurecan be directly observed, and the pleasure-mechanisms remain fixed.On Wed, Aug 27, 2008 at 12:32 PM, Mark Waser <[EMAIL PROTECTED]>wrote:But, how does your description not correspond to giving the AGI thegoals of being helpful and not harmful? In other words, what moredoesit do than simply try for these? Does it pick goals randomly suchthat they conflict only minimally with these?Actually, my description gave the AGI four goals: be helpful, don'tbe harmful, learn, and keep moving. Learn, all by itself, is going to generate an infinite number of subgoals.Learning subgoals will be picked based upon what is most likely tolearn the most while not being harmful. (and, by the way, be helpful and learn should both generate aself-protection sub-goal in short order with procreation followingimmediately behind) Arguably, be helpful would generate all three of the other goals but learning and not being harmful without being helpful is a *much* bettergoal-set for a novice AI to prevent "accidents" when the AI thinksit is being helpful. In fact, I've been tempted at times to entirely drop the be helpful since the other two will eventually generate it with a lessened probability of trying-to-be-helpful accidents. Don't be harmful by itself will just turn the AI off. The trick is that there needs to be a balance between goals. Any single goal intelligence is likely to be lethal even if that goal is to help humanity. Learn, do no harm, help. Can anyone come up with a better set of goals?(and, once again, note that learn does *not* override the other two-- there is meant to be a balance between the three). ----- Original Message ----- From: "Abram Demski" <[EMAIL PROTECTED]> To: <agi@v2.listbox.com> Sent: Wednesday, August 27, 2008 11:52 AM Subject: **SPAM** Re: AGI goals (was Re: Information theoretic approaches to AGI (was Re: [agi] The Necessity of Embodiment))Mark, I agree that we are mired 5 steps before that; after all, AGI is not"solved" yet, and it is awfully hard to design prefab concepts ina knowledge representation we know nothing about! But, how does your description not correspond to giving the AGI thegoals of being helpful and not harmful? In other words, what moredoesit do than simply try for these? Does it pick goals randomly suchthat they conflict only minimally with these? --Abram On Wed, Aug 27, 2008 at 11:09 AM, Mark Waser <[EMAIL PROTECTED]> wrote:It is up to humans to define the goals of an AGI, so that it will do what we want it to do.Why must we define the goals of an AGI? What would be wrong withsetting it off with strong incentives to be helpful, even stronger incentives to not beharmful, and let it chart it's own course based upon the vagariesof the world? Let it's only hard-coded goal be to keep it's satisfaction above a certain level with helpful actions increasing satisfaction, harmful actions heavily decreasing satisfaction; learning increasing satisfaction, andsatisfaction naturally decaying over time so as to promote action. . . . Seems to me that humans are pretty much coded that way (with evolution'sadditional incentives of self-defense and procreation). The realtrick of the matter is defining helpful and harmful clearly but everyone is still mired five steps before that. ----- Original Message ----- From: Matt Mahoney To: agi@v2.listbox.com Sent: Wednesday, August 27, 2008 10:52 AM Subject: AGI goals (was Re: Information theoretic approaches to AGI (was Re: [agi] The Necessity of Embodiment)) An AGI will not design its goals. It is up to humans to define the goals of an AGI, so that it will do what we want it to do.Unfortunately, this is a problem. We may or may not be successfulin programming the goals of AGI to satisfy human goals. If we are not successful, then AGI will be useless at best and dangerous at worst. If weare successful, then we are doomed because human goals evolved inaprimitive environment to maximize reproductive success and not inan environment where advanced technology can give us whatever we want. AGI will allow us to connect our brains to simulated worlds with magic genies, or worse, allow us to directly reprogram our brains to alter our memories, goals, and thought processes. All rational goal-seeking agents must have a mental state of maximum utility where any thought or perception would be unpleasant because it would result in a different state. -- Matt Mahoney, [EMAIL PROTECTED] ----- Original Message ---- From: Valentina Poletti <[EMAIL PROTECTED]> To: agi@v2.listbox.com Sent: Tuesday, August 26, 2008 11:34:56 AM Subject: Re: Information theoretic approaches to AGI (was Re: [agi] The Necessity of Embodiment) Thanks very much for the info. I found those articles very interesting. Actually though this is not quite what I had in mind with the term information-theoretic approach. I wasn't very specific, my bad. What I am looking for is a a theory behind the actual R itself. These approaches (correnct me if I'm wrong) give an r-function for granted and work fromthat. In real life that is not the case though. What I'm lookingfor is howthe AGI will create that function. Because the AGI is created byhumans,some sort of direction will be given by the humans creating them.What kind of direction, in mathematical terms, is my question. In other words I'm looking for a way to mathematically define how the AGI will mathematically define its goals. Valentina On 8/23/08, Matt Mahoney <[EMAIL PROTECTED]> wrote:Valentina Poletti <[EMAIL PROTECTED]> wrote: > I was wondering why no-one had brought up the > > information-theoretic > aspect of this yet.It has been studied. For example, Hutter proved that the optimalstrategy of a rational goal seeking agent in an unknown computable environment isAIXI: to guess that the environment is simulated by the shortestprogram consistent with observation so far [1]. Legg and Hutter also propose as a measure of universal intelligence the expected reward over a Solomonoff distribution of environments [2]. These have profound impacts on AGI design. First, AIXI is (provably) not computable, which means there is no easy shortcut to AGI. Second, universalintelligence is not computable because it requires testing in aninfinite number of environments. Since there is no other well accepted test of intelligence above human level, it casts doubt on the main premise of thesingularity: that if humans can create agents with greater thanhuman intelligence, then so can they. Prediction is central to intelligence, as I argue in [3]. Legg proved in [4] that there is no elegant theory of prediction. Predicting all environments up to a given level of Kolmogorov complexity requires a predictor with at least the same level of complexity. Furthermore, above a small level of complexity, such predictors cannot be proven because of Godel incompleteness. Prediction must therefore be an experimental science. There is currently no software or mathematical model of non-evolutionary recursive self improvement, even for very restricted or simple definitionsof intelligence. Without a model you don't have friendly AI; youhave accelerated evolution with AIs competing for resources. References 1. Hutter, Marcus (2003), "A Gentle Introduction to The Universal Algorithmic Agent {AIXI}",in Artificial General Intelligence, B. Goertzel and C. Pennachineds., Springer. http://www.idsia.ch/~marcus/ai/aixigentle.htm 2. Legg, Shane, and Marcus Hutter (2006), A Formal Measure of Machine Intelligence, Proc. Annual machine learning conference of Belgium and The Netherlands (Benelearn-2006). Ghent, 2006. http://www.vetta.org/documents/ui_benelearn.pdf 3. http://cs.fit.edu/~mmahoney/compression/rationale.html4. Legg, Shane, (2006), Is There an Elegant Universal Theory ofPrediction?, Technical Report IDSIA-12-06, IDSIA / USI-SUPSI, Dalle Molle Institute for Artificial Intelligence, Galleria 2, 6928 Manno, Switzerland. http://www.vetta.org/documents/IDSIA-12-06-1.pdf -- Matt Mahoney, [EMAIL PROTECTED] ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com-- A true friend stabs you in the front. - O. Wilde Einstein once thought he was wrong; then he discovered he was wrong. For every complex problem, there is an answer which is short, simple and wrong. - H.L. Mencken ________________________________ agi | Archives | Modify Your Subscription ________________________________ agi | Archives | Modify Your Subscription ________________________________ agi | Archives | Modify Your Subscription------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?& Powered by Listbox: http://www.listbox.com------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/Modify Your Subscription: https://www.listbox.com/member/?&Powered by Listbox: http://www.listbox.com
------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com