If I were to make a short list of the most important human qualities—

—and yes, this is a fool's errand, because human nature is immensely complicated, and we don't even notice all the tiny tweaks that fine-tune our moral categories, and who knows how our attractors would change shape if we eliminated a single human emotion—

—but even so, if I had to point to just a few things and say, "If you lose just one of these things, you lose most of the expected value of the Future; but conversely if an alien species independently evolved just these few things, we might even want to be friends"—

—then the top three items on the list would be sympathy, boredom and consciousness.

Boredom is a subtle-splendored thing.  You wouldn't want to get bored with breathing, for example—even though it's the same motions over and over and over and over again for minutes and hours and years and decades.

Now I know some of you out there are thinking, "Actually, I'm quite bored with breathing and I wish I didn't have to," but then you wouldn't want to get bored with switching transistors.

According to the human value of boredom, some things are allowed to be highly repetitive without being boring—like obeying the same laws of physics every day.

Conversely, other repetitions are supposed to be boring, like playing the same level of Super Mario Brothers over and over and over again until the end of time.  And let us note that if the pixels in the game level have a slightly different color each time, that is not sufficient to prevent it from being "the same damn thing, over and over and over again".

Once you take a closer look, it turns out that boredom is quite interesting.

One of the key elements of boredom was suggested in "Complex Novelty":  If your activity isn't teaching you insights you didn't already know, then it is non-novel, therefore old, therefore boring.

But this doesn't quite cover the distinction.  Is breathing teaching you anything?  Probably not at this moment, but you wouldn't want to stop breathing.  Maybe you'd want to stop noticing your breathing, which you'll do as soon as I stop drawing your attention to it.

I'd suggest that the repetitive activities which are allowed to not be boring fall into two categories:

  • Things so extremely low-level, or with such a small volume of possibilities, that you couldn't avoid repeating them even if you tried; but which are required to support other non-boring activities.  You know, like breathing, or obeying the laws of physics, or cell division—that sort of thing.
  • Things so high-level that their "stability" still implies an immense space of specific possibilities, yet which are tied up with our identity or our values.  Like thinking, for example.

Let me talk about that second category:

Suppose you were unraveling the true laws of physics and discovering all sorts of neat stuff you hadn't known before... when suddenly you got bored with "changing your beliefs based on observation".  You are sick of anything resembling "Bayesian updating"—it feels like playing the same video game over and over.  Instead you decide to believe anything said on 4chan.

Or to put it another way, suppose that you were something like a sentient chessplayer—a sentient version of Deep Blue.  Like a modern human, you have no introspective access to your own algorithms.  Each chess game appears different—you play new opponents and steer into new positions, composing new strategies, avoiding new enemy gambits.  You are content, and not at all bored; you never appear to yourself to be doing the same thing twice—it's a different chess game each time.

But now, suddenly, you gain access to, and understanding of, your own chess-playing program.  Not just the raw code; you can monitor its execution.  You can see that it's actually the same damn code, doing the same damn thing, over and over and over again.  Run the same damn position evaluator.  Run the same damn sorting algorithm to order the branches.  Pick the top branch, again.  Extend it one position forward, again.  Call the same damn subroutine and start over.

I have a small unreasonable fear, somewhere in the back of my mind, that if I ever do fully understand the algorithms of intelligence, it will destroy all remaining novelty—no matter what new situation I encounter, I'll know I can solve it just by being intelligent, the same damn thing over and over.  All novelty will be used up, all existence will become boring, the remaining differences no more important than shades of pixels in a video game.  Other beings will go about in blissful unawareness, having been steered away from studying this forbidden cognitive science.  But I, having already thrown myself on the grenade of AI, will face a choice between eternal boredom, or excision of my forbidden knowledge and all the memories leading up to it (thereby destroying my existence as Eliezer, more or less).

Now this, mind you, is not my predictive line of maximum probability.  To understand abstractly what rough sort of work the brain is doing, doesn't let you monitor its detailed execution as a boring repetition.  I already know about Bayesian updating, yet I haven't become bored with the act of learning.  And a self-editing mind can quite reasonably exclude certain levels of introspection from boredom, just like breathing can be legitimately excluded from boredom.  (Maybe these top-level cognitive algorithms ought also to be excluded from perception—if something is stable, why bother seeing it all the time?)

No, it's just a cute little nightmare, which I thought made a nice illustration of this proposed principle:

That the very top-level things (like Bayesian updating, or attaching value to sentient minds rather than paperclips) and the very low-level things (like breathing, or switching transistors) are the things we shouldn't get bored with.  And the mid-level things between, are where we should seek novelty.  (To a first approximation, the novel is the inverse of the learned; it's something with a learnable element not yet covered by previous insights.)

Now this is probably not exactly how our current emotional circuitry of boredom works.  That, I expect, would be hardwired relative to various sensory-level definitions of predictability, surprisingness, repetition, attentional salience, and perceived effortfulness.

But this is Fun Theory, so we are mainly concerned with how boredom should work in the long run.

Humanity acquired boredom the same way as we acquired the rest of our emotions: the godshatter idiom whereby evolution's instrumental policies became our own terminal values, pursued for their own sake: sex is fun even if you use birth control.  Evolved aliens might, or might not, acquire roughly the same boredom in roughly the same way.

Do not give into the temptation of universalizing anthropomorphic values, and think:  "But any rational agent, regardless of its utility function, will face the exploration/exploitation tradeoff, and will therefore occasionally get bored with exploiting, and go exploring."

Our emotion of boredom is a way of exploring, but not the only way for an ideal optimizing agent.

The idea of a steady trickle of mid-level novelty is a human terminal value, not something we do for the sake of something else.  Evolution might have originally given it to us in order to have us explore as well as exploit.  But now we explore for its own sake.  That steady trickle of novelty is a terminal value to us; it is not the most efficient instrumental method for exploring and exploiting.

Suppose you were dealing with something like an expected paperclip maximizer—something that might use quite complicated instrumental policies, but in the service of a utility function that we would regard as simple, with a single term compactly defined.

Then I would expect the exploration/exploitation tradeoff to go something like as follows:  The paperclip maximizer would assign some resources to cognition that searched for more efficient ways to make paperclips, or harvest resources from stars.  Other resources would be devoted to the actual harvesting and paperclip-making.  (The paperclip-making might not start until after a long phase of harvesting.)  At every point, the most efficient method yet discovered—for resource-harvesting, or paperclip-making—would be used, over and over and over again.  It wouldn't be boring, just maximally instrumentally efficient.

In the beginning, lots of resources would go into preparing for efficient work over the rest of time.  But as cognitive resources yielded diminishing returns in the abstract search for efficiency improvements, less and less time would be spent thinking, and more and more time spent creating paperclips.  By whatever the most efficient known method, over and over and over again.

(Do human beings get less easily bored as we grow older, more tolerant of repetition, because any further discoveries are less valuable, because we have less time left to exploit them?)

If we run into aliens who don't share our version of boredom—a steady trickle of mid-level novelty as a terminal preference—then perhaps every alien throughout their civilization will just be playing the most exciting level of the most exciting video game ever discovered, over and over and over again.  Maybe with nonsentient AIs taking on the drudgework of searching for a more exciting video game.  After all, without an inherent preference for novelty, exploratory attempts will usually have less expected value than exploiting the best policy previously encountered.  And that's if you explore by trial at all, as opposed to using more abstract and efficient thinking.

Or if the aliens are rendered non-bored by seeing pixels of a slightly different shade—if their definition of sameness is more specific than ours, and their boredom less general—then from our perspective, most of their civilization will be doing the human::same thing over and over again, and hence, be very human::boring.

Or maybe if the aliens have no fear of life becoming too simple and repetitive, they'll just collapse themselves into orgasmium.

And if our version of boredom is less strict than that of the aliens, maybe they'd take one look at one day in the life of one member of our civilization, and never bother looking at the rest of us.  From our perspective, their civilization would be needlessly chaotic, and so entropic, lower in what we regard as quality; they wouldn't play the same game for long enough to get good at it.

But if our versions of boredom are similar enough —terminal preference for a stream of mid-level novelty defined relative to learning insights not previously possessed—then we might find our civilizations mutually worthy of tourism.  Each new piece of alien art would strike us as lawfully creative, high-quality according to a recognizable criterion, yet not like the other art we've already seen.

It is one of the things that would make our two species ramen rather than varelse, to invoke the Hierarchy of Exclusion.  And I've never seen anyone define those two terms well, including Orson Scott Card who invented them; but it might be something like "aliens you can get along with, versus aliens for which there is no reason to bother trying".

 

Part of The Fun Theory Sequence

Next post: "Sympathetic Minds"

Previous post: "Dunbar's Function"

101 comments, sorted by
magical algorithm
Highlighting new comments since Today at 3:23 PM
Select new highlight date
Moderation Guidelines: Reign of Terror - I delete anything I judge to be annoying or counterproductiveexpand_more

Are you sure this isn't the Eliezer concept of boring, instead of the human concept? There seem to be quite a few humans who are happy to keep winning using the same approach day after day year after year. They keep getting paid well, getting social status, money, sex, etc. To the extent they want novelty it is because such novelty is a sign of social status - a new car every year, a new girl every month, a promotion every two years, etc. It is not because they expect or want to learn something from it.

sex

Maybe for some people more shallow forms of novelty suffice e.g. sex with new women.

Robin, do they eat the same foods every day? Drive to the same places every day? Buy the same things every time they shop? Have sex in the same position every time? Watch the same movie each time they go to the theater? Since you're standing back, you see them at a level of abstraction from which their life looks mostly "the same" to you, but I doubt they're playing the same level of the same video game over and over again "every time they sit at the computer".

An easy way to differentiate the two kinds for those who like games is: People who can play Mario Kart thousands of times and have a lot of fun. People who must play the new final fantasy.

There are those who do both, and those who only enjoy games designed for doing the same thing, better and better, every five minutes.

Compare the complexity of handball with the complexity of bowling.

Maybe bowling is Eliezer::boring but it isn't boring for a lot of people.

It would be a waste of energetic resources if FAI gave those people Final Fantasy 777 instead of just letting them play Mario Kart 9.

The tough question then becomes: Are those of us who enjoy Mario Kart and bowling willing to concede the kind of fun that the Eliezer Final Fantasy, pro-increasing-rate-of-complexity find desirable? They will be consuming soooo much energy for their fun.

Isn't it fair that we share the pie half in half, and they consume theirs exponencially, while we enjoy for subjectively longer?

People who can play Mario Kart thousands of times and have a lot of fun. People who must play the new final fantasy.

Do you really play Mario Kart thousands of times because you love repeating the same thing? Or do you love it because you have a finer eye for small detail than the new FF player, and so are noticing new novelty each time you play? I know I can watch "King Kong" or "Halloween" over and over again partly because I notice something new each time I watch those films.

That being said, I think a proper fun theory would probably have some sort of error bars around the level of boredom that is acceptable. In other words a creature that gets bored x% less easily than the median human is might still be a worthwhile creature, but creating a creature x+1% less easily bored would be bad (unless such a creature is instrumentally useful, obviously).

Another question came to my mind while thinking about this today. How often do you play Mario Kart alone vs. with friends? Adding social interaction to the game vastly increases its complexity. Probably part of the reason it's more enduring than the FF games is that most of those games are single player, so the complexity is limited by your inability to play against other humans. Good multiplayer games are probably so replayable partly because they are venues for social interaction, which is a very, very complex form of novelty.

Perhaps consistent with Robin's comment, I don't see any reason not to "collapse ... into orgasmium," at least after our other utilitarian obligations (e.g., preventing suffering by others in the multiverse) are completed.

... and if you consider the class of "subconscious activities done in order to reach another goal", you'll see that if covers both "low-level" stuff like breathing, and "high-level" stuff like thinking (or at least, the mechanics of thinking - retrieving memories, updating beliefs, etc.). So you get one category instead of two.

Another way to get one category instead of two... Think of boredom as a signal of not incorporating new, useful physical info. Breathing and thinking (usefully) are not boring because those processes facilitate the body's exploitation or incorporation of physical info. In other words, boredom arises from a lack of novelty on the level of physics, though the process of breathing may seem repetitive or non-novel on the level of biomechanics.

I'm not sure breathing needs a special exclusion from boredom, for the same reasons people don't get bored from jumping in Mario: we don't get bored with something if it's only a mean to something else.

This blog post talks about that a bit.

You could also say that you only get bored with conscious activities, and that breathing is unconscious, just as jumping is in mario. I'm not sure which explanation is the best way of putting things "not boring because unconscious" and "not boring because it's a means to a goal".

But anyway, those explanations seem to fit reality closer than "Things so extremely low-level, or with such a small volume of possibilities, that you couldn't avoid repeating them even if you tried; but which are required to support other non-boring activities." (I don't think "low-level" and "small volume of possibilities" are necessary conditions. Some pretty high-level and complex things like driving a car can still be non-boring if it's unconscious / used with another goal in mind.

I always think of boredom as the chorus of brain agents crying out that 'whatever you are doing right now, it has not recently helped ME to achieve MY goals'. Boredom is the emotional reward circuit to keep us rotating contributions towards our various desired goals. It also applies even if we are working on a specific goal, but not making progress.

I think as we age our goals get fewer, narrower and a bit less vocal about needing pleasing, thus boredom recedes. In particular, we accept fewer goals that are novel, which means the goals we do have tend to be more practical with existing known methods of achieving them such that we are more often making progress.

Interest in previously boring (due to repetition) things regenerates over time. Eating strawberries every six months may not be as good as the first time (although nostalgia may make it better), but it's not obvious that it declines in utility.

We may also actively value non-boredom in some mid-level contexts, e.g. in sexual fidelity, or for desires that we consider central to our identity/narratives.

"Eating strawberries every six months may not be as good as the first time (although nostalgia may make it better), but it's not obvious that it declines in utility."

Isn't "not being as good" just what "declines in utility" means?

Read that as "continues to decline in utility every six months."

Maybe the problem is trying to assign utility to "eat strawberries" rather than to the whole timeless state "ate strawberries Tuesday, blueberries Wednesday, bananas Thursday" etc.

We do seem to run into some weird problems if we say that the marginal utility of strawberries declines each time you eat a strawberry... though maybe we could say that this is true, everything else held constant or something like that.

Maybe they meant that it doesn't continue getting less and less good. I dunno.

Ben Goertzel's patternist philosophy of mind is suggestive here. It is boring when the same patterns repeat themselves. It's not about whether the pixels change or stay the same. It's about whether you can detect a unique pattern each time.

This may explain Eliezer's preference for the mid-level. If you are too specific, you don't see any patterns. If you are too general you miss many lower-level patterns. And also the balance between triviality and intractability of problems. If it's trivial you already have the pattern, if it is intractable you can't see any.

Intelligence may be the same algorithm each time it is applied, but as long as it generates/detects new patterns in the problems it encounters, there is no cause for boredom. Intelligent people get bored more quickly, because they can assimilate more patterns per unit time. But I like to think we all experience the same 'subjective novelty-time', so there is no need to make yourself stupid to extend your experience of novelty.

"it's easier to become bored with the game as a whole than with that particular part of it."

That has to do with our limited working memory capacity. When we conceive of "the game as a whole" we don't download the whole game into working memory. There is no space. Since the game really isn't in working memory, the conscious you does not detect any pattern. Playing the game though, is fun. Not because jumping in mario is instrumental to scoring points, but because arriving at a particular goal through particular means constitutes a pattern. Of course, mere permutations on fragments of the journey aren't as exciting techniques within strategies within stories because well, 'permutation of elements in a set' is a pretty common pattern...

What about detecting patterns in the clouds, in coincidences, in astrology, in pi, in the noise? Very quickly boring because when we go meta, we see there are no patterns linking these disparate atomic patterns.

So the recipe for interestingness is: objects, patterns, recursion. This is best demonstrated in axiomatic systems constructed by mathematicians. But it is also the same recipe with an interesting life in general.

Emile and Caledonian are right. Eliezer should've defined exceptions to boredom instead (and more simply) as "activities that work towards your goal". Those are exempt of boredom and can even be quite fun. No need to distinguish between high, low and mid-level.

The page at Lostgarden that Emile linked to is a bit long, so I'll try to summarize the proposed theory of fun, with some of my own conclusions:

You naturally find activities that provide you with valuable insights fun (the "aha!" moment, or "fun"). Tolerance to repetition (actually, finding a repetitive act "fun" as well) is roughly proportional to your expectation of how it will provide you with a future fun moment.

There are terminal fun moments. Driving a car is repetitive, but at high speeds adrenaline makes up for that. Seeing Mario jump for the first time is fun (you' found a way of impacting the world [the computer screen] through your own action). I'm sure you can think of other examples of activities chemically wired to being fun, of course ;)

Working in the financial business might be repetitive and boring (or at least it seems that way at first), but if it yields good paychecks, which give you the opportunity to buy nice things, gain social status, etc, you'll keep doing it.

Jumping in Mario is repetitive, and if jumping didn't do anything, you'd never touch that button again after 10 jumps (more or less). But early on it allows you to get to high platforms, which kinda "rekindles" the jumping activity, and the expectation that it will be useful in the future/yield more fun. Moving from platform to platform gets repetitive, unless it serves yet another purpose.

(The above is all described in Lostgarden and forms the basis of their theory of fun, and how to build a fun game. Following are some of my own conclusions.)

The highest goal of all is usually to "beat the game"/"explore the game world"/"have the highest score", and you set it upon yourself naturally. This is like the goal of jumping over a ledge, even if you don't know what's beyond it (in the Mario world). You ran out of goals so you're exploring, which usually means thinking up an "exploratory goal", ie, trying something new.

You can say that finding goals is fun in itself. If you start in a blank state, nothing will seem fun at first. You might as well just sit down and whither away! So a good strategy is to set yourself a modest goal (an exploratory goal), and the total fun had will be greater than the fun you assigned earlier to the goal in itself, which might be marginal. A more concrete example: The fun in reading "You win!" is marginal, but you play through Mario just to read those words. So I guess that the journey is more important than getting to the end.

Few people become bored with jumping in SMB because 1) becoming skilled at it is quite hard, 2) it's used to accomplish specific tasks and is quite useful in that context, 3) it's easier to become bored with the game as a whole than with that particular part of it.

Zaphod: kind of funny, given the many foreign words in English - ennui, weltschmerz, melancholy etc.

EY: (Do human beings get less easily bored as we grow older, more tolerant of repetition, because any further discoveries are less valuable, because we have less time left to exploit them?)

Anonymous: For a 3 year old, every day is like your first day in Paris.

Sample of one but my personal experience in my seventh decade is that is just gets damn hard to find interesting new hings, especially easily accessible interesting new things.

My tolerance of boredom actually seems far lower than it used to be but I have to work a lot harder to access good new stuff.

The argument that you would loose interest if you could explain boredom away, which is what I have to conclude from your stance:

All novelty will be used up, all existence will become boring, the remaining differences no more important than shades of pixels in a video game.

seems a bit thin to me. Does a magician loose interest because he knows every single trick that wows the audience?

Does the musician who has spent a lifetime studying the intricacies of Bach's partita No 2 loose interest just because he can deconstruct it entirely?

Douglas Hoefstadter expressed a similar concern a decade or so ago when he learnt of some "computer program" able to "generate Mozart music better then Mozart himself" only to recant a bit later when facing the truism that there is more to the entity than the sum of its parts.

I do not know that we will someday be able to "explain magic away", and if that makes me irrational (and no, I don't need to bring any kind of god in the picture: I'm perfectly happy being goddless and irrational :) so be it.

the time when we gain enough experience to find all boring will be near to the time we can eliminate boredom as an emotion.

If you still wish to feel novelty you are free to wallow in ignorance; all will be new to you then. if you wish to move forward then remember self-modification to be an option.

If there's still a forward to move in then it would seem we don't have enough experience to find all boring.

If most of us have a "terminal preference for a stream of mid-level novelty" those with ADD/ADHD find that unstimulating. They require a stream of high level novelty or, at least, a much faster stream of mid-level. See ADHD posts on ThePowerOfBoredom.com

Robin: It is not because they expect or want to learn something from it.

A major component of fun in video games is the emotional reward when the brain learnt something; that probably generalizes to why we find a lot of activities enjoyable, even though we might not label them as "learning" which is often associated to "memorizing useless facts because you're forced to".

Right. Football players don't think of themselves as "learning" when they perfect their field goal kicks, but neurologically that is exactly what they are doing.

You can't just be "intelligent over and over", because discovery and insight are essentially random processes. You can't just find insight, you have to look for it, in the same way that evolution searches the option space.

Yes, you can always have better heuristics or search algorithms. But those heuristics are not themselves intelligence. And there are always new heuristics to discover...

So, I don't think mere insight into the process of intelligence would allow you to be bored, since the things to be discovered by intelligence would still be "out there" rather than "in here", if you get my drift. And it's those subjects of discovery that are the intended targets of novelty and interest, anyway.

because discovery and insight are essentially random processes.

In that case, roll a die over and over, and you'll have to worry about boredom.

But those heuristics are not themselves intelligence.

True, but one of those heuristics is you, and you can't stop doing that.

Of course, I don't think any of us think this is very likely to be a problem. I'm basically just playing devil's advocate.

I'm pretty certain you'd have to do some significant self-modification to understand it all on a level that makes it boring, and at that point, you could just self-modify so that it isn't boring.

Robin, I suspect that despite how it may look from a high level, the lives of most of the people you refer to probably do differ enough from year to year that they will in fact have new experiences and learn something new, and that they would in fact find it unbearable if their world were so static as to come even a little close to being video game repetitive.

That said, I would agree that many people seem not to act day-to-day as if they put a premium on Eliezer-style novelty, but that seems like it could be better explained by Eliezer's boredom being a FAR value than by the concept being specific to Eliezer :-)

I wonder what kinds of boredom, unfamiliar to us resource-limited kind, do billionaires suffer from. It seems it's the same things all over and over again, only on a different scale - probably the closest to the boredom experienced widely in the Culture.

Science, it seems, is ultimately the only reliable escape from boredom. Until everything is solved - any estimation when we might call the project called 'Science' "Done!"?

"All science is either physics or stamp collecting." -Ernest Rutherford

If you get bored of being a billionaire, give away all your money.

In fact, give it to me.

Cool stuff. It's philosophy for our present times. I like all the cultural references.

Now this is probably not exactly how our current emotional circuitry of boredom works. That, I expect, would be hardwired relative to various sensory-level definitions of predictability, surprisingness, repetition, attentional salience, and perceived effortfulness.

It is interesting to read this after reading Yvain's classic essay on wanting, liking, and approving. In Yvain's terms, the value of boredom could be construed as an instance where our "wanting, "liking," and "approving" systems are in relative harmony.

Couched in Yvain's terms, Eliezer praises boredom because he and most other people approve very strongly of the things that boredom motivates us to do, such as explore, engage in personal growth, learn new things, seek out new experiences, etc. In addition to this, the "wanting" and "liking" aspects of our characters also motivate us to engage in these positive behaviors, because as they become used to certain experiences they start to like them less and want to do them less frequently. This means that in addition to approving of seeking out new experiences, we also want to and like to.

But this is Fun Theory, so we are mainly concerned with how boredom should work in the long run.

Based on Yvain's work, it would seem that the way to create the improved boredom of the future would be to greatly enhance the power of the "approving" parts of our minds, so that they can more easily override the "wanting" and "liking" parts. If done properly this would give us free reign to improve the "liking" parts of our minds so that we can feel wonderful and happy all the time without fear of this causing us to lose our motivation to do awesome things with our lives. The people of the future could engage in all sorts of complex and challenging activities and feel like orgasmium while they're doing them.

This reminds me a little of how Eliezer speculates elsewhere that we might ultimately find a way to improve pain so that the more negative aspects of it are removed without getting rid of other facets of it that we approve of, such as enjoying Serious Stories.

What bores me is that we live in a binary universe. Sort of limits your options.

An appropriate post : I've come to find EY's posts very boring. Subtle, intelligent, all that, sure. A mind far finer than my own, sure. But it never gets anywhere, never goes anywhere. He spends so much time posting he's clearly not moving AI forward. His book is still out of sight, two years down the line. I can understand the main thrust of his posts, and the comments, if I invest enough, my intelligence and knowledge are just about up to that. But why bother ? It's sterile. Boredom = sterility. As for Robin's comment, which is pertinent and bears on the real world of lived emotions, the connection is that boredom is not a result of what you are doing, it's a result of what you're not doing. Think about it.

You get boredom when you are attempting to efficiently explore a mostly-unexplored search space - or at least you get a tendency to avoid repeatedly sampling the same region of the search space - which is boredom's primary behavioural manifestation.

In the example in the post - of an optimiser that doesn't get bored - that happens because the search space it is exploring has become exhausted.

That is simply a property of the particular example which was selected. It isn't a general property of efficient optimisers and it doesn't mean that efficient optimisers don't exhibit boredom. They do exhibit boredom - when exploring mostly-unexplored search spaces.

IMO, boredom is best seen as being a universal instrumental value - and not as an unfortunate result of "universalizing anthropomorphic values".

Update 2011-07-27: Yudkowsky responds to roughly this point - 39 minutes in here. He claims

If you look at an ideal Bayesian decision maker as ask: "what kind of boredom is a convergent instrumental value?" then there is a convergent solution to the exploration-exploitation tradeoff and that solution leads to what we would regard as a boring, worthless, valueless future.

He doesn't give any references, or much of a supporting argument - it is more of an assertion. Maybe he thinks the material about paperclips in this post is sufficiently convincing. Our civilisation maximises entropy - not paperclips - which hardly seems much more interesting. Also, it isn't clear that Yudkowsky realises the extent to which boredom is implemented an instrumental value in modern humans. Nature can't easily wire in some kind of universal "boredom rate" - since boredom is task-specific (we mostly don't get bored of sex and food) and context specific. That's not to say it is entirely instrumental - but it is partly instrumental in humans.

Yudkowsky then goes on immediately to say:

this is why we have to solve the Friendly AI problem

...so it seems to be a significant point.

My position is that we had better wind up approximating the instrumental value of boredom (which we probably do pretty well today anyway - by the wonder of natural selection) - or we are likely to be building a rather screwed-up civilisation. There is no good reason why this would lead to a "worthless, valueless future" - which is why Yudkowsky fails to provide one.

Our civilisation maximises entropy - not paperclips - which hardly seems much more interesting.

No it doesn't. It tries to maximize fun. It might maximize entropy as a side effect, but saying that we act to maximize entropy is as ludicrous as saying cows act to maximize atmospheric methane content. You're confusing a side-effect with the real goal.

A constant theme I've noticed in your posts is that you take some (usually evolutionary) trend that is occurring in our society or history and then act as if that trend is an actual conscious goal of human beings, and life. You then exhibit confusion when someone makes a statement that their real conscious goals are in opposition to this trend, and tell them that the fact that this trend is occurring means they must really want it to occur, and that it is part of their real goals. This demonstrates rather confused thinking, both on how human minds work, an on what a "goal" really is.

Scientists often metaphorically describe trends as having goals and "maximizing" things, you seem to have taken these metaphors excessively literally and act like these trends literally have goals and literally maximize things. Terms like "goals" and "maximization" only refer, in the literal sense, to the computations of consequentialist thinking beings (consequentialist in this case meaning a being that can forecast the future, not the moral theory). A goal is a forecast of the future a consequentialist assigns favorable utility to. Maximizing refers to a consequentialist that values a certain property in the future so much it assigns very favorable utility to increasing it as much as possible. This is the only appropriate time to literally use the terms "goal" and "maximize," all other times are metaphorical.

Evolution and other trends do not literally maximize anything. They have no goals. It is just sometimes useful to metaphorically pretend they do because it makes thinking about it easier for human brains, which are better at modeling other consequentialist than they are at modeling abstract trends. Any claim that evolution has a goal in a non-metaphorical sense is a blatant falsehood. Until you realize this fact you will fail to understand virtually everything that Eliezer is talking about.

There is no good reason why this would lead to a "worthless, valueless future" - which is why Yudkowsky fails to provide one.

Yes there is. Such a future would be boring. It is bad for things to be boring and good for things to be interesting, so such a future would be bad. And don't ask me why boringness is bad, that's like asking why water is H2O. You are asking Eliezer to provide some meaning of good seperate from things like truth, fun, novelty, life, etc, when he has clearly explained that there is no meaning of good seperate from those things. Those things are good, so a world where they don't exist, or are reduced, would be bad, full stop.

No it doesn't. It tries to maximize fun. It might maximize entropy as a side effect, but saying that we act to maximize entropy is as ludicrous as saying cows act to maximize atmospheric methane content. You're confusing a side-effect with the real goal.

Really? How do you know that? Are plants trying to maximise "fun"? Is "fun" even a measurable quantity?

If "fun" is being maximised, why is there so much suffering in the world? If two systems are in contention, is it really the one that is having the most fun that will win? The "fun-as-maximand" theory seems trivially refuted by the facts.

"Fun" - if we are trying to treat the concept seriously - is better characterised as the proxy that brains use for the inclusive fitness of their associated organism.

There's a scientific literature on the subject of what God's utility function is. Entire books have been written about the topic. I'm familiar with this literature, are you?

Terms like "goals" and "maximization" only refer, in the literal sense, to the computations of consequentialist thinking beings (consequentialist in this case meaning a being that can forecast the future, not the moral theory). A goal is a forecast of the future a consequentialist assigns favorable utility to. Maximizing refers to a consequentialist that values a certain property in the future so much it assigns very favorable utility to increasing it as much as possible. This is the only appropriate time to literally use the terms "goal" and "maximize," all other times are metaphorical.

We had better talk about "optimization" then, or we will talk past each other.

Evolution and other trends do not literally maximize anything.

Really? How do you know that? Evolution is a gigantic optimization process with a maximand. You claimed above that it is "fun" - and my claim is that it is entropy. As I say, there's a substantial scientific literature on the topic - have you looked at it?

Evolution is a gigantic optimization process with a maximand

Success for the fox is failure for the rabbit; success for the rabbit is failure for the fox. What is the maximand?

So: entropy, as far as we can tell. See the works of Dewar, referenced here. Or for a popular version, try: Whitfield, John Survival of the Likeliest? for a popular version from someone other than me.

OTOH, as rabbits become better fox-evaders, foxes become better rabbit-hunters. If there exists some thing X that fox-evasion and rabbit-hunting have in common, it's possible (I would even say likely) that X is increasing throughout this process.

Increasing != maximising, though. Methane is increasing in both cases - but evolution doesn't maximise methane production.

Not sure why it's relevant, but certainly true.

Really? How do you know that?

You said: "Our civilisation maximises entropy." Our civilization consists of all the humans in the world. When you're asking what our civilization is trying to maximize you're asking what the humans of the world are trying to maximize. Humans try to do things they enjoy, things that are fun. Therefore our civilization tries to maximize fun.

I know that because that's basic human psychology 101. Humans want to be happy and have fulfilled preferences.

Are plants trying to maximise "fun"?

We're talking about our civilization. In other words, all the humans in the world. Plants aren't human, so whether they maximize fun is irrelevant. I suppose if you regarded human tools and artifacts as part of our civilization then agricultural plants could be regarded as part of it. But they aren't the part of our civilization that makes decisions on what to maximize, humans are.

Plants aren't trying to maximize anything. They're plants, they don't have minds. If I was to use the word maximize as liberally as you do I could actually argue that agricultural plants do try to maximize fun, because humans grow them for the purpose of eating, and eating is fun. But that wouldn't be strictly accurate, plants just execute their genetically coded behaviors, any purpose they have is really the purpose of the consequentalist minds that grow them, not of the plants. Saying that agricultural plants have any purpose at all is the mind-projection fallacy.

If "fun" is being maximised, why is there so much suffering in the world?

Because some humans are selfish and try to maximize their own fun at the expense of the fun of others. And sometimes we make big mistakes when trying to make the world more fun. But still, most of the time we try to work together to have fun. We aren't that good at it yet, but we're trying and keep improving. The world is getting progressively more fun.

If you systems are in contention, is it really the one that is having the most fun that will win?

Yes. Humans who are enjoying life the most are generally regarded as being more successful at life than humans who are not. This is a basic and easily observable fact.

The "fun-as-maximand" theory seems trivially refuted by the facts.

It's easily confirmed by the facts. As humans have grown richer and more technologically advanced they have devoted more and more of their resources to having fun. Look at the existence of places like Disneyworld for evidence.

"Fun" - if we are trying to treat the concept seriously - is better characterised as the proxy that brains use for the inclusive fitness of their associated organism.

No it isn't. Brains don't care about inclusive genetic fitness. At all. They never have. If you want evidence for that, note the fact that humans do things like use condoms. Also note that the growth of the world's population is slowing and will probably stop by the end of the 21st century if trends continue.

There's a scientific literature on the subject of what God's utility function is. Entire books have been written about the topic. I'm familiar with this literature, are you?

That literature has exactly zero relevance to our current discussion, which is what human beings, value, care about, and try to maximize. You learn about that by studying basic psychology. Evolutionary theory may give us insights into how humans came to have our current values, but it has no relevance on what we should do now that we have them.

Our values are what we value, how we came to have them is irrelevant. If our values were bestowed on us by an alien geneticist rather than evolution we would behave exactly the same as we do now. Humans don't give a crap about "god's utility function." If they end up increasing entropy it is as a side effect to obtaining their real goals.

We had better talk about "optimization" then, or we will talk past each other.

Optimization has the same problem. Optimization literally refers to a consequentialist creature using its future forecasting abilities to determine how an object or meme would better suit its goals and altering that thing accordingly. Evolution can be metaphorically said to optimize, but that isn't strictly true. It's just a form of personification to make thinking about evolution easier.

Strictly speaking, evolution is just a description of a series of trends. Since human minds are bad at modeling trends, but good at modeling other consequentialists, sometimes it's useful to pretend that evolution is a consequentialist with "goals" and a "utility function" to help people understand it. It's less scientifically accurate than modeling evolution as a series of trends, but it makes up for it by being easier for a human brain to compute. The problem is that, while most scientists understand this, there are some people who who misinterpret this to mean that evolution literally has goals, desires, and utility functions. You appear to be one of these people.

Really? How do you know that?

Because literally speaking, only consequentialist minds maximize things. You might be able to say evolution maximizes things as a useful metaphor, but literally speaking it isn't true.

Evolution is a gigantic optimization process with a maximand.

No it isn't. It is useful to pretend that it is because doing so makes it a little easier for the human mind to think about evolution. But really, evolution is just an abstract series of mindless trends.

You claimed above that it is "fun" - and my claim is that it is entropy.

I never claimed evolution tries to maximize fun. I claimed our civilization does. In other words, that the consequentialist minds making up human civilization use their forecasting abilities to foresee possible futures, and then steer the universe towards the one where they are having the most fun.

As I say, there's a substantial scientific literature on the topic - have you looked at it?

I'm familiar with some of the literature, and I've looked at your website. You constantly confuse the metaphorical "goals" evolution has with the real goals that consequentialist minds such as human beings have. For instance you say:

Another example: currently, researchers at ITER in France are working on an enormous fusion reactor, to allow us to accelerate the conversion of order into entropy still further.

This is trivially false, the reason researchers are working on a fusion reactor is to secure human beings cheap renewable energy to have more fun with. The fact that it increases entropy is a side-effect. The consequentialist human minds do not foresee a future with more entropy and take action in order to secure that future. They foresee a future where humans are using cheap energy to have more fun and take actions to secure that future. The entropy increase is an unfortunate, but acceptable side effect.

What you remind me of is one of those theologians who describe God as an "unmoved mover" or something like that and suggest such a thing must exist (which was a reasonable hypothesis at one point in history, even if it isn't now). They then make the ridiculous leap of logic that because an unmoved mover must exist, and you can call such a thing "God," that therefore a God with all the ludicrously specific human-like properties described in the Bible must exist.

Similarly, you take some basic facts about evolution and physics that every educated person agrees are true. Then you make bizarre leaps of logic to conclude that human beings care about maximizing IGF and maximizing entropy and other obvious falsehoods. I am not objecting to the evolutionary biology research you cite, I am objecting to the bizarre and unjustified inferences about human psychology and moral philosophy you use that research to make.

We had better talk about "optimization" then, or we will talk past each other.

Optimization has the same problem. Optimization literally refers to a consequentialist creature using its future forecasting abilities to determine how an object or meme would better suit its goals and altering that thing accordingly.

Nonsense. Look it up.

Strictly speaking, evolution is just a description of a series of trends. Since human minds are bad at modeling trends, but good at modeling other consequentialists, sometimes it's useful to pretend that evolution is a consequentialist with "goals" and a "utility function" to help people understand it. It's less scientifically accurate than modeling evolution as a series of trends, but it makes up for it by being easier for a human brain to compute. The problem is that, while most scientists understand this, there are some people who who misinterpret this to mean that evolution literally has goals, desires, and utility functions. You appear to be one of these people.

Feel free to substitute "maximisation" terminology if my preferred lingo causes you conceptual problems. Selfishness, progress and optimisation can all be "cashed out" in more long-winded terms. Remember: teleonomy is just teleology in new clothes.

Another example: currently, researchers at ITER in France are working on an enormous fusion reactor, to allow us to accelerate the conversion of order into entropy still further.

This is trivially false, the reason researchers are working on a fusion reactor is to secure human beings cheap renewable energy to have more fun with. The fact that it increases entropy is a side-effect. The consequentialist human minds do not foresee a future with more entropy and take action in order to secure that future. They foresee a future where humans are using cheap energy to have more fun and take actions to secure that future. The entropy increase is an unfortunate, but acceptable side effect.

This line of reasoning is intuitive, but, I believe, wrong. Destroying energy gradients is actively selected for in lots of ways. For example, it actively deprives competitors of resources. Organisms compete to dissipate sources of order by reaching them quicky and eliminating before others can. The picture of entropy as an inconvenient side effect seems attractive initially, but doesn't withstand close inspection.

I don't deny that properly functioning brains act like hedonic maximisers. Hedonic maximisation is a much weaker explanatory principle than entropy maximisation, though. The latter explains why water flows downhill. Hedonic maximisation is a narrow and weak idea - by comparison.

Then you make bizarre leaps of logic to conclude that human beings care about maximizing IGF and maximizing entropy and other obvious falsehoods.

To clarify, many humans fail to maximise their own inclusive fitnesses - largely because they are malfunctioning - with many of the most common malfunctions being caused by parasites - and the most common parasites being responsible for memetic hijacking. Humans and the ecosystems they are part of really do maximise entropy (subject to constraints) - or at least the MEP is a deep and powerful explanatory principle - when it comes to CAS and living systems.

"Fun" - if we are trying to treat the concept seriously - is better characterised as the proxy that brains use for the inclusive fitness of their associated organism.

No it isn't. Brains don't care about inclusive genetic fitness. At all. They never have. If you want evidence for that, note the fact that humans do things like use condoms. Also note that the growth of the world's population is slowing and will probably stop by the end of the 21st century if trends continue.

You are misunderstanding me. Pleasure is literally evolution's way of getting organisms do things that help them to increase their inclusive fitness. This idea is true, and is in no way refuted by condoms or the demographic transition.

You are misunderstanding me. Pleasure is literally evolution's way of getting organisms do things that help them to increase their inclusive fitness.

It is evolution's (metaphorical) way. When you say that brains use it as a proxy for genetic fitness that gave the impression that you thought brains literally cared about fitness and were optimizing for it.