Dissenting Views

Long overdue:

In May when I composed this post, I saw the LW community as having a dominant ideology, which I have since learned to label 'physical materialism'. I refrained from publically defining this ideology because of some kind of reluctance.

I didn't expect the community to change over time, but it seems to me there has been drift in the type of discussions that occur on Less Wrong away from epistemological foundations. So I feel more comfortable now outlining the tenets of average LW epistemology, as I perceived it, as a ‘historical’ observation.

The first and fundamental tenet of this epistemology is that there is a real, objective reality X that we observe and interact with. In contrast, persons with a metaphysical bent are less definitive about the permanent existence of an objective reality, and believe that reality alters depending on your thoughts and interactions with it. On the other extreme are skeptics that believe it is meaningless to consider any objective reality, because we cannot consider it objectively. (There are only models of reality, etc.)

For formalism and precision, I will here introduce some definitions. Define objective reality as a universe X = the set of everything that we could ever potentially observe or interact with physically. (This is what we consider “real”.) We cannot know if X is a subset of a larger universe X-prime. Suppose that it is: The component of X-prime that is outside X (X-complement) may ‘exist’ in some sense but is not real to us.

The second tenet is that anything we observe or interact with is a subset of X, the real physical world. While this trivially follows from the definition of X, what is being argued with physical materialism is not the tautology itself but the value of seeing things from this point of view. Trivially, there is nothing metaphysical in X; we either interact with something or we don’t.

In contrast, the metaphysical view is to consider reality = X-prime, and consider that everything we interact with physically/scientifically/objectively may only be a subset of our total experience of reality.

Comparing the views: Physical Materialism verses Metaphysical View

Consider the hypothetical, real sighting of a ghost: a white floating image is observed in front of two observers. The physical materialist observes the ghost, and knows that either (a) the ghost exists outside subjective experience, in which case the ghost must be reflecting light in such a way as to appear white and hazy, and the interaction of the light with the ghost could be studied and reproduced or (b) the ghosts exists as a subjective experience, in which case it is still physically manifested as a hallucination that may be equal to certain neural patterns, etc. The metaphysicist, in contrast, considers a third possibility as potentially reasonable: the ghost has a physical component (same cases a,b) AND ALSO a metaphysical component that explains the ‘existence’ of the ghost in some deeper way. For the metaphysicist, the physical materialist’s ghost is a subset of the “whole” ghost that actually straddles X and X- complement.

In my view, the physical materialist view is more coherent. We cannot know if the ghost straddles X and X-complement, but if it does, in no sense is the part of the ghost contained within X- complement “real” to us. It is not real because we can never observe or interact with this component in any way.

The epistemological question, all along; the debate over the ages, is whether holding any X- complement component (imaginary component) in our theory of the ghost will give us a better understanding of the X component (real part) of the ghost.

Personally, I see no evidence that the physical world X should not be informationally/theoretically complete. The labor of science is the belief that X can be understood within X itself. On the other hand, there is no proof that X is not dependent upon or manipulated in (scientifically) unfathomable ways by a larger X-prime, and it is conceivable that interactions occur between X and X-complement in ways that cannot be understood within X. Physical materialism really is just a matter of ideological preference, not fact. But it is the direction modern culture is certainly going; metaphysical religious views seem increasingly anachronistic and ‘separate’.

Another point of view typically held on Less Wrong: that since reality is 'just' physical that this implies that it is coarse, or simple or stupid. I think this is just a backlash to metaphysical accounts describing reality as divine and inspired. We can look around and see that reality is structured, patterned and organized/directed*. Physical materialism is not the belief that these observations are nonsense, but that we can explain them without resorting to the supernatural.

*Please allow materialistic interpretations of these anthropomorphic words… I'm not aware of adequate alternatives and suspect language is evolving too slowly.

I concur. We seem more interested in phenomena which are interesting psychologically than which are useful. This should not be surprising - interesting phenomena are fun to read about. Implementing a new cognitive habit takes hard work and repetition. Perhaps it is like divorcing warm fuzzies from utilons - we should differentiate from "biases that are fun to read/think about" and "practices which will help you become less wrong."

As a metaphor, consider flashy spinning kicks vs. pushups in martial arts. The former are much more fun to watch and think about, but boring exercises to build strength and coordination are much more basic and important.

This pretty vague for a heresy. Can you link to a comment or post that explains what you're referring to, or why we should condition on wide-scale adoption?

so there isn't a set of objectively rational goal values

There probably is for each individual, but none that are universal.

I contend otherwise. The utilitarian model comes down to a subjective utility calculation which is impossible (I use the word impossible realizing the extremity of the word) to do currently. This can be further explicated somewhere else but without an unbiased consciousness - without one which does not fall prey to random changes of desires and mis-interpretations or mis-calculations (in other words the AI we wish to build), there cannot be a reasonable calculation of utility such that it would accurately model a basket of preferences. As a result it is not a reasonable nor reliable method for determining outcomes or understanding individual goals.

True there may be instances in which a crude utilitarian metric can be devised which accurately represents reality at one point in time, however the consequential argument seems to divine that the accumulated outcome of any specific action taken through consequential thought will align reasonably if not perfectly with the predicted outcome. This is how utilitarianism epistemologically fails - the outcomes are impossible to predict. Exogeny anyone?

In fact, what seems to hold truest to form in terms of long term goal and short term action setting is the virtue ethics which Aristotle so eloquently explicated. This is how in my view people come to their correct conclusions while falsely attributing their positive outcomes to other forms such as utilitarianism. E.G. someone thinking "I think that the outcomes of this particular decision will be to my net benefit in the long run because from this will lead to this etc..". To be sure it is possible that a utilitarian calculation could be in agreement with the virtue of the decision if the known variables are finite and the exogenous variables are by and large irrelevant, however it would seem to me that when the variables are complicated past current available calculations understanding the virtue behind an action, behavior or those which are indigenous to the actor will yield better long term results.

It is odd because objective Bayesian probability is rooted in Aristotelian logic which is predicated on virtue ethics, and since Eliezer seems to be very focused on Bayesian probability that would seem to conflict with consequential utilitarianism.

However I may be read the whole thing wrong.

ED: If there is significant disagreement please explicate so I can see how my reasoning is not clear or believed to be flawed.

This is perhaps a useful disctinction:

When it comes to knowledge of the world you want to be more right.

But when it comes to reasoning I do think it is more about being less wrong... there are so many traps you can fall into, and learning how to avoid them is so much of being able to reason effectively.

The group title is attempting to be modest - which is cool.

Whether you try something new or try to formulate a model or hypothesis, you must at minimum be prepared for the result to be more wrong at first.

Disagree. You don't have to believe your new model or hypothesis.

Results are neither right nor wrong - they just are.

In contrast, you can be "less wrong" just by doing nothing, or by being a critic of those who do something.' But in the real world (and even in science), you can never win BIG -- and it's often hard to win at all -- if you never place any bets.

To expand a little on what timtyler said, I think you're mixing up beliefs and actions.

Doing nothing doesn't make your beliefs less wrong, and placing bets doesn't make your beliefs more right (or wrong).

Wanting to be 'less wrong' doesn't mean you should be conservative in your actions.

That it's not aimed at being "more right" -- which is not at all the same as being less wrong.

I've also had mixed feelings about the concept of being "less wrong." Anyone else?

Of course, it is harder to identify and articulate what is wrong than what is right: we know many ways of thinking that lead away from truth, but it harder to know when ways of thinking lead toward the truth. So the phrase "less wrong" might merely be an acknowledgment of fallibilism. All our ideas are riddled with mistakes, but it's possible to make less mistakes or less egregious mistakes.

Yet "less wrong" and "overcoming bias" sound kind of like "playing to not lose," rather than "playing to win." There is much more material on these projects about how to avoid cognitive and epistemological errors, rather than about how to achieve cognitive and epistemological successes. Eliezer's excellent post on underconfidence might help us protect an epistemological success once we somehow find one, and protect it even from our own great knowledge of biases, yet the debiasing program of LessWrong and Overcoming Bias is not optimal for showing us how to achieve such successes in the first place.

The idea might be that if we run as fast as we can away from falsehood, and look over our shoulder often enough, we will eventually run into the truth. Yet without any basis for moving towards the truth, we will probably just run into even more falsehood, because there are exponentially more possible crazy thoughts than sane thoughts. Process of elimination is really only good for solving certain types of problems, where the right answer is among our options and the number of false options to eliminate is finite and manageable.

If we are in search of a Holy Grail, we need a better plan than being able to identify all the things that are not the Holy Grail. Knowing that an African swallow is not a Holy Grail will certainly help us not not find the true Holy Grail because we erroneously mistake a bird for it, but it tells us absolutely nothing about where to actually look for the Holy Grail.

The ultimate way to be "less wrong" is radical skepticism. As a fallibilist, I am fully aware that we may never know when or if we are finding the truth, but I do think we can use heuristic to move towards it, rather than merely trying to move away from falsehood and hoping we bump into the truth backwards. That's why I've been writing about heuristic here and here, and why I am glad to see Alicorn writing about heuristics to achieve procedural knowledge.

For certain real-world projects that shall-not-be-named to succeed, we will need to have some great cognitive and epistemological successes, not merely avoid failures.

Perhaps there are intuitive notions of "less wrong" that are different from "more right", but in a technical sense, they seem to be the same:

At this point it may occur to some readers that there's an obvious way to achieve perfect calibration - just flip a coin for every yes-or-no question, and assign your answer a confidence of 50%. You say 50% and you're right half the time. Isn't that perfect calibration? Yes. But calibration is only one component of our Bayesian score; the other component is discrimination.

Suppose I ask you ten yes-or-no questions. You know absolutely nothing about the subject, so on each question you divide your probability mass fifty-fifty between "Yes" and "No". Congratulations, you're perfectly calibrated - answers for which you said "50% probability" were true exactly half the time. This is true regardless of the sequence of correct answers or how many answers were Yes. In ten experiments you said "50%" on twenty occasions - you said "50%" to Yes-1, No-1; Yes-2, No-2; .... On ten of those occasions the answer was correct, the occasions: Yes-1; No-2; No-3; .... And on ten of those occasions the answer was incorrect: No-1; Yes-2; Yes-3; ...

Now I give my own answers, putting more effort into it, trying to discriminate whether Yes or No is the correct answer. I assign 90% confidence to each of my favored answers, and my favored answer is wrong twice. I'm more poorly calibrated than you. I said "90%" on ten occasions and I was wrong two times. The next time someone listens to me, they may mentally translate "90%" into 80%, knowing that when I'm 90% sure I'm right about 80% of the time. But the probability you assigned to the final outcome is 1/2 to the tenth power, 0.001 or 1/1024. The probability I assigned to the final outcome is 90% to the eighth power times 10% to the second power, (0.9^8)*(0.1^2), which works out to 0.004 or 0.4%. Your calibration is perfect and mine isn't, but my better discrimination between right and wrong answers more than makes up for it. My final score is higher - I assigned a greater joint probability to the final outcome of the entire experiment. If I'd been less overconfident and better calibrated, the probability I assigned to the final outcome would have been 0.8^8 * 0.2^2, 0.006.

Accounting for the uncertainty in your own mind only gets you so far, to a certain minimum of wrongness. To do better, to be less wrong, you have to actually be right about the rest of the universe outside your mind.

I think this is an excellent point; I'm not sure it's a valid criticism of this community.

As the wiki link points out, the word "ideology" has a fairly neutral sense in which it simply refers to "a way of looking at things", which seems to reflect Byrnema's focus on the underlying assumptions this community brings to things.

I don't think it's a stretch to suggest that many of us here probably do share particular ways of looking at things. It's possible that these general ways of looking-at-things do in fact let us track reality better than other ways of looking-at-things; but it's also possible that we have blind spots, and that our shared "ideology" may sometimes get in the way of the x-rationality we aspire to.

It seems like, to some extent, you are confusing rationality with being "Spock".

Emotion is not irrational. Luck can't be irrational, because it doesn't exist. Aspects of human thought, such as imagination, are the bedrock of human rationality.

The universe is irrational and we have to live in it. [...] Dealing with all of this, a perfectly rational being would forfeit. There is too much chaos, too much unpredictability.

Interesting. I see this as some kind of anti-thesis to rationality; being in some sense exactly what rationalists deny. Sure, we may believe in chaos and unpredictability, but we still believe that rationality is the best way to deal with it.

While I can sympathize with the view that the universe is sometimes too complex, I do believe that predictable success is possible to some extent (probabilistically, for example), and that being rational is the way to achieve that. If being irrational predictably gives better results in any specific context, then our rational theory needs to be expanded to include that irrational behavior as rational. My strongest belief is that the theory of rationality can always be expanded in a consistent way to include all behavior that yields success. I realize this is a substantial assumption.

I would like to learn more about what sorts of things are nevertheless "beyond" rationality, and whether there are some ways to be more rational about these things, or if it's just separate (so that the label rational/irrational doesn't apply.) For example, I think rationalists generally agree that preferences and values are outside rationality.

I think you need to define what you mean by "morality" a lot more carefully. It's hard to attribute meaning to the statement "People should act without morals." Even if you mean "Everyone should act strictly within their own self-interest", evolutionary psychology would demand that you define the unit of identity (the body? the gene?), and would smuggle most of what we think of as "morality" back into "self-interest".

Moral skepticism is not particularly impressive as it's the simplest hypothesis. Certainly, it seems extremely hard to square moral realism with our immensely successful scientific picture of a material universe.

The problem is that we still must choose how to act. Without a morality, all we can say is that we prefer to act in some arbitrary way, much as we might arbitrarily prefer one food to another. And...that's it. We can make no criticism whatsoever about the actions of others, not even that they should act rationally. We cannot say that striving for truth is any better than killing babies (or following a religion?) anymore than we can say green is a better color than red.

At best we can make empirical statements of the form "A person should act in such-and-such manner in order to achieve some outcome".

Some people are prepared to bite this bullet. Yet most who say they do continue to behave as if they believed their actions were more than arbitrary preferences.

My position may be one of those you criticize. I believe something that bears an approximation to "morality" is both worth adhering to and important.

I think a particular kind of morality helps human societies win.

Morality, as I understand it, consists of a set of constraints on acceptable utility functions combined with observable signals of those constraints.

Do I believe that this type of morality is in any sense ultimately correct? No. In a technical sense, I am a complete and total moral skeptic.

However, I do think publicly-observable moral behavior is useful for coordination and cooperation, among other things. To the extent that this makes us better off--to the extent it makes me better off--I would certainly think that even a moral skeptic might find it interesting.

Perhaps LWers are "too uncritical toward their moral prejudices." But it's at least worth examining which of those "moral prejudices" are useful, where this doesn't conflict with other, more deeply held values.

Finally, morality broadly enough construed is a condition of rationality: if morality is taken to simply be your set of values and preferences, then it is literally necessary to a well-defined utility function, which is itself (arguably) a necessary component of rationality.

Certainly, the main ideological tenet of Less Wrong is rationality. However, other tenets slip in, some more justified than others. One of the tenets that I think needs a much more critical view is something I call "reductionism" (perhaps closer to Daniel Dennetts "greedy reductionism" than what you think of). The denial of morality is perhaps one of the best examples of the fallacy of reductionism. Human morality exists and is not arbitrary. Intuitively, the denial of morality is absurd and ideologies that result in conclusions that are intuitively absurd should require extraordinary justification to keep believing in them. In other words, you must be a rationalist first and a reductionist second.

First, science is not reductionist. Science doesn’t claim that everything can be understood by what we already understand. Science makes hypotheses and if the hypotheses don’t explain everything, it looks for other hypotheses. So far, we don’t understand how morality works. It is a conclusion of reductionism – not science – that morality is meaningless or doesn’t exist. Science is silent on the nature of morality: science is busy observing, not making pronouncements by fiat (or faith).

We believe that everything in the world makes sense. That everything is explainable, if only we knew enough and understood enough, by the laws of the universe, whatever they are. (As rationalists, this is the only single, fundamental axiom we should take on faith.) Everything we observe is structured by these laws, resulting in the order of the universe. Thus a falsehood may be arbitrary, but a truth, and any observation, will always not be arbitrary, because it must follow these laws. In particular, we observe that there are laws, and order, over all spatial and temporal scales. At the atomic/molecular scale, we have the laws of physics (classical mechanics). At the organism level, we have certain laws of biology (mutation, natural selection, evolution, etc). At the meta-cognitive level, there will be order, even if we don’t perceive or understand that order.

Obviously, morality is a natural emergent property of sapience. (Since we observe it.) Perhaps it is not necessary… concluding necessity would require a model of morality, that I don’t have. But imagine: the space of all sapient beings over all time in the universe. Imagine the patterning of their respective moralities. Certainly, their moralities will be different from each other. (This I can conclude, because I observe differences even among human moralities.) However, it is no leap of faith but just the application of the most important assumption, that their morality will also have certain features in common; will obey certain laws and will evidence order. Even if not readily demonstrated in a single realization. Our morality – whatever it is – is meant to be, is natural, is without question obeying laws of the universe.

By analogy with evolution (here I am departing from science and am reverting to reductionism, trying to understand something in the context of what I do understand -- the analogy doesn’t necessarily hold, and one must use their intuition to estimate if the analogy is reasonable) there may not be a unique emergent “best” morality, but it may be the case that certain moralities are better than others, just as some species are more fit than others. So instead of thinking of the existence of different moralities in humanity as evidence that morality is “relative” and arbitrary or meaningless, I see the variations as evidence that morality is something that is evolving, competing, striving even among humans to fit an idealized meta-pattern of morality, whatever it may be. Like all idealized abstractions, the meta-morality would be physically unobtainable. The meta-morality itself could only be deduced by looking at the pattern of moralities (across sapient life forms would be most useful) and abstracting what is essential, what is meaningful.

It is a constant feature of life to want to live. Every species has an imperative to do their utmost to live, each species contributes itself to the experiment, to participate in the demonstration of which aspects of life are most fit. Paradoxically, every species has an imperative to do its utmost to live, even if it means changing from what they were. There is a trade-off between fighting to stay the same (and “win” for your realization of life), and changing (a possible win for the life of the next species). Morality might be the same: we fight for our idea of morality (with a greater drive, not less, than our drive for life) but we will forfeit our own morality, willingly, for a foreign morality that our own morality recognizes as better. Morality wants to achieve this ideal morality that we only see the vaguest features of. (In complex systems, “wanting” means inexorably moving towards, globally.)

I’m not always sure when one moral position is better than another – there seems to be plenty of gray at this local level of my understanding. However, some comparisons are quite clear. That morality exists is a more moral position than denying that it exists. Also, morality is not just doing what’s best for the community by facilitating cooperation: that explanation is needlessly reductionist. We can see this by the (abstract) willingness of moral people to sacrifice themselves – even in a total loss situation – for a higher moral ideal. Morality is not transcendent however; “transcendent” is an old word that has lost its usefulness. We can just say that morality is an emergent property. An emergent property of something. A month ago, I would have said intelligence, but I’m not sure. A certain kind of intelligence, surely. Social intelligence, perhaps. That even ants possess, but not a paperclip AI.

[Later edit: I've convinced myself that a paperclip AI does have a morality, though a really different one. Perhaps morality is an emergent property of having a goal. Could you convince a paperclip AI to not make any paperclips if the universe would have more "paperclipness" without them? Maybe it would decide that everything being paperclips results in an arbitrary number, and it would be a stronger statement to eradicate all paperclips...)

Sorry. I thought about things a little and realized that a few things about prospect theory definately need to be scrapped as bad ideas.. The probability weighing for instance. But other quirks (such as loss aversion or having different utilities for loss vs gain) might be useful to retain...

It would really be good if I knew a bit more about the different descision theories at this point. Does anyone have any good references from where one would get an overview and good references?

I think the tendency to use terms like "utility function" and "prior" stems more from a desire to be precise than to show off. Both have stuck with me as seemingly-useful concepts far outside the space of conversations in which they're potentially intelligible to others.

Unless you know it's superfluous, give jargon the benefit of the doubt. When communication is more precise, we all win.

What do you mean by "reason"?

I think this is a problem of Rhetoric. I dialed it back to Plato and Aristotle and have made my way up to "The New Rhetoric: A Treatise on Argumentation". I shall report back if I have any success as I make my way to the here and now. In the meantime I note the following: as a sceptic I hope I'm ready to cast aside any idea, however cherished, that fails of its purpose - that fails the acid bath test of falsifiability. I cast aside the snake handlers that kept Romney from being nominated and I cast aside the "empathy" that leads Sotomayor to believe that single Latina Moms make better judgments than white males. But what's left? Is there really room for a party of Rationalists? Won't the purest rationalist sell out his brethren for a better deal offered by the emotionalists? Isn't that what a good rationalist would do? Are we ultimately the victims of our own good sense? Or are we able to deal with the negative externailities of personal rationalism. And if so, how? Alas, even Spock has now decided "if it feels right, do it!"

I don't discuss religion much - but here is my list of "Viable Intelligent Design Hypotheses":

http://originoflife.net/intelligent_design/

I'd replace your item 1 with physicalism. The "rational religious" example you propose might get criticized here, but not for belief in the supernatural.

FWIW, this doesn't describe me:

http://alife.co.uk/essays/the_singularity_is_nonsense/

Mission accomplished!