If two things are correlated, there is causation. Either A causes B, B causes A, they have common cause, or they have a common effect you're conditioning on.
That doesn't seem to be strictly true. Of all the things that are correlated it would seem that there would be some that have none of the listed causal relationships. It is merely highly probable that one of those is the case.
That doesn't seem to be strictly true.
It goes against the spirit of "15 words" to insist on strict truth. The merit of the quote lies in the fourth clause.
or they have a common effect you're conditioning on.
That's the big surprise. The point of boiling it down to "15 words" is to pick which subtlety makes it into the shortest formulation.
Dennett's heterophenomenology seems to offer some of the good points of Skinner's behaviorism without a lot of the bad points.
Heterophenomenology notices that people's behavior includes making descriptions of their conscious mental states: they emit sentences like "I think X" or "I notice Y". It takes these behaviors as being as worthy of explanation as other behaviors, and considers that there might actually exist meaningful mental states being described. This is just what behaviorism dismisses.
In Beyond Freedom and Dignity Skinner writes (page 21)
A more important reason is that the inner man seems at times to be directly observed. We must infer the jubilance of a falling body, but can we not feel our own jubilance? We do, indeed, feel things inside our own skin, but we do not feel the things which have been invented to explain behaviour. The possessed man does not feel the possessing demon and may even deny that one exists. The juvenile delinquent does not feel his disturbed personality. The intelligent man does not feel his intelligence or the introvert his introversion. (In fact, these dimensions of mind or character are said to be observable only through complex statistical procedures.) The speaker does not feel the grammatical rules he is said to apply in composing sentences, and men spoke grammatically for thousands of years before anyone knew there were rules. The respondent to a questionnaire does not feel the attitudes or opinions which lead him to check items in particular ways. We do feel certain states of our bodies associated with behaviour, but as Freud pointed out we behave in the same way when we do not feel them; they are by-products and not to be mistaken for causes.
Dennett writes (page 83)
The heterophenomenologlcal method neither challenges nor accepts as entirely true the assertions of subjects, but rather maintains a constructive and sympathetic neutrality, in the hopes of compiling a definitive description of the world according to the subjects.
So far Skinner and Dennett are not disagreeing. Skinner did say "We do, indeed, feel things inside our own skin,...". He can hardly object to Dennett writing down our descriptions of what we feel, as verbal behaviour to be explained in the future with a reductionist explanation.
Dennett continues on page 85
My suggest, then, is that if we were to find real goings-on in people's brains that had enough of the "defining" properties of the items that populate their heterophenomenological worlds, we could reasonably propose that we had discovered what they were really talking about --- even if they initially resisted the identifications. And if we discovered that the real goings-on bore only a minor resemblance to the heterophenomenological items, we could reasonably declare that people were just mistaken in the beliefs they expressed.
Dennett takes great pains to be clear. I feel confident that I understand what he is taking 500 pages to say. Skinner writes more briefly, 200 pages, and leaves room for interpretation. He says that we do not feel the things that have been invented to explain behaviour and he dismisses them.
I think it is unambiguous that he is expelling the explanatory mental states of the psychology of his day (such as introversion) from the heterophenomenological world of his subjects, on the grounds that they are not things that we feel or talk about feeling. But he is not, in Dennett's phrase "feigning anesthesia" (page 40). Skinner is making a distinction, yes we may feel jubilant, no we do not feel a disturbed personality.
What is not so clear is the scope of Skinner's dismissal of say introversion. Dennett raises the possibility of discovering meaningful mental states that actually exist. One interpretation of Skinner is that he denies this possibility as a matter of principle. My interpretation of Skinner is that he is picking a different quarrel. His complaint is that psychologists claim to have discovered meaningful mental states already, but haven't actually reached the starting gate; they haven't studied enough behaviour to even try to infer the mental states that lie behind behaviour. He rejects explanatory concepts such as attitudes because he thinks that the work needed to justify the existence of such explanatory concepts hasn't been done.
I think that the controversy arises from the vehemence with which Skinner rejects mental states. He dismisses them out-of-hand. One interpretation is that Skinner rejects them so completely because he thinks the work cannot be done; it is basically a rejection in principle. My interpretation is that Skinner rejects them so completely because he has his own road map for research in psychology.
First pay lots of attention to behaviour. And then lots more attention to behaviour, because it has been badly neglected. Find some empirical laws. For example, One can measure extinction times: how long does the rat continue pressing the lever after the rewards have stopped. One can play with reward schedules. One pellet every time versus a 50:50 chance of two pellets. One discovers that extinction times are long with uncertain rewards. One could play for decades exploring this stuff and end up with quantitative "laws" for the explanatory concepts to explain. Which is when serious work, inferring the existence of explanatory concepts can begin.
I see Skinner vehemently rejecting the explanatory concepts of the psychology of his day because he thinks that the necessary work hasn't even begun, and cannot yet be started because the foundations are not in place. Consequently he doesn't feel the need to spend any time considering whether it has been brought to a successful conclusion (which he doesn't expect to see in his life-time).
This example pushed me into formulating Crowe's Law of Sarcastic Dismissal: Any explanation that is subtle enough to be correct is turbid enough to make its sarcastic dismissal genuinely funny.
Skinner had a subtle point to make, that the important objection to mentalism is of a very different sort. The world of the mind steals the show. Behaviour is not recognized as a subject in its own right.
I think I grasped Skinner's point after reading something Feynman wrote on explanations in science. You can explain why green paint is green by explaining that paint consists of a binder (oil for an oil paint) and a pigment. Green paint is green because the pigment is green. But why is the pigment green? Eventually the concept of green must ground in something non-green, wavelengths of light, properties of molecules.
It is similar with people. One sophisticated way of having an inner man without an infinite regress is exhibited by Minsky's Society of Mind. One can explain the outer man in terms of a committee of inner men provided that the inner men, sitting on the committee, are simpler than the outer man they purport to explain. And the inner men can be explained in terms of interior-inner-men, who are simpler still and whom we explain in terms of component-interior-inner-men,... We had better have remember that 1+1/2+1/3+1/4+1/5+1/6+... diverges. It is not quite enough that the inner men be simpler. They have to get simpler fast enough. Then our explanatory framework is philosophically admissible.
But notice the anachronism that I am committing. Skinner retired in 1974. Society of Mind was published in 1988. Worse yet, the perspective of Society of Mind comes from functional programming, where tree structured data is processed by recursive functions. Does your recursive function terminate? Programmers learn that an infinite regress is avoided if all the recursive calls are on sub-structures of the original structure, smaller by a measure which makes the structures well-founded. In the 1930's and 1940's Skinner was trying to rescue psychology from the infinite regress of man explained by an inner man, himself a man. It is not reasonable to ask him to anticipate Minsky by 50 years.
Skinner is trying to wake psychology from its complacent slumber. The inner man explains the outer man. The explanation does indeed account for the outer man. The flaw is that inner man is no easier to explain than the outer man.
We could instead look at behaviour. The outer man has behaved before. What happened last time? If the outer man does something unexpected we could look back. If the usual behaviour worked out badly the time before, that offers an explanation of sorts for the change. There is much to be done. For example, if some behaviour works ten times in a row, how many more times will it be repeated after it has stopped working? We already know that the inner man is complicated and hence under-determined by our experimental observations. This argues for caution and delay in admitting him to our explanations. We cannot hope to deduce his character until we have observed a great deal of his behaviour.
But let us return to humour and the tragedy of Sidney Morgenbesser's sarcastic dismissal
Let me see if I understand your thesis. You think we shouldn't anthropomorphize people?
The tragedy lies in the acuteness of Morgenbesser's insight. He grasped Skinner's subtle point. Skinner argues that anthropomorphizing people is a trap; do that an your are stuck with folk psychology and have no way to move beyond it. But Morgenbesser makes a joke out of it.
I accept that the joke is genuinely funny. It is surely a mistake for biologists to anthropomorphize cats and dogs and other animals, precisely because they are not people. So there is a template to fill in. "It is surely a mistake for psychologists to anthropomorphize men and women and other humans, precisely because they are not people." Hilarity ensues.
Morgenbesser understands, makes a joke, and loses his understanding somewhere in the laughter. The joke is funny and sucks every-one into the loss of understanding.
Can someone please let me know why this is the most down-voted I have ever been since de-lurking on this site? I'm not whining, I genuinely want to know what intellectual standards I'm not meeting or what social rules I'm violating by posting this.
My goal in posting this was to identify possible dangling units within the friendly AI concept.
Readers don't know what your post is about. Your comment explains "My goal ..." but that should be the start of the post, orienting the reader.
How does your hypothetical help identify possible dangling units? You've worked it out in your head. That should be the second part of post, working through the logic, here is my goal, here is the obstacle, here is how I get round it.
For the most part the objects which approve themselves to us are not so much the award of well-deserved certificates --- which is supposed by the mass of unthinking people to be the main object --- but to give people something definite to work for; to counteract the tendency to sipping and sampling which so often defeats the aspirations of gifted beings,...
--- Sir Hubert Parry, speaking to The Royal College of Music about the purpose of music examinations
Initially I thought this a wonderful quote because, looking back at my life, I could see several defeats (not all in music) attributable to sipping and sampling. But Sir Hubert is speaking specifically about music. The context tells you Sir Hubert's proposed counter to sipping and sampling: individual tuition aiming towards an examination in the form a viva.
The general message is "counter the tendency to sipping and sampling by finding something definite to work for, analogous to working ones way up the Royal College of Music grade system". But working out the analogy is left as an exercise for the reader, so the general message, if Sir Hubert intended it at all, is rather feeble.
Any examples of total recursive functions that are not primitive recursive and do not violently explode?
The set of primitive recursive functions is interesting because it is pretty inclusive, (lots of functions have a primitive recursive implementation) and primitive recursive functions always terminate. I'm interested in trying to implement general purpose machine learning by enumerating primitive recursive functions. Which raises the question of just how general the primitive recursive functions really are.
Ackermann's function gives an example of what you miss out when you confine yourself to primitive recursive functions. But Ackermann's function explodes violently. The values of the function rapidly become too large for any practical use. If that were typical, I would think that I miss out on nothing of practical importance when I restrict myself to primitive recursive functions.
But I suspect that the violent explosion is only there to meet the needs of the proof. Given a function that is specified by an implementation that is not primitive recursive, how can you tell whether it is primitive recursive? There might be a clever primitive recursive way of implementing the function that is hard to find. One needs to come up with a proof, and that is also hard to find. The violent explosion of Ackermann's function lets one construction a proof; no primitive recursive function goes BOOM! quite as dramatically.
Are there any other proof techniques and hence other, more practically important, functions that are (known to be) total recursive but not primitive recursive?
It is an important topic, but the Institute of Economic Affairs landing page that you link to is pretty lame.
Emphasizing "Evidence" gives one a hefty shove towards evidence that is quick and easy to gather.
QUICK The IEA say
A disregard for substitution effects.
but the actual problem is that substitution takes time. If you want to gather evidence about substitution effects you have to be patient. "Evidence based policy making" is biased towards fast-evidence, to accommodate the urgency of policy making. So of course substitution effects get under-estimated.
EASY Computing correlations is easy, tracing causality is hard. Worse, you can hardly hope to unravel the network of causal connections in a real world problem without making some theoretical commitments. An emphasis on "Evidence" leaves you relying on the hope that correlation does imply causality because you can get evidence for correlations. Causality? Not very practical. Then you get kicked in the teeth by Goodhart's Law
The IEA say
Calculating the external costs of harmful activities.
which is true, but hardly the worst of the problems. Ideally one would estimate the benefits of an economic policy based on adding the consumer surplus and the producer surplus. But this is too hard. Instead one tots up the market prices of things. This leads to GDP, which is a notoriously crap measure of welfare. But if you insist on "evidence" you are going to end up throwing out theoretical considerations of consumer and producer surplus in favor of GDP.
This submission is getting down voted. You might what to blog about the topic and try again with a link to your blog post. It shouldn't be too hard to provide a substantial improvement on the IEA landing page.
However, Polya's notion of being a good guesser has been formalised as Bayesian reasoning. If a theorem prover, facing a fork in its search tree, uses Bayes Theorem to pick the most likely branch first, it may find much deeper results.
I don't think there is currently much overlap between those working on theorem proving and those working on Bayesian statistics. There is perhaps even a clash of temperaments between those who seek absolute certainty and those who seek an edge in the messy process of muddling through. Nevertheless I foresee great possibilities of using Bayesian reasoning to guide the internal search of theorem provers, thus giving them a sense of direction.
What I question is whether the Bayesian reasoning algorithms are at all computationally feasible to implement to solve nontrivial problems. See the final two paragraphs of my comment here about computational complexity. Do you have evidence in the other direction?
Do you have evidence in the other direction?
No. I think one typically has to come up with a brutally truncated approximation to actually Bayesian reasoning. For example, if you have n propositions, instead of considering all 2^ n basic conjunctions, ones first idea is to assume that they are all independent. Typically that is a total failure; the independence assumption abolishes the very interactions that were of interest. So one might let proposition n depend on proposition n-1 and reinvent Markov models.
I don't see much hope of being able to anticipate which, if any, crude approximations to Bayesian reason are going to work well enough. One just has to try it and see. I don't think that my comment goes any deeper than saying that there are lots of close to practical things due to be tried soon, so I expect one or two pleasant surprises.
Bayesian probability theory tells you how to process evidence as well as possible. If you know what's evidence of a correct path, you can make it into an ad-hoc hueristic more easily than a part of a Bayesian update. Seems like the real insight required is to figure out what's evidence of a correct path.
You've put your finger on a weakness of my optimistic vision. If the guesses are calling it 90% of the time, they significantly extend the feasible depth of search. But 60:40? Meh! There is a lot of room for the insights to fail to be sharp enough, which turns the Bayesian stuff into CPU-cycle wasting overhead.
= f037147d6e6c911a85753b9abdedda8d){kind=logo}
= 27d567bc2d48d9f40e9ccc20ea370b15)
= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
Saunders Mac Lane, Categories for the Working Mathematician