Tuesday

Dec282010

## RS25 - Q&A With Massimo and Julia

Release date: January 2, 2011

Massimo and Julia answer listeners' questions, while trying to stay away from politics. In this installment the topics include: Is quantitative research more scientific than qualitative? Can philosophers really claim to have expertise on something? Is skepticism just another name for intelligence? How valuable is feminist philosophy? What is Bayesian reasoning? And what are M&J's anti-*akrasia* strategies?

Comment on the episode teaser.

## Reader Comments (15)

Great podcast and thanks for answering my question. I admit that Massimo change my mind a little about qualitative research but for the most part I am with Julian on this one. Anyone that was exposed to qualitative studies in most social sciences and humanities and the generalizations they make to much larger populations based on very small samples could cringe sometimes :)

Massimo,

You said there were results in computer science showing that on complex issues, one can reasonably compute multiple contrary answers to the same problem. Which results were you speaking of?

I'm most familiar with the research following from Aumann's Agreement Theorem, newer versions of which state that ideal Bayesian reasoners, given a few conditions, should never disagree.

That said, I'm not really that well-read on the epistemology of disagreement.

On Bayesianism...

I don't see too many people claiming that Bayesianism

describesscientific process. Scientific process today appears to be more clearly Popperian, and focused on NHST and computing p-values and so on.And while certain brain processes work in Bayesian terms, there are a huge raft of cognitive heuristics that depart from ideal Bayesian reasoning, so the human mind is far from Bayesian.

I'm most familiar with those who think science

shouldbe Bayesian, which I think is hard to argue with after a book like Jaynes'Probability Theory: The Logic of Science.I used to agree with Julia that 1 + 1 = 2 was 100% certain, but I now disagree. I'm just really, really certain. A short post on this is here:

http://lesswrong.com/lw/jr/how_to_convince_me_that_2_2_3/

Oh, Luke, I wasn't saying that 1+1=2 is 100% certain! That was Massimo's claim, not mine. I was just saying that it's tricky to decide how/if that statement fits into Bayesian reasoning. I'm still making up my mind about how to think about statements like that.

Julia,

Ah, okay.

I walk into the CBD on my lunch breaks whilst listening to the rationally speaking podcast. Over the past several weeks people walking past me must have been wondering why I looked constipated. The reason is that I'm a transhumanist, memeticist and feminist!

I'll stick with just the feminism for now, though. Whilst I loathe post-modernism and the kinds of sesquipedalian superfluity that you read out in the podcast, I am impelled to point out that feminism as a field is very much based in empirical sociological data, and that that kinds of literary obfuscation you're referring to is somewhat cherry-picked and not necessarily representative.

Also, because it's a political movement there tends to be a certain amount of hyperbole and absurd claims made from passion rather than rationality, but the vast majority of feminist research isn't this kind of thing at all. Rather, it deals very much with issues such as gendered pay disparity, divisions in household labour, quantitative surveys of perceptions of gender, and other much more academically-sound things at the more sciencey end of the social science spectrum.

It's also a very divided political/philosophical field. I'm a male radical feminist (radical means 'root', not 'way out') and our camp is very distinct from 'liberal' post-modernist feminists. Even amongst radical feminists there is much very vehement disagreement, and so to tar all 'feminist philosophers' with the same brush is probably not very realistic nor fair. Many feminists demonstrate extremely well-developed sceptical and rational minds - it's something of a pre-requisite to be able to examine one's identity and society as a whole without the confirmation biases of ingrained gender roles.

Here are a couple of formal proofs that 1+1=2, derived from Peano axioms.

http://mathforum.org/library/drmath/view/61208.html

http://mathforum.org/library/drmath/view/51551.html

Of course, there's a possibility that those proofs have an error, like the "proofs" that 1+1=0.

In any case, Bayesian reasoning allows for 100% certainty, so why is that even an issue?

NHST and p-values can be a problem. Dr. Daryl Bem, a distinguished psychologist at Cornell, used them to find evidence of psychic powers (psi) where there was none. Check out this critique:

http://www.ruudwetzels.com/articles/Wagenmakersetal_subm.pdf

"In a recent article, Dr. Bem conducted nine studies with over a thousand participants in an attempt to demonstrate that future events retroactively affect people’s responses. Here we discuss several limitations of Bem’s experiments on psi; in particular, we show that the data analysis was partly exploratory, and that one-sided p-values may overstate the statistical evidence against the null hypothesis. We reanalyze Bem’s data using a default Bayesian t-test and show that the evidence for psi is weak to nonexistent."

They basically used Bayes factors instead of p-values, which makes more sense to me.

http://en.wikipedia.org/wiki/Bayes_factor

Here's my problem with null hypothesis significance testing.

If I flip a coin 5 times and get THHHT, the chance of getting that exact sequence with a fair coin is 1/32 = 0.03 < 0.05. Therefore, reject the null hypothesis that it's a fair coin.

Now, you might say I should've found the probability of getting 3 or more heads, but why should I look at that statistic instead of, say, the probability of getting a palindrome?

Max, in your calculation of the probability of the certain sequence you didn';t conduct any hypothesis testing because you didn't really have any hypothesis or calcualted a test statistics. In order of doing a formal test you need to state your hypothesis in advance a-priori and not post them after the fact. If you indeed predicted this particular sequence well before you tossed the coin and got it, then there is a chance that something might be wrong with the coin.

Gil,

Bingo! If my alternative hypothesis is a magic coin that's programmed to output THHHT, I should compare the probability of this magic coin outputting THHHT (100%) with the probability of a fair coin outputting THHHT (1/32). The ratio is 1/(1/32)=32, and that's the Bayes factor indicating strong evidence in favor of the magic coin. No p-value.

Or if my alternative hypothesis is a random-palindrome-making coin, the Bayes factor is (8/32)/(1/32)=8, indicating substantial evidence in favor of it.

Or if my alternative hypothesis is a two-headed coin, the Bayes factor is 0, indicating very strong evidence against it.

Using the Bayes factor forces you to state your alternative hypothesis explicitly.

Msx, you don't really need to use Bauseian thinking into this sort of question. You can arrive to the same result using traditional hypothesis testsing. Note that your logic in the premise is a bit flowed. You didn't calculate and test statistic. All you did was to compare the probablity of getting a certain sequence to an arbitrary number that is 0.05. That's not how hypothesis testing works. In a true hypothesis test you know that the probability of getting what you got is 1/32. To check if it's any different than that you will need to repeat the experiment several times. To see if the proportion of your observations differ from 1/32 over the long run you will need to calculate the right statistics which is p-hat and then look what the p-value that correspond to that. Only then, if the p-value is less than 0.05 you will be able to say that there is something wrong with the coin with some certainty.

@Jesse Richardson -- Thanks for listening despite your ideological differences with some of the content! As a self-described memeticist, maybe you could tell me what you disagreed with in Massimo's and my analysis of the "meme" concept? I'd be interested to hear another opinion.

Regarding the feminism issue -- I should emphasize that "feminist philosophy" Is a very different creature from feminism as a political movement or an ideology. We were specifically asked about feminist philosophy, and I don't think its questionable validilty has any bearing on the validity of what people more generally refer to as "feminism."

Hey Julia,

Hmm, I'm not sure if delineating between feminist philosophy and feminism as a political movement can be quite so clear - ultimately any feminist philosophy must concern itself with the politics and sociological foundations of feminism. That is to say that feminist philosophy is necessarily political philosophy. Having said that I do recognise that there is a not insubstantial amount of utter bollocks in some of the more navel-gazing pomo feminist academia.

If you're interested I find Bell Hooks to be a particularly insightful feminist author - she comes from a perspective of looking at power dynamics from a basis of race, class and gender, and despite being ostensibly pomo herself, seems to manage to make cogent, succinct arguments with

actual content. Her latest book is on teaching critical thinking, so this gives some indication of why she's worth reading. This talk by Emily Maguire on the myth of post feminism is also well worth watching: http://www.themonthly.com.au/emily-maguire-accidental-feminist-anu-1507Now, memetics. The main criticism I'd have of your analysis would be that you were examining memetics' usefulness in an empirical rather than philosophical context. One could criticise any philosophy if the criteria for its worth was that it be subject to scientific empiricism. I think I've discerned that you're sceptical of philosophy in general because of this fact, and whilst I disagree with you, that's a whole other argument. Curiously, though, it was the full time philosopher who was most critical of memetics whilst you seemed more receptive to its potential.

Personally I think that there

isan amount of fuzziness to it, but it's still an extremely fascinating and useful model to examine human behaviour, sociology and psychology through. I view the human mind as a kind of parliament in which memes are constituents, memeplexes are members of parliament representing all the memes, and thoughts are like legislation attempting to be passed. When we have a thought, every single meme seeks to perpetuate itself through exerting a kind of self-interested power upon the 'legislation', but whilst all this is going on the conscious mind only gets to hear the 6 o clock news to hear that a bill has been passed one way or the other.This model diverges quite a lot from the classical 'meme's eye view' of singular self-replication of an idea as it looks at things on a group level, dynamically. Someone like Susan Blackmore sees a more pure and singular memetic function, but I'm not really attached to the analogous continuity of genes and memes, I'm more interested in how self-interested replicating ideas function as a psychological dynamic.

As an interesting aside Dawkins' attachment to the gene's eye view means that he rejects group selection as a concept, and I think he's likely wrong about this (but I love him, so please don't tell him so). We humans like to attach to ideas and attempt to force logical compliance and continuity to them post hoc, even the best of us.

By looking at ideas as self-interested agents in their own right, I think we can build a model of understanding human consciousness and sociological functions that may be very useful in terms of understanding how and why some ideas are more influential than others. What ideas must already be working in their own self-interest in our minds for other ideas to be able to replicate and spread? As it stands I think we view psychology in too linear a fashion, and memetics is a valuable way to approach consciousness and psychology from a more analytical perspective. It's not a perfect model, and it doesn't mirror genetic evolution in any especially meaningful way (and this is where I think it can become tenuous, when memeticists attempt to forge this connection more than is warranted) but it is, however, potentially very useful and definitely interesting and worth exploring.

Sorry, I appear to have banged on a bit :P

Massimo,

In regards to the brief portion on Bayesian epistemology in the podcast, though you were correct to point out the that her worries do not (directly) extend to scientific methodology, I fear that you did not adequately answer her misunderstanding regarding the Bayesian assignment of degrees of belief to logical truths (tautologies) and (though she did not mention them) logical falsehoods.

In any standard text on probability theory, logical truths are assigned 1 and logical falsehoods are assigned 0. As Bayesian probability theory is an extension of classical probability theory, and thus constrained by the suitable axioms (Kolmogorov axioms, Cox's postulates, whatever), Bayesian assignments are constrained by logical probabilities (i.e., assigning tautologies 1 and logical falsehoods 0).

Anyone have a reference on the Julius Caesar article? I can't find it :/