Wednesday, June 24, 2015

How's the Muslim-hunt working out for you, Sam Harris?

Three years ago, Sam Harris wrote in defense of racial profiling of Muslims because they are overwhelmingly more likely to commit acts of terrorism than non-Muslims, specifically:
"We should profile Muslims, or anyone who looks like he or she could conceivably be Muslim."
(Emphasis added.)

Turns out there is actual data to inform this debate.  As The New York Times reports:
Since Sept. 11, 2001, nearly twice as many people have been killed by white supremacists, antigovernment fanatics and other non-Muslim extremists than by radical Muslims: 48 have been killed by extremists who are not Muslim, compared with 26 by self-proclaimed jihadists, according to a count by New America, a Washington research center.
...
Non-Muslim extremists have carried out 19 such attacks since Sept. 11, according to the latest count, compiled by David Sterman, a New America program associate, and overseen by Peter Bergen, a terrorism expert. By comparison, seven lethal attacks by Islamic militants have taken place in the same period.
So, anyone want to place a bet as to whether this will prompt Sam to issue a retraction?  I'll give you long odds against.

22 comments:

Luke said...

A while ago, you sent me a link to Sam Harris' The Limits of Discourse, where he attempts a conversation with Noam Chomsky. You didn't say whether you leaned toward Chomsky or Harris. :-p

Ron said...

That's like asking if I prefer to beat with wife with a ruler or a belt. Harris thinks the U.S. can do no evil and Chomsky thinks it can do no good, so they both run afoul of Ron's First Law. At least Chomsky can back up his arguments with actual data. It was fun watching Harris get smacked down by him, but it was also a little sad that Harris didn't seem to realize that that's what was happening.

coby said...

I think that is an inaccurate presentation of Harris and Chomsky, more so of Chomsky. I think it would be just as pithy, but much more correct, to say that Harris thinks the US has only good (moral) intentions and Chomsky thinks the US gives morality practically zero priority. Splitting that difference does not land you anywhere close to the reality.

Publius said...

Hard to Choose

The data* you present do not support your thesis. You demonstrate the fallacy of relative privation, perhaps with a measure of red herring and false equivalence mixed in.

The data* you discuss can be summarized:
A. 26 killed by self-proclaimed jihadists
B. 48 killed by non-muslim extremists

The desire to prevent "A" has nothing to do with "B".
Say Sam is interested in reducing the death rate from jihadists - reducing it from about 2/year to 0/year. Sam suggests a specific strategy to address problem "A". The strategies to address "A" are not dependent on problem "B" existing, or the strategies one might use for "B".

You may argue that "B" is more important, but this is different discussion than what strategies could be used to reduce problem "A".

Solving Problem A

Say you have a large dataset of data about people. Each person is classified as jihadist (1) or non-jihadist (0) - this is the "Y" variable. For each person, you also have 30 attributes about that person, which are X1, X2, ..., X30.

You run your dataset through a machine learning algorithm - say, a C4.5 Decision Tree -- to determine which X factors are useful in predicting Y.

Now say 5 factors come back as useful in the classification tree (model) - X7, X13, X23, X27, and X29. If X13 is an indicator variable for Muslim (1) or non-Muslim (0), is it forbidden to use this variable for predicting if someone is a jihadist?

If it's forbiddent to use X13, then is the use of the C4.5 algorithm forbidden?


* the "data" is crap, but that's a post for another day.

Ron said...

> The desire to prevent "A" has nothing to do with "B".

Of course it does. The resources you put towards preventing A are resources you can no longer deploy to prevent B.

Coincidentally, this came across the wires today:

https://www.washingtonpost.com/graphics/national/police-shootings/?hpid=z1

461 people shot dead by police this year. 62 in the last 30 days.

Publius said...

You're moving the pea (fallacy of relative privation).
The question isn't should we prevent A or prevent B (or prevent both).

The question is: how do we prevent A?

Ron said...

@Publius

Don't believe everything you read in wikipedia. The "fallacy" of relative privation is not actually a fallacy. In a world of finite resources, the question of which problems should have resources devoted to them and which should not is very much germane.

Publius said...

Ron's Brain vs. Ron's Writing

The "fallacy" of relative privation is not actually a fallacy.

Another name for it would be demagoguery.

In a world of finite resources, the question of which problems should have resources devoted to them and which should not is very much germane.

Yes, absolutely. Hundreds of other problems and questions are also worthy of consideration.

But . . . I think I see where I am misunderstanding you. Your post wasn't against profiling per se, but wrong probabilities and relative risk.

Breaking it Down

Here is how I would summarize your post into a series of "logical" statements:
1.0 Sam Harris called for profiling of Muslims, as they are overwhelmingly more like to commit acts of terrorism than non-Muslims
2.0 Since 9/11/01, 48 people have been killed by home grown extremists vs. 26 by self-proclaimed jihadists.
2.5 [implied - see below]
3.0 Sam Harris should issue a retraction

2.5 is an unstated conclusion in your statements. Some possibilities are:
i) Therefore, profiling Muslims is not effective or useful
ii) Therefore, Muslims are not more likely to be terrorists than non-Muslims

I conclude you meant "ii". I was reading "i" into it [which is a logical non-sequitur].

Left unsaid is whether you would support profiling of Muslims if the body counts were reversed.

Luke said...

Nowhere in either article did I see:

     (1) the population sizes each group is drawn from
     (2) profiling mechanisms in-place
     (3) thwarted plots with estimates of success and body count

Without these data, it seems your thesis is on pretty iffy grounds. In particular, it seems notable that there hasn't been an explicit foreign terrorist attack on American soil since 9/11. Had there been just one—like Umar Farouk Abdulmutallab the underpants would-be bomber—then your numbers might be completely different, and yet it's not clear a different conclusion would thereby be merited.

I'm also with Publius: I want to know if you're against profiling on principled grounds, consequentialist grounds, or some other grounds.

Ron said...

@Luke:

I'm not against profiling. I'm against *stupid* profiling. And trying to profile *Muslims* based on how they *look* (which is what Harris originally advocated) is stupid because:

1. You can't tell if someone is a Muslim by how they look.

2. Even if you could tell if someone is a Muslim by how they look, there are a billion Muslims in the world, and only a tiny fraction of them are terrorists.

That, of course, has nothing to do with this article, but you asked me to clarify my position.

The reason I think this article is relevant is because Harris and others claim that it is justified to profile Muslims because Muslims are uniquely predisposed to be terrorists than non-Muslims. (The slogan is that "Islam is the motherlode of bad ideas.") That is the argument that this article debunks. There are equally bad ideas to be found in non-Muslim traditions, and those bad ideas are every bit as likely to drive non-Muslim radicals to commit acts of terrorism as they are to drive Muslim extremists to do the same.

Muslims are not the enemy. *Extremists* are the enemy. To quote Sam Harris, who could fail to understand this?

Ron said...

@Publius

> whether you would support profiling of Muslims if the body counts were reversed

No. What makes these numbers debunk the theory that "[Muslims] are overwhelmingly more like to commit acts of terrorism than non-Muslims" is not that the non-Muslim number is bigger (though that does help to drive the point home). It is that they are *comparable*. And that they are *small*. There are three million Muslims in the U.S. and 7 acts of terror since 2001. This is a tiny, tiny signal. There are 300 million non-Muslims and 19 acts of terror. That is also a tiny, tiny signal. Both of these are virtually indistinguishable from zero, totally lost in the noise. Mix and match them however you like, there is no justification there for profiling anyone.

You don't want to profile Muslims or Christians, you want to profile *crazy people* no matter what flavor of crazy they happen to be. You want to profile *Islamists*, not Muslims. You want to profile white supremacists. Or something like that.

Luke said...

@Ron:

> Ron: 1. You can't tell if someone is a Muslim by how they look.

Very true, but profiling need only be probabilistically correct, not perfectly correct. Of course, that is also where the injustice lies.

> Ron: There are three million Muslims in the U.S. and 7 acts of terror since 2001. This is a tiny, tiny signal. There are 300 million non-Muslims and 19 acts of terror.

     7 / 3 = 2.3 Muslim terror acts per million
     19 / 300 = 0.063 non-Muslim terror acts per million

By this reasoning (which I don't accept; see my (1)), Muslims are 2.3 / 0.063 = 37x as likely to commit an act of terror. Now, you say that the numbers are too small for that 37x to mean anything; what is your justification for this? Again by this reasoning, it strikes me that if law enforcement agencies were to give Muslims 2x as much scrutiny as non-Muslims, there would likely be fewer terror acts per million, total.

I guess what was confusing is that when you said "there is actual data to inform this debate", it seemed like you thought the data issued a verdict other than "It is unknown whether Muslims are more likely to commit terrorism than non-Muslims." And yet, you seem to reject this verdict in a comment:

> Ron: There are equally bad ideas to be found in non-Muslim traditions, and those bad ideas are every bit as likely to drive non-Muslim radicals to commit acts of terrorism as they are to drive Muslim extremists to do the same.

Do we know, or not know, this "every bit as likely"?

> Ron: Muslims are not the enemy. *Extremists* are the enemy.

This can be true, simultaneously with the fact that Muslims are more likely to be extremists. Whether the latter is actually the case, I don't know. (see (1)–(3))

Ron said...

> profiling need only be probabilistically correct, not perfectly correct

OK, fine, let's do the math:

P(Terrorist|Muslim) = P(M|T) x P(T) / P(M)

P(M|T) = 7/26 = 0.27
P(M) = 0.01
P(T) = 26/300000000 = 0.000000009

P(Terrorist|Muslim) = 0.00000023
P(~Terrorist | Muslim) = 99.99998%

So the probability of false negatives is indistinguishable from 100% to seven significant figures.. That's about as far from "perfectly correct" as you can get.

Luke said...

I don't know why you ignored my 37x figure; did I make an error? As to your probabilities, you surely know that law enforcement will use a number of factors, many of which may be quite weak by themselves, but which can be significant when combined. Your reasoning would defeat the use of every single such factor, which seems to be a very bad idea.

If law enforcement has finite resources and has to surveil either a Muslim or a non-Muslim, ceteris paribus, a naive (and probably wrong) analysis of the numbers you provided indicates that the Muslim should be surveilled. Or have I made an error?

Ron said...

> I don't know why you ignored my 37x figure

Because 37 times a number nearly indistinguishable from zero is still a number nearly indistinguishable from zero. If you buy 37 lottery tickets your odds of winning the lottery are 37 times higher than if you'd bought one ticket. But you still, with nearly 100% certainty, will not win the lottery. (If you buy one lottery ticket your odds of winning are infinitely greater than if you'd bought zero lottery tickets. That doesn't change the fact that the expected value of every lottery ticket you buy is negative.)

> have I made an error?

More than one:

> If law enforcement ... has to surveil either a Muslim or a non-Muslim

What you meant to say is if law enforcement has to *choose* to surveil a Muslim or a non-Muslim (otherwise it's a tautology). But your premise if false: they don't have to make that choice. In fact, they *cannot* make that choice, because, as I have pointed out many times before, it is not possible to tell whether or not a given person is or is not a Muslim.

You don't have to buy lottery tickets. In fact, the number of tickets that maximizes your expected return from the lottery is zero, despite the fact that every ticket you buy increases your odds of winning.

Luke said...

Just FYI, I'm probably against profiling when the benefit is sufficiently low (but still extant), on the basis that we do not abridge the rights of the few for the benefit of the many. My concern here is that your reasoning, completely aside from such principles, is flawed.


> Because 37 times a number nearly indistinguishable from zero is still a number nearly indistinguishable from zero.

I don't see why this is relevant, given that I said:

> Luke: As to your probabilities, you surely know that law enforcement will use a number of factors, many of which may be quite weak by themselves, but which can be significant when combined. Your reasoning would defeat the use of every single such factor, which seems to be a very bad idea.

Your lottery ticket example simply doesn't capture the idea that one factor would indicate a 37x increase in probability, another a 50x increase, another a 12x increase, ..., such that all combined, the result is significant.

> In fact, they *cannot* make that choice, because, as I have pointed out many times before, it is not possible to tell whether or not a given person is or is not a Muslim.

I tried to respond to this before:

> Ron: 1. You can't tell if someone is a Muslim by how they look.

> Luke: Very true, but profiling need only be probabilistically correct, not perfectly correct.

> Ron: So the probability of false negatives is indistinguishable from 100% to seven significant figures.. That's about as far from "perfectly correct" as you can get.

Your math had nothing to do with whether one can be probabilistically correct in identifying someone as Muslim. So, that is merely a bald assertion—that likelihood to be Muslim is insufficiently correlated with appearance. (Note that if it were, we could go a bit beyond appearance and result in Harris being wrong in letter but possibly right in spirit—at least, according to principles which are held by a significant number of people.)

Ron said...

> Your lottery ticket example simply doesn't capture the idea that one factor would indicate a 37x increase in probability, another a 50x increase, another a 12x increase, ..., such that all combined, the result is significant.

First, this analysis tacitly assumes that all of these hypothetical factors would be independent of one another, and that is far from clear.

And second, this is a straw man because it is not what Harris was advocating. I have no idea what law enforcement is actually doing, and I'm not critiquing what law enforcement is actually doing. I'm critiquing Sam Harris's full-throated and unrepentant advocacy of profiling "Muslims, or anyone who looks like he or she could conceivably be Muslim", full stop. (Well, it wasn't quite full stop. He went on to say that "we should be honest about it" but that just cements the irony.)

And none of this addresses the real issue: even if you concede that profiling "Muslims, or anyone who looks like he or she could conceivably be Muslim" might be a good idea, how do you do it? The phrase presumes that there is such a thing as someone who looks like they could not conceivably be Muslim, and who can therefore be eliminated from scrutiny. What does such a person look like? Harris gives a few examples (old people, three-year-olds, and members of the Mormon Tabernacle Choir, by which he cannot possibly mean anything other than white males) but these are completely bogus. There are old Muslims, there are young Muslims, and there are white male Muslims. We can eliminate old people from scrutiny not because old people look like they could not conceivably be Muslim, but because very few terrorists are old. (Unfortunately, we can't eliminate young people on the same basis.)

Publius said...

And none of this addresses the real issue: even if you concede that profiling "Muslims, or anyone who looks like he or she could conceivably be Muslim" might be a good idea, how do you do it?

It appears that Google has already solved the problem.

But seriously, you don't determine "Muslim" or "not Muslim" by a person's appearance. Instead, you look up their name and birthday in your database. Given the software and database technology that exists today - where companies like Acxiom can deliver a detailed user profile to a website in less than 1 second - you just build a database of every Muslim in the world. Should Acxiom's software systems not be sufficient, I'm sure the NSA could deliver a system.

Or you could ask the person, "are you a Muslim?"

Ron said...

> It appears that Google has already solved the problem.

You're kind of making my point for me there.

> you just build a database of every Muslim in the world

George Orwell would be proud.

> Or you could ask the person, "are you a Muslim?"

Of course! Why didn't I think of that?

Publius said...

George Orwell would be proud.

What would he think of the Democrat Party having the religion of every person in the United States in their database?

Publius said...

Another Day

Following up to my 6/30 comment, the New American Foundation "study" has been debunked.
Popular press account of debunking.
The article by Prof. Holt

The NAF study left out a few things:
1) The extraordinary security measures implemented after 9/11 (e.g., the Patriot Act and the creation of Homeland Security) to prevent attacks by Islamic terrorists.
2) Leaving out several incidents of radical Islamic terrorism deaths.

One killing spree that I didn't know was Islamist related was that of the Beltway snipers, John Allen Muhammad and Lee Boyd Malvo, who in 2002 killed 10 people.

Publius said...

Update

International Security has updated their body count after the Orlando massacre. The Jihadists are now 2X ahead of the "far right wingers", even though they undercount the Jihadist attacks.