28 June 2012

Look to Shannon thou sluggard; consider his definitions and be wise.

The Economist: Graphic Detail | KNC | Tracking social media: The mood of the market

Markets, the classical economists instructed, rely on information. But what if there is too much of it, not too little?
Information is data which reduces entropy, or equivalently, helps you to make more accurate predictions. If you have "too much information," it's not really information.

(Yes, yes. I know there are many definitions of "information" and for many of them you can indeed have "too much." But I think using the sense of information I use above makes the whole situation KNC describes (the introduction of sentiment analysis data in quantitative finance) easier to understand. MarketPsych is not selling people information. It is selling them data. How much information is contained in that data is TBD.)

26 June 2012

Apple Stores

asymco | Horace Dediu | The face and the brand

The corollary to this would be that the job of the [Apple] store is not primarily to sell things. This is confirmed by the fact that store employees are not on commission and there is no sales pressure on visitors. Indeed, the sheer number of employees in a store of modest size (117 employees on an average of about 8k sq. ft.) implies a brazen disregard for the economic orthodoxy of retail efficiency and incentives to sell.
The last three times I was in an Apple store I literally could not get an employee to take my money. Each time I have been left standing in the middle of the floor with a product in one hand and my wallet in the other asking every uniformed employee I could get my hands on if they would please sell me this thing. Each time it took me ~90 seconds to find the product I needed on the shelf, but 10-15 minutes to track down someone who was authorized to let me exchange that product for money.

I don't want to second guess their strategy. It's obviously working for them. But come on. I'm doing all the work for them: just let me give you some money.

If we take a visitor-centric point of view, the analysis of the retail business simplifies considerably. We can even derive an “income statement per visitor”. Revenues are measured per visitor (currently about $52/visit). Costs are measured in terms of employee time. A $15/hr average wage and 15 min interaction time per visit puts the service cost at $3.75/visit. Profit per visit is also measurable and it’s currently at $13/visit. Therefore the employee “FaceTime” costs 7% of sales, while other costs, including the cost of goods sold, add up to 66% of sales.
$52/visit is a good mean, but I want to see the distribution of sales per visit. I bet the variance on that is huge.

25 June 2012

Mystery: Something which is Foo behaves like a Foo, and something which is not Foo doesn't

Free Exchange | S.C. | America's fiscal union: Blithe acceptance

In a blogpost on Friday, Paul Krugman pointed out that income gaps between euro-members are no greater than income gaps between American states. The difference, he writes, is that

we think of ourselves as a nation, and blithely accept fiscal measures that routinely transfer large sums to the poorer states without even thinking of it as a regional issue
There's a good reason for that: we actually are a nation, and Europe is not.

Mrs SB7 and I have frequent "fiscal transfers" between ourselves. My unmarried friend Anne and her boyfriend Bob do not. Would anybody look at that situation and decide the difference is that Mrs SB7 and I "think of ourselves as a family"?

24 June 2012

Measure results, not inputs

Hit & Run | Tim Cavanaugh | Orszag: Compulsory Voting for All!

Writing in Bloomberg, Orszag says it's "our own fault" we get such poor results when we vote:
Compulsory voting, as exists in Australia and more than two dozen other countries, would fix that problem. As William Galston of the Brookings Institution argues, “Jury duty is mandatory; why not voting?”

Mandating voting has a clear effect: It raises participation rates. Before Australia adopted compulsory voting in 1924, for example, it had turnout rates similar to those of the U.S. After voting became mandatory, participation immediately jumped from 59 percent in the election of 1922 to 91 percent in the election of 1925.
Those "more than two dozens other countries," by the way, include such envies of the world as Libya, Argentina, Congo, Lebanon, Egypt and Nauru. (Actually, Nauru seems interesting.)
Hrrrrrrrrrr. Talk about begging the question. Voting isn't an end in itself: voting is a method to get good policy. You can't judge the success of a law which dictates voting by counting how many people vote. The goal is better policy. Counting increased turnout as a victory would be like claiming that stomping your foot on the gas pedal is ipso facto a good idea because now the pedal is closer to the floor.

This Orszag dude, by the way, was one of the Certified Very Smart People who rode into town with Obama to replace all of Bush's supposed hillbillies and frat boys.  If this is the level of logic we get out of the technocratic, "reality-based," pro-science whateverthehellyouwanttocallit people, we're toast. The people I'm told are idiots act like idiots, and the people I'm told are geniuses act like arrogant idiots.

PS For the record, I think not only is compulsory voting is a bad idea, but even the "Rock the Vote" type haranguing is downright immoral. If people aren't already motivated to get themselves to a polling place then they almost certainly aren't motivated to stay informed and make coherent decisions. Encouraging people so unmotivated to exert control over society is irresponsible.

I would love to see a boxing match between Aaron Sorkin and David Mamet

That title has almost nothing to do with this post. But it is true. I would pay a lot of money to see that.
NPR: Monkey See | Linda Holmes | Sorkin's 'Newsroom' Is No Place For Optimism

The underlying thesis of The Newsroom is that the problems of TV news – no, the problems of news media – no, the problems of American political life – are really pretty easy to solve. What could turn things around, the story suggests, is one newsman who will look into a camera and speak the objective and easily discernible truth. And, it suggests, the only reason that hasn't happened anywhere (and is thus so revelatory) is that everyone in every media organization in the country is so obtuse that they've never thought of offering objective facts in a civil manner before, and is such a money-grubbing coward that they'd never do it if they did.
That's essentially the same problem I had with Sorkin's West Wing. Many of my conservatives friends think it's biased towards leftism; I think it's biased towards Great Men.* In Sorkinland all of America's problems could be fixed if only the right people were given the reins. This is the antithesis of how I see the problem: it's not who holds the reins, it's that the team they're hitched too is far too large.

Putting your faith in finding just the right people to be in charge is not only juvenile but also dangerous. We need to design systems to work even when the wrong people are in the driver's seat, because sooner or later (it'll be sooner) you always get the wrong people.

I find Sorkin's world deeply disturbing. It's cynical and a fairy-tale all at the same time. In my world there are political divisions because people fundamentally disagree about what is right, and there are structural issues which reinforce those disagreements. In his world there are divisions because everyone is an idiot, and everything could be fixed if everyone who is wrong just shut up and listened to the few Smart People who are right.

I used to think that Sorkin was born several centuries late, that he would have been an excellent playwright creating dramas to boost the egos of courtiers in the Doge's Palace or Versailles. It's too bad, I thought, that he missed his calling putting on plays that flatter the intellectual and political elite about how smart and powerful they are, and how everyone needs their wisdom and guidance. But I just realized something: he didn't miss his chance at all. That's exactly what he does now, except instead of flattering some inbred Bourbons who actually could change history with a proclamation, he's flattering all the people who wish they could do that, all the people who would be courtiers if such still existed.

* I used to think this was in large part a result of general Narrative Bias. Sorkin is telling stories, and stories have a small and finite number of characters, and so a small and finite number of people must be the ones solving all the problems. But then I meditated on The Wire a little more and realized that David Simon well and truly proved that good narratives do not need to fall back on a handful of relevant people around whom the rest of the narrative universe revolves. Fiction, even political fiction, does not need Great Men. Indeed, it is better off without it.

22 June 2012

On "let there be (some) amnesty"

Here are some unordered responses to two recent Bryan Caplan posts about Obama's semi-amnesty program. First, Caplan:
EconLog | Bryan Caplan | Is Obama's Semi-Amnesty for Real?

I don't just think that immigration restrictions are bad policy; I think they're a grotesque crime against humanity [pdf] — with all that implies. Given this starting point, Obama's semi-amnesty for hundreds of thousands of illegal immigrants sounds like the best U.S. political news of the 21st-century. I can't remember the last time any American policy change actually made me happy or even hopeful. I'd like to believe this is for real.

EconLog | Bryan Caplan | Six Theses on Extremely Unjust Laws I Dare You to Dispute
  1. Extremely unjust laws are conceivable.
  2. Extremely unjust laws exist.
  3. It is morally permissible to break an extremely unjust law.
  4. It is morally permissible to evade punishment for breaking an extremely unjust law.
  5. It is morally impermissible to enforce an extremely unjust law.
  6. It is morally permissible to punish a person for enforcing an extremely unjust law.
The laws underlying slavery or the Holocaust are obvious examples of extremely unjust laws. To dispute my six theses in any fundamental way, you have deny them for these notorious cases. To reject #4, for example, commits you to the view that it was morally impermissible for slaves to hide from runaway slave patrols. To deny #6, similarly, commits you to the view that the Nuremberg Trials were morally impermissible.

Why bring this up now? Because Obama's semi-amnesty predictably provoked the complaint that his action violates the "rule of law." Yet if you accept my six theses, any discussion about the rule of law is premature. The first question on the agenda has to be the justice of the laws Obama has decided to undercut. If we're going to argue about his decision, this is what we should be arguing about.
(I) Like Caplan, I find our immigration laws to be morally equivalent to apartheid. I would be happy to see them go.

(II) Unlike Caplan, I am not so cheery about how this was done, i.e., by way of Presidential fiat. His response is a good one, but ultimately unconvincing:
I say the laws on the books are so overwhelmingly wrong that even random Presidential nullification would be a huge expected improvement.

My question for Arnold: What's the best law any future President is likely to nullify due to Obama's precedent? I just don't see this slippery slope leading anywhere we should fear to slide.
I'm not used to arguing against people who hold US legislation in even more contempt than I do. Nevertheless, I believe I can think of several laws which are not good ones, but given all the rest of our legislation we're better off with them being enforced than not enforced.

For instance, I oppose both ObamaCare in particular and the idea of an individual mandate in general, but we'd be a lot worse off if we kept ObamaCare but POTUS decided not to impose the mandate. I don't think income tax is moral, but assuming the government keeps spending (safe assumption, I think)  I'd much rather have the income tax enforced than not enforced.  Several jurisdictions stopped enforcing all evictions during the housing crash. I think removing people from property they have not paid for, and is therefor not theirs, is perfectly moral. True, this wasn't a federal decision, but it's also not speculation: it actually happened. I think collecting payments on student loans is entirely moral, but I would not be surprised if a President ordered that those debts not be enforced. This particular President has already had all sorts of moral laws regarding bankruptcy abrogated. None of this is based on this immigration decision in particular as a precedent, but I think it sets a more general precedent that laws only exist at the pleasure and discretion of the President.

(III) Stephen Landsburg asks two very piercing questions:
  1. What morally relevant criterion protects the rights of the relatively rich foreigners who are already here but allows us to continue trampling the rights of the desperately poor foreigners we’re continuing to turn away at the border?
  2. Now that you’ve at least admitted it’s a bad idea to throw current residents out, what are we going to do about the folks you’ve thrown out over the past three-and-a-half years (and the ones your predecessors threw out before that)? They were, after all, studying in our schools and playing in our neighborhoods at the time. Will we apologize and invite them back?
(IV) Back to Caplan. I think he misses a couple of steps and a couple of dimensions in his six theses.

• The first of these is authority. I think the moral requirements are different for people in power. Having put yourself in a position of authority means you have, to a certain degree, forfeited the right to make these decisions. Having agreed to uphold the law, and promised your fellow citizens that you will do so, you don't then get to pick and choose which laws you like. That wasn't part of the deal by which Obama gained power. If you feel some laws are too immoral to enforce, don't put yourself in a position where it is explicitly your responsibility to enforce them.

I'd draw an analogy to Catholic clerics who want to teach doctrine the Church disavows. They don't get to pick and choose. They are, of course, perfectly free to teach whatever their heart leads them to believe. But they don't (shouldn't) get to do it while simultaneously remaining authorities of the Church. If you think premarital sex is a swell idea or think transubstantiation is silly, that's fine. But you don't get to say that and remain a bishop.

If Obama wanted this law to change (and I hope he did!) and thought it was too immoral to enforce, he should have remained a Senator, where he could work to change it and would not have taken on the responsibility of enforcing it. There's no way he was blindsided by this. There's simply no way he thought immigration law was all fine and dandy and then three years into his term realized it was immoral and refused to enforce it. He gained power with the full knowledge that he was also gaining the responsibility to participate in the enforcement of these laws.

I see the POTUS as I understand Washington did: as the Chief Magistrate of the country. I want a President whose job is to keep the machine of State in fine whack. I'm comfortable with a POTUS who falls back on "the forms must be obeyed" because I think his sole role is to do so, not to determine what the forms are. The power to determine what the law should and should not be is explicitly given to a different set of people. (And for a very good reason, I would add.)

• In a more general sense, I'm much more comfortable with people out of authority saying "screw this, I'm not coloring within the lines" than when people in power say that. I guess that most people would say they agree, but when push comes to shove I don't think they really do. That's just speculation though.

• I also think it's important to look at people's expectations. A President enters office knowing he's going to have to enforce immoral immigration laws. He goes in eyes-open, swears on the constitution that he's willing to do so, and then three years later decides not to. It's not like this was just sprung on him by congress out of the blue last month.

Imagine a young man joins the army to defend his country and then, years down the road, is ordered to slaughter POWs as part of a new policy. I hope and pray that young man has the courage to refuse to enforce his orders. But whether he does or not he's in a very different moral position than a recent recruit who signs up explicitly knowing that his army's policy is to execute prisoners and furthermore volunteered for prison guard duty. Obama is in the same moral position as the latter man, not the former.

• There's another temporal problem with adopting amnesty-by-executive-fiat: it can be reversed at any time. Laws need to be consistent. How is a young immigrant supposed to plan for his or his children's futures if at any time the deportation switch can be flipped back the other way? True, congressional legislation can be reversed, but that's a lengthy (and relatively rare) process. This immigrant children amnesty thing is something that people can expect to be reversed at least every four years on average, if we leave it up to the pleasure of the president.

Caplan would probably respond that these children are better with a temporary chance of amnesty than a permanent lack of chance, and there's definitely something to that. But I still think this temporariness makes this policy a pretty half-hearted lollygag of a nod toward morality.

• Maybe this is just me being cynical, but I don't believe Obama is refusing to enforce these laws because he believes them to be immoral. He's been perfectly willing to enforce them for years. He's deported a record number of people. There's no apparent moral principle for this new position (as Landsburg establishes). So no, I don't believe he's doing this because he thinks to do otherwise would be immoral. At best, he's doing it because he believes it's moral but more importantly electorally useful.

He may actually think this is the right thing to do, but he wasn't willing to do it until it was politically useful. That means morality is second fiddle for him. You don't get to hide behind realism when its convenient and then brag about being an idealist when its opportune.

I'm not going to congratulate Obama for accidentally blundering into the moral position like Caplan seems willing to do.

• Caplan's theses also ignore structural issues. If I, a random citizen, refuse to help enforce a law, the structure of law does not change. If our President, the chief law enforcement officer of the nation, picks and chooses then the entire system of legality collapses. That is a meta-change.

Maybe the President chooses not to enforce an immoral law. Then we all win some. But we've put all our eggs in one basket, labeled "The Moral Sense of the Current President." That's a dangerous situation. If you or me or John Q. Beatcop refuses to participate in immoral legislation we don't face any such structural danger.

Our legislation is a loose approximation of morality. In many ways it is immoral, but our system of generating laws is moral one. (Well, as moral as anything else in wide use.) Caplan would have us trade more moral legal particulars for a less moral law generating system. His apparent position is that our system has already generated so many immoral laws that we're better off without it. That's consistent at least, but every time I pick up a history book I'm shocked about how much immorality results from leaving legislation up to the moral sense of the head honcho.

21 June 2012

Use gallons per mile!

Via Tyler Cowen:
StrategyProfs.net | Steve Postrel | Marginalism and the Higher Ed Paradox

Typical graduate business school education has indeed become less rigorous over time, as has typical college education. But typical high school education has declined in quality just as much. As a result, the human capital difference between a college and high-school graduate has increased, because the first increments of education are more valuable on the job market than the later ones. It used to be that everybody could read and understand something like Orwell’s Animal Farm, but the typical college graduates could also understand Milton or Spencer. Now, nobody grasps Milton but only the college grads can process Animal Farm, and for employers the See Spot Run–>Animal Farm jump is more valuable than the Animal Farm–>Milton jump.

So the value of a college education has increased even as its rigor has declined, because willingness to pay for quality is really willingness to pay for incremental quality. This principle holds true in many markets. For example, a roof with mean time to failure of 5 years is a lot more valuable than one with a MTF of 2 years, but a 25-year MTF isn’t that much better than a 22-year MTF for most owners. A fuel economy increase from 12 to 15 miles per gallon is a bigger deal than an increase from 27 to 30 MPG.
Yes, yes I agree with his point. But that is a truly terrible analogy. Truly, truly, terrible.

The increase from 12 to 15mpg is not just better in some perceptual sense than the increase from 27 to 30mpg, it is larger in a literal, arithmetic way! It's not about margins or incremental quality, it's just straight up quality. This is why we should use gallons per mile. Doing so makes it immediately apparent that the first shift (12→15mpg) is bigger than the second (27→30mpg).

car miles/gallon gal/100 miles cost per 10,000 miles difference
A 12mpg 8.33 $2918.83 $585.17
B 15mpg 6.66 $2333.66
C 27mpg 3.70 $1296.48 $129.65
D 30mpg 3.33 $1166.38

(Using today's average gas price of $3.504/gal.)

Thinking about mpg makes it look like the difference between A & B and C & D is the same, but it's obvious looking at the third column that they're not. I've covered this before:
People don't realize that improving from 10 to 11 mpg saves as much gas as improving from 33 to 50 mpg. To rephrase [Fuqua's] closing argument, "Which is more useful to know: How far you can drive on a gallon of gas? Or, how much gas will you use to cover a certain distance?" MPG answers the first question. GPM answers the second question." The first question is valuable when gas stations are scarce and population thin, and you need to know if you can make it to the next stop, which is to say it's valuable if you live in the desert or the 1930's. These days we're far more interested in "How many gallons of gas (or dollars) will it take to commute X miles to work 200 times this year plus trips to grandma's house and the beach?"

I'm going to take this opportunity to bitch about something else mileage-related that's been grinding my gears.

Chevy has been running these adds for the Volt in which owners (or actors pretending to be owners?) brag about how infrequently they have to gas up. One of them ends with a woman saying she's using her savings from not buying gas to go on vacation.

I find this unbearably dishonest, preying as it does on American innumeracy and general financial kunckle-headedness. The reason is simple: she already spent her gas savings on the Volt! She spent a lot of extra money (over $10,000) to get the Volt. That's the money she isn't spending on gas right there. She can't double count those dollars to spend them on vacation too. Not until she's out of the payback period from that initial outlay, which is probably longer than she'll own the car!

If that Edmunds article is too much, I'll let Wikipedia spell it out:

According to Edmunds.com, the price premium paid for the Volt, after discounting the US$7,500 U.S. federal tax credit, takes a long time for consumers to recover in fuel savings, often longer than the normal ownership time period. Edmunds compared the Volt (priced at US$31,712) with the same-size gasoline-powered Chevrolet Cruze (priced at US$19,656) and found that the payback period for the plug-in hybrid is 15 years for gasoline prices at US$3 per gallon, 12 years at US$4 per gallon, and drops to 9 years with gasoline prices at US$5 per gallon. At February 2012 prices, the break even period is 14 years. These estimates assume an average of 15,000 miles (24,000 km) annual driving and vehicle prices correspond to Edmunds.com's true market value estimates.
To make matters worse, I think Edmunds ignored the time value of money, so the payback periods will be even longer.

The Dude is now done carping about gas.

20 June 2012

TSA & Private Screeners

Schneier on Security: Rand Paul Takes on the TSA: Paul Rand has introduced legislation to rein in the TSA. There are two bills:
One bill would require that the mostly federalized program be turned over to private screeners and allow airports with -- Department of Homeland Security approval -- to select companies to handle the work.
This seems to be a result of a fundamental misunderstanding of the economic incentives involved here, combined with a magical thinking that a market solution solves all. In airport screening, the passenger isn't the customer. (Technically he is, but only indirectly.) The airline isn't even the customer. The customer is the U.S. government, who is in the grip of an irrational fear of terrorism.

It doesn't matter if an airport screener receives a paycheck signed by the Department of the Treasury or Private Airport Screening Services, Inc. As long as a terrorized government -- one that needs to be seen by voters as "tough on terror," wants to stop every terrorist attack regardless of the cost, and is willing to sacrifice all for the illusion of security -- gets to set the security standards, we're going to get TSA-style security.
I think Scheneier is partially right, but still missing something. Privitization would not solve the major problems with airport screening, but it would address a lot of the smaller ones.

For instance, the TSA can vandalize David Friedman's luggage in a way that private companies can not afford to do. (See also: Is TSA Vandalism Deliberate Policy?) Privitization won't much help if your objection is the scanning and x-rays and pat-downs and such. But if you're worried about things being stolen from luggage, screener incompetence, rudeness, spite, imperiousness, etc., then privatization will help. (Assuming, of course, that the airport has a profit motive to increase customer satisfaction.)

17 June 2012

"The Iron Law of Shoes"

Via Marginal Revolution:
Journal of Research in Personality | Omri Giliath et al. | Shoes as a source of first impressions

Surprisingly minimal appearance cues lead perceivers to accurately judge others’ personality, status, or politics. We investigated people’s precision in judging characteristics of an unknown person, based solely on the shoes he or she wears most often. Participants provided photographs of their shoes, and during a separate session completed self-report measures. Coders rated the shoes on various dimensions, and these ratings were found to correlate with the owners’ personal characteristics. A new group of participants accurately judged the age, gender, income, and attachment anxiety of shoe owners based solely on the pictures. Shoes can indeed be used to evaluate others, at least in some domains.
This is only interesting if you routinely get to see just a stranger's shoes, and none of the rest of them. This has happened exactly one time in my life: an artificially constructed scenario during college orientation.

The more relevant study would be to determine how much more accurately (if at all) people can judge others when they see everything-including-shoes than when they see everything-except-shoes. For instance, if I can see you but not your shoes already, letting me peak at your feet will give me very little extra ability to judge your gender or age.

These researchers studied the correlation between shoe and personal characteristics. I want to know the information gain of shoes for personal characteristics.

PS I took a (very, very) cursory look at this paper, and the authors looked for correlations between shoes and at least 11 personal characteristics (ie, they report results from 11; I don't know how many others they tested). Shoe observers were able to predict four of those characteristics, including the (fairly obvious) gender. The reportage on this article could very well be "observers unable to judge most things about people by looking at their shoes."

15 June 2012

"The Google-Trolley Problem"

Following up from the previous post on the ethics of autonomous vehicles:
Marginal Revolution | Alex Tabarrok | The Google-Trolley Problem

As you probably recall, the trolley problem concerns a moral dilemma. [...]

I want to ask a different question. Suppose that you are a programmer at Google and you are tasked with writing code for the Google-trolley. What code do you write? Should the trolley divert itself to the side track? Should the trolley run itself into a fat man to save five? If the Google-trolley does run itself into the fat man to save five should Sergey Brin be charged? Do your intuitions about the trolley problem change when we switch from the near view to the far (programming) view?

I think these questions are very important: Notice that the trolley problem is a thought experiment but the Google-trolley problem is a decision that now must be made [because Google has driverless cars].
I think it's most important to think of this not as a dilemma but as a trilemma:
  1. The robot car runs over one person; one person dies.
  2. The robot car runs over five people; five people die.
  3. Robot cars are outlawed because some people do not trust them to make them to make ethical decisions between #1 and #2; ten thousand people die because human drivers are terrible.
Sadly, status quo bias agitates hard for #3.

Robocar Ethics

The Economist | Morals and the machine: As robots grow more autonomous, society needs to develop rules to manage them: ethical robots

One way of dealing with these difficult questions [of machine ethics] is to avoid them altogether, by banning autonomous battlefield robots and requiring cars to have the full attention of a human driver at all times. [...] Driverless cars are very likely to be safer than ordinary vehicles, as autopilots have made planes safer. Sebastian Thrun, a pioneer in the field, reckons driverless cars could save one million lives a year.
Autonomous cars will have to make ethical decisions. But note:
  1. In a lot of these cases we don't even know what the "right" decision is for people, so why would we expect people's tools to make the "right" decision?
  2. Even when we do, people often chose incorrectly due to pressure.
  3. In any situations we let people off the hook for their decisions, we ought to be willing to let their tools off as well.
  4. Making the demand for ethical machines too strongly may actually be highly unethical, for the reason given by Thrun above.
If my dog gets run over by a human it's small consolation that at least the human is capable of being ethical.

Why give legislative favor to moral people, who nonetheless routinely make bad decisions over amoral machines whose decisions rarely result in tragedy? When it comes to road safety I would rather have one person injured by an amoral machine than a thousand people injured by other human drivers, who do have moral agency but also have too much to drink, and talk on their cell phones, and get distracted by their kids fighting, and are plain and simple bad at driving.

Later on the Economist says that we might limit robots to techniques which are open to audit, so they (or their owners, or makers) can explain why they made the decisions they did. This means, as they point out, neural networks might not be allowed:
First, laws are needed to determine whether the designer, the programmer, the manufacturer or the operator is at fault if an autonomous drone strike goes wrong or a driverless car has an accident. In order to allocate responsibility, autonomous systems must keep detailed logs so that they can explain the reasoning behind their decisions when necessary. This has implications for system design: it may, for instance, rule out the use of artificial neural networks, decision-making systems that learn from example rather than obeying predefined rules.
Driving is one area where neural networks have excelled for a long time (e.g. ALVINN) and seem likely to in the future. So the Economist is just replaying the ethics-vs-outcomes debate on a smaller scale and this time they come to the opposite conclusion. They're saying they're willing to trade a more ethically-understandable system — not actually one which is more ethical, just one whose decisions are easier to pass ethical judgement on — for one which creates more ethical outcomes. They'll sacrifice ethics for road safety in general, but in the specifics they want to do the opposite: they'll sacrifice safety in order to benefit ethics.

If you're willing to allow amoral machines for the sake of reduced harm, you should be willing to allow amoral neural networks for the sake of further reduced harm. I don't see what principal allows you to draw a line in the middle allowing you to accept safe-but-amoral machines as long as they aren't too safe. You either except good outcomes as a relevant measure or you don't.

13 June 2012

Smart people being stupid (?)

Via JamulBlog and 3qd:
New Yorker | Jonah Lehrer | Why Smart People Are Stupid

West also gave a puzzle that measured subjects’ vulnerability to something called “anchoring bias,” which Kahneman and Tversky had demonstrated in the nineteen-seventies. Subjects were first asked if the tallest redwood tree in the world was more than X feet, with X ranging from eighty-five to a thousand feet. Then the students were asked to estimate the height of the tallest redwood tree in the world. Students exposed to a small “anchor”—like eighty-five feet—guessed, on average, that the tallest tree in the world was only a hundred and eighteen feet. Given an anchor of a thousand feet, their estimates increased seven-fold.
First: understanding cognitive bias is important. Okay, now that that's out of the way, here's my problem with these Kahnenman-type studies:

The types of questions people use in the lab to expose biases and cognitive failures is alien to the way people actually think. Many things that are clearly errors from a strictly objective calculation perspective are very useful in the real world. Useful enough that I think they may not be built-in biases, but learned adaptations.

Let's take an analog of the redwood situation. If a friend asks you if you think his suitcase is over or under 50 pounds, that's probably not idle speculation on his part. He probably knows something about the weight of the suitcase that you don't. Maybe he's used it before, and has memory of its weight in the past. In that case is makes a lot of sense to anchor your estimate somewhere around 50lbs. If the real weight of the suitcase was 10lbs or 125lbs your (rational, observant) friend wouldn't have asked you if you thought it was over or under 50lbs. In the real world that framing question is often relevant; in the lab it isn't. How valid of a conclusion is it to go "Aha!! People are distracted by the often-relevant question specifically chosen to be irrelevant in our study!"
And here’s the upsetting punch line: intelligence seems to make things worse. The scientists gave the students four measures of “cognitive sophistication.” As they report in the paper, all four of the measures showed positive correlations, “indicating that more cognitively sophisticated participants showed larger bias blind spots.” This trend held for many of the specific biases, indicating that smarter people (at least as measured by S.A.T. scores) and those more likely to engage in deliberation were slightly more vulnerable to common mental mistakes.
A lot of the questions these researches use are structured like the "word problems" we all had for 13 years of primary and secondary education. We all got good at thinking the way the people who write word problems want us to think, which often means accepting a very simplified world view, ignoring outside knowledge, etc.*
I suspect one of the things "smart" people learn with these tests is not to overthink things: you don't want to give the right answer, you want to give the answer the test-maker wants you to give. That's often a stupid answer. For a fictional representation, refer to Law-rence Waterhouse's US Navy induction test in Cryptonomicon
The questions in these research studies are presented like K12 word problems, but don't actually function like that. The researchers shouldn't be surprised that people answer them as if they are K12 word problems.

West et al. are surprised that people who did better on standardized tests had larger biases in their studies. But these are people who have trained themselves (or are naturally good) at accepting the universe of the test question. Just like a friend doesn't ask you if their 15lb bag is over 50lbs, a test writer doesn't ask you if the tallest tree is over 85ft if it's really a few hundred. That initial question is not there randomly; it's actually an additional piece of information that should give you a clue about the rest of the test.

When you expose people who are good at uncovering and using that information to a situation that is superficially the same as the tests they are accustomed to, but now being written by a person with completely different objectives to test a completely different skill set, you should not be surprised that they use their test-taking skills anyway. This is interesting, but I don't think it tells us as much about intelligence and cognitive bias as the Kahnemans, Wests, and especially Lehrers of the world would like us to think it does.

12 June 2012

Regulating Algorithms

The Guardian | Cory Doctorow | Google admits that Plato's cave doesn't exist:

Google implies that a page of search results is effectively the table of contents for a custom-made magazine assembled on the fly in response to a user's query

Google has, to date, always refused to frame itself [in an editorial position]. The pagerank algorithm isn't like an editor arguing aesthetics around a boardroom table as the issue is put to bed. The pagerank algorithm is a window on the wall of Plato's cave, whence the objective, empirical world of Relevance may be seen and retrieved.

That argument is a convenient one when the most contentious elements of your rankings are from people who want higher ranking. "We have done the maths, and your page is empirically less relevant than the pages above it. Your quarrel is with the cold, hard reality of numbers, not with our judgement."

The problem with that argument is that maths is inherently more regulatable than speech. If the numbers say that item X must be ranked over item Y, a regulator may decide that a social problem can be solved by "hard-coding" page Y to have a higher ranking than X, regardless of its relevance. This isn't censorship – it's more like progressive taxation.
That strikes me as the exact opposite of the truth. You have some numbers, you have some mathematical function, you get some output. The end. A regulator can decide that the output should be different, he can order you to claim the output is different, he can fine and jail you for refusing to state that the output is different. But the output is the output. Reality is reality. A regulator can demand that student pocket calculators must report that 23*3=70, but that neither changes the maths nor is it easier than regulating speech.

And no, this would be nothing like progressive taxation or censorship, except in that all are forms of absolutist dirigisme.

I see what Google is trying to do here, but if they had gifted lawyers they would make their arguments one level higher up. Their search results are not free-speech protected editorial content. Those are still the results of objective calculations. But their design of the PageRank algorithm is subjective and editorial. They are exercising a meta-editorial technique unavailable to people in pre-computational societies. The design of the algorithm is a form of free speech, the results it produces are mathematical and unarguable. If the judges are smart enough to think on two different levels of abstraction*
not a sure thing, to be sure
then they're in the clear.

The implication is that Google has discovered a mathematical model of relevance, a way of measuring some objective criteria that allows a computer to score and compare the relevance of different web-pages.

But there is no such mathematics. Relevance is a subjective attribute. The satisfaction you experience in regards to a search-results page is generated by your mind, and it reflects the internal state of your neurons just as much as it reflects the external reality of the results.
Yes, relevance is subjective. But once you have some definition or goal or standard you are using as an approximation for relevance then matters are not subjective any longer. Minwise hashing is minwise hashing, no matter how you define "relevance." (Minwise hashing is family of fast techniques used, among other things, for measuring similarity between sets of large documents, and thus useful for search.) There is a difference between subjectivity in the definition of the problem space and subjectivity in algorithms which run in that space.

For instance: what constitutes a noisy nuisance is subjective. Indeed, the very concept of "loudness" is subjective. But the mathematics of sound pressure, and the circuits of a sound level meter are not subjective. We can write perfectly objective rules regarding sounds over certain decibels after certain times at night despite the subjectivity of loudness or noise or nuisance.

Biometric phones

Via Tyler Cowen, who comments "I like this one:"
NY Times | 32 innovations that will change your tomorrow

A team of Dutch and Italian researchers has found that the way you move your phone to your ear while answering a call is as distinct as a fingerprint. You take it up at a speed and angle that’s almost impossible for others to replicate. Which makes it a more reliable password than anything you’d come up with yourself. (The most common iPhone password is “1234.”) Down the line, simple movements, like the way you shift in your chair, might also replace passwords on your computer. It could also be the master key to the seven million passwords you set up all over the Internet but keep forgetting.
I don't want to disparage these reserchers; I havent read their papers and I don't have time to soon. But from what I know from hanging around a biometrics lab, this is exactly the sort of thing that works great until it doesn't.

I've heard schemes to unlock computers based on the unique rhythm of people's typing, to enable car ignition based on how people get in, and to unlock doors based on people's walking gates. They all work great in the lab, then you get in the real world and someone can't unlock their laptop because they have a blister on their thumb or a split on their finger, and they can't use their phone because they have a brimming cup of coffee in their dominant hand, and they can't get in their office because they have a heavier-than-usual gym bag over their shoulder. Then people get frustrated because the machine is locking them out, and it's completely opaque about why, so they start thinking about how they're typing or walking or lifting their phone, and then it's even worse because they're doing in consciously not unconsciouly, essentially trying to imitate themselves, and the whole thing spirals into resentful anger.

Maybe this system for phones works great. I'd like to see more things like this, where our computing devices recognize us instead of us actively authenticating ourselves. Beyond the general problem, I see two specifics with this phone scheme.

(1) How do you unlock the phone when you're not answering a call? What if you're answering a text message, or just want to look something up, or initiate a call? Do you have to put the phone up to your ear as if you're answering a call, then put it back down again to look at it? And do people do this the same was as if the call was real?

(2) How will my phone respond when this occurs:
Mrs SB7: Honey, your phone is ringing.
SB7: Who is it?
Mrs SB7: Brian.
SB7: Can you get it; I'm up to my elbows in raw chicken. Just tell him we'll be there at seven.
I suppose you can have a back-up or override password for cases like this, or for when your phone isn't responding because, e.g. you're trying to answer while jogging across a parking lot in rain instead of sitting calmly at your desk. But (a) there's nothing to stop an attacker from using this secondary system to bypass their inability to mimic you biometrics, and (b) if people pick terrible passwords when passwords are the only means of security, won't they choose even worse passwords when they are the secondary, less-used, method?

Again: don't take this as a specific criticism of the unnamed Dutch & Italian researchers. I only make this critique because this is exactly the sort of research that gets put into some mangled and fanciful university press release and unskeptical journalists pass on without really understanding and then never comes to fruition and then subliminally causes people to be disappointed by the progress of technology.

PS I have not kept up with this research in about five years, but last I knew it was believed that the shape of your ear could be used as a unique identifier. If phones have front-facing cameras (and you have a short haircut) that would be an interesting way of unlocking the phone, at least for answering calls.

People consumed instead of investing; wealth dropped. Shocking!

NY Times | Binyamin Applebaum | Family Net Worth Drops to Level of Early ’90s, Fed Says

The recent economic crisis left the median American family in 2010 with no more wealth than in the early 1990s, erasing almost two decades of accumulated prosperity, the Federal Reserve said Monday.
But the problem is that we did not actually accumulate that wealth. Income came in, and people consumed it. There was no accumulation. (It may look like there was some accumulation, but the single largest factor there was mis-marking the value of people's largest asset.)

If you get a season of heavy rainfall, but keep using it all on throwing lavish water balloon fights, you can't stare down into your dry cisterns at the beginning of the dry season and say "where did all our accumulated water go?"

PS The converse of this statement is a good summation of why I don't care much about income inequality:
The data does provide the latest indication, however, that the recession reduced income inequality in the United States, at least temporarily.

08 June 2012

Radio Tragedy!

Car Talk Staff Blog | Tom & Ray | Time to Get Even Lazier

Work-Averse Brother Decides that Even One Hour a Week Is Too Much — An Actual Semi-serious Note from Click and Clack to Their Listeners

Now I will never win the puzzler, never find out if my old mechanic was scamming me with his "seized wheel" thing, and never know if I can (jokingly) blame my wife for killing my first car.

I'm not a petrolhead by any standard, but I don't think I've missed a single Car Talk since sometime in 2004.

(See also: LA Times | Rene Lynch | 'Car Talk's' last lap: Click and Clack to retire; fans mourn)

07 June 2012

If they don't have right, they don't have rights

LifeHacker | These Are the Companies that Protect Your Data from the Government (And the Ones That Don't)

Just a reminder to anyone out there who has their undies in a twist over Citizens United: a "People's Rights Amendment" would put every single company in the country on the "Don't Protect Your Data from the Government" list. Or more properly, the "Can't Protect Your Data" list, because they would have no rights to privacy, security from search & seizure, petitioning of congress, etc.

Companies not having rights means "companies won't have rights," not "companies won't have rights to do things I dislike such as criticize my favorite politicians but will still be able to exercise all the rights I like, especially the ones that protect me."

But if that's what all you statists want, go right ahead.

06 June 2012

Socialism is Broken, and Only More Socialism is the Answer

DCist | Martin Austermuhle | Capital Bikeshare's Socialism is Broken, and Only More Socialism is the Answer

Last week, Washington Times columnist Charles Hurt made quite the splash by calling Capital Bikeshare "broken-down socialism." Today we admit that he might have been right.

A few Bikeshare users—including DCist's own former editor-in-chief, Aaron Morrissey—complained this morning that every bike at their closest station was gone. This isn't an isolated complaint, of course—plenty of stations see heavy use during rush hour, notably along popular commuting routes.

In a sense, the quasi-socialism of Capital Bikeshare fails in this respect. Weirdly enough, though, it fails in a very capitalistic way: the bikes operate based on supply and demand, with the most sought-after bikes going most quickly. The only response is for Bikeshare to move bikes back to the most-used locations as quickly as possible, or simply expand capacity to meet demand. In essence, we need more socialism to make the system work—not less.
I don't think Austermuhle understand what "capitalism" and "socialism" mean. Capitalism is not "a system where supply and demand exist." That's all systems. Socialism is not "a system where someone gives me more of what I want." Though that's what most First World socialists often seem to think.

There is always, always, ALWAYS supply and demand. There is no escaping desire (ie demand) and scarcity (ie supply). None. People throughout history have tried to mask them, paper over them, ignore them, and destroy people's ability to measure them — all hoping that would somehow eliminate the supply and demand themselves, but it's impossible.

The difference between capitalism and other systems is not the presence of supply and demand, but how they are balanced: through markets operating in response to price signals. Capital Bikeshare currently charges one price no matter what the state of bike supply or bike demand. If they wanted to make sure that they didn't have rush-hour shortages they could raise the price of their bikes during busy hours, or raise the price to rent a bike from stations with dwindling supply. That would be the capitalist way,* not just a top-down directive that "there shall be more bikes."

* To the extent you can even meaningfully consider a single firm to be an entire economic system categorizable by "capitalism" or "socialism" to begin with, which isn't the best assumption to make.

PS I love that Austermuhle lists two things as "the only response" to having supply run out, neither of which is prices, and one of which is "just have more stuff." The latter in particular is amusingly juvenile.

02 June 2012

Misunderstanding Misfits

The Economist: Schumpeter Column | In praise of misfits: Why business needs people with Asperger’s syndrome, attention-deficit disorder and dyslexia

Entrepreneurs also display a striking number of mental oddities. Julie Login of Cass Business School surveyed a group of entrepreneurs and found that 35% of them said that they suffered from dyslexia, compared with 10% of the population as a whole and 1% of professional managers. Prominent dyslexics include the founders of Ford, General Electric, IBM and IKEA, not to mention more recent successes such as Charles Schwab (the founder of a stockbroker), Richard Branson (the Virgin Group), John Chambers (Cisco) and Steve Jobs (Apple). There are many possible explanations for this. Dyslexics learn how to delegate tasks early (getting other people to do their homework, for example). They gravitate to activities that require few formal qualifications and demand little reading or writing.
What?! Dyslexics become entrepreneurs because they have experience cheating on their homework? What mental slag heap did they drag that idea out of?

Take dyslexia out of the picture. Who would possibly contend that the people most likely to be entrepreneurs are those who are good at passing work off onto others? If that's your goal you're far better off in a big, byzantine organization than going it alone. If you want to avoid responsibility for your own work why in the name of St. Drucker would you start your own business?

Here's an alternative explanation. If you have dyslexia you learn early that you have two options: use it as an excuse, or work harder than everyone else to overcome your limitation.* The ones who choose the second path make natural entrepreneurs. We live in a world of written language. Since six or seven years old dyslexics have been surrounded by people who move through this world effortlessly, seemingly without thought, while they've had to grab it by the short hairs and bend it to their will. Having succeeded in doing that to language, why wouldn't they do the same to business?

* I am sickened by the number of students in Mrs SB7's classes who tell her they should have less work because of their learning impairments. They have it exactly backwards. I'm dyslexic. I didn't need less work, I needed MORE. That's your only hope of overcoming your bad neurocognitive luck. Gear the hell up and work harder than everyone else instead of wasting time whining to your teacher about how its not fair that you "only" get twice as much time as everyone else to do the assignment instead of an infinite amount of time you want.

There is, of course, an entire bureaucratic structure arranged to enforce and normalize these students' outlook; something like half of her students have official "accommodations" from the county.

Recruiters have noticed that the mental qualities that make a good computer programmer resemble those that might get you diagnosed with Asperger’s syndrome: an obsessive interest in narrow subjects; a passion for numbers, patterns and machines; an addiction to repetitive tasks; and a lack of sensitivity to social cues.
Maybe programmers do like repetitive tasks. I don't know; I'm not a sociologist. In my experience this isn't the case. But I do know this: if your programming is a series of repetitive tasks, you're a shitty programmer. Paul Graham (or Joel Spolsky? or...? I can't find the quote) said that if you're typing the same line over and over, or using copy-paste much, you're Doing It Wrong. If your coding is boring or repetitive you're at the wrong level of abstraction: move up until there is no repetition.

01 June 2012

Local Education Foundations

The Atlantic | Laura McKenna | How Good Parents and Good Intentions Lead to Dramatically Unequal Schools

Local education foundations (LEF) are 503(c)(3) groups formed in local school districts. They usually provide funding for auxiliary educational activities or after school activities, like Lego Leagues, music programs, or special field trips. Unlike PTA groups, they tend to focus on gathering large scale donations from local businesses or corporations.

There are wide variations in how much these groups can collect based on the wealth of the community. [...]

In Slate magazine in November, Helaine Olen described the impact of wealthy school foundations in California.
One elementary school is so adept at getting its wealthy parents to open their checkbooks that it is able to spend an additional $2,000 per student on enrichment activities, which include employing multiple reading and instructional assistants, as well as classes in chorale music, marine science, and art. But at another Santa Monica-Malibu Unified school, located just a few miles away, a significant percentage of the kids come from economically disadvantaged homes and the local PTA can't even muster up an additional $100 per child, leaving the students to make do with a truncated music program, a few art classes, and one measly instructional assistant.
I think there are two mutually exclusive ways to look at the justice of situations like this one. Option 1: LEFs are Pareto improvements. Some children are made better off, other stay the same. This is good. Option 2: the children who are already better off get further improvements, increasing inequality. This is bad.

I wish there were handy labels for people whose primary concern is the relative size of the pie slices and those whose concern is the size of the pie. That's a much more useful spectrum to discuss politics than left/right.
Reich also writes that individuals and business who donate to these foundations receive tax breaks. In other words, government is subsidizing foundations that channel money to wealthy school districts. The poor school districts that are unable to create these foundations receive nothing. Ultimately, charitable donations are not going to areas where it is needed most.
Well yes, but that's true of all sorts of donations. Museum patrons are of above-average wealth, as are familes who send children to college. Is allowing tax deductions for contributions to these institutions a "subsidy for the wealthy"?

Maybe your answer is "yes." That's fine. So long as your answer accords with how you would answer that question w.r.t. LEFs.

(I'm actually pretty open to the idea of ending tax deductions for charitable donations. I just want to do it consistently rather than drawing up an ad hoc list of "bad," unpopular charities like LEFs that don't deserve that tax treatment.

Okay I didn't intend to get into a digression about this, but I can't help justifying my position here since I think I said the opposite on this blog some years back. The short answer is that some portion of donations are actually consumption in another form. Furthermore I don't think non-profit status lends any kind of moral difference from any other organization. (They can have lots of profits, they just have to spend them.)

In recognition that some non-zero portion of contributions is concealed consumption, perhaps a compromise whereby donations of $X lowers your taxable income $0.5X is appropriate.)

Also, shouldn't we be happy that LEFs target corporate donors rather than local residents? I imagine it's much easier to get the car dealership to contribute to schools in the poorer neighborhood than it is not get rich citizens to contribute to a school on the wrong side of town.

PS "Dramatically Unequal Schools"? Let's say the typical LEF in a rich neighborhood comes up with $1000 per student, and the typical school district spending per pupil is $12500. Let's further make the very unlikely supposition that there is no other transfer from richer to poorer students' families, i.e. there is no progressive taxation in the district, the poorer school gets no grants the richer school does not, the booster club at the richer school does not contribute to a district-wide fund, etc. In that case the LEF would make spending at the richer school a dramatic eight percent higher. Eight percent's not nothing. I sure could use an 8% revenue boost in my budget. But seeing as how educational outcome is almost entirely unresponsive to orders-of-magnitude increases in spending, is 8% really that dramatic? (See also: "GDP Yield Per Dollar in Education Spending.")

Remember, spending isn't the point. Increasing knowledge and intelligence is the point. Spending is just what's convenient to measure.

Speaking of which, an extra $1000 contribution from Mr and Mrs Moneypants to their LEF appears to increase inequality by $1000. But this is only the case if none of that money would have been spent on their children had the LEF not existed. If they give the school a hunk of money for after-hours language program that shows up in the statistics as an increased expenditure in the rich neighborhood. But if they instead spend that money on a tutor for their child that will never show up in the numbers, making the distribution appear to be more even despite the opposite being the case.

A nearby town with an average income of $45,000 does not have a foundation, but it could use one to provide basic necessities for the school like textbooks and classroom supplies.
I'm really tired of the "they can't even afford textbooks!" cry. That's only going to sway me if you can also demonstrate that there isn't any fat anywhere else in the system. You want books? Fire a supernumerary vice principal. Fire the director of sustainability programs. Stop spending half a million on new furntiture for the headquarters building. (*AHEM* Baltimore!) Then you can buy literally truckloads of books.

The "we can't afford books" thing is the school equivalent of the way the parks service threatens to close the Washington Monument whenever they don't get the budget they want. They try to frame the debate as as dichotomy between getting all the funding they want and making the most dramatic, visible cut possible. Not the most reasonable cut or the most efficient cut or a cut to the program with the lowest benefit-cost ratio or the one furthest from meetings its objectives. It's always the cut which is most likely to draw public outrage or sympathy.

(Is there a name for this maneuver? I feel like there is but I can't for the life of me figure it out.)