"Affirmative Action" at Ivy Leagues - what is the value?

John Stossel is not the first, and will not be the last, to call out the inequities of letting different races into competitive colleges. "Diversity" is a positive feature, because it is not academic achievement alone that determines one's position later in life. Evolutionary biology teaches us that monocultures are extremely vulnerable to perturbance, and an all-asian-and-white, legacy-children-from-Exeter-and-Andover student body is a sitting duck for any major cultural or sociatal shift. That is why it makes sense to admit athletes with lower test scores and grades - athletes tend to make significantly more money later in life (and then gift it their alma maters) than non-athletes with comparable admissions scores.

The question I would ask is not what the cause of these policies is, but what is the effect? We have a generation and then some of diversity admissions programs. What has the effect been? Have the diversified schools benefited in any objective measure, and which measures? Do African American Ivy graduates go on to make more money, or start more non-profits, or file more patents, or win more elections than white ivy graduates? Do hispanics Ivy graduates acheive executive positions, or sell more albums, or publish more journal articles, or defend more cases, than their asian brethren?

I never understood the use of race as a stand-in for diversity. I understand cultural diversity - Economics, religion, ethnicity, country of origin, whether you were raised by one or two parents - these all are huge contributors to value systems and learned experience that can make a community of any kind more resilient. A school enriches itself by bringing in new perspectives; but it enriches itself more by bringing in the white South African ghetto superstar than well-educated, upper middle class black South African. Without evidence to the contrary, I believe using race as a stand-in for diversity is lazy, divisive, and counter-productive.

Good news... but you can't put it in your pipe and smoke it

The White House released a memo to federal prosecutors and top DEA executives today that told them not to prosecute medical marijuana users so long as they were in full accordance with conflicting state laws. A breath of fresh air for marijuana activists, and one of the things I thought Obama could realistically do to make his presidency suck considerably less.

But, similar to his empty words on every other promise he has chosen not to break outright, this is pretty empty. As a first policy initiative, it's fine, but if it is not followed by real change it is a limp carrot next to the deadly stick of the Drug War. Things it doesn't do:

1. Prohibit federal prosecutors from prosecuting cases... it just makes it a low priority
2. Prohibit the DEA from investigating and permanently seizing property of suspected drug users without actually charging them with a crime
3. Make the people who use or sell medical marijuana lawfully any less vulnerable to the next rex publius
4. Similarly, it does not make the actions of users and sellers legal - things they do now even if prosecutions cease for a few years can still be used against them in court down the road

The ruling seems nice enough, but it essentially estabilishes precedent that the Drug War is completely at the favor of the whims of the Executive. If it goes to the Supreme Court, it will be struck down as both a legal defense and a policy measure, because it essentially tells the DOJ and DEA to ignore Congress. In other words, it is typical of Obama: it is a nice propaganda measure with no teeth, the unintended consequences will probably end up being worse for those who support it than the status quo, and he will probably stand pat and consider his job done now that he's given a flowery statement.

It need not be terribly damaging politically either - Obama could simply introduce a bill with edits to existing drug bills that essentially suborns federal regulation to state regulation in drug cases when the two are in conflict. It could be pushed as a state's rights issue and would likely have a sizable number of Republican co-sponsors. This is pretty weak, and dangerous, if the administration does not follow it up with an actualy policy or legislative push.

I'm sick of being right all the time... no, seriously.

The WSJ has an article today about the origins of the housing bubble. The main point: so-called "predatory" lenders needed to sell their mortgages to someone. They did not hold them like normal banks, but instead securitized bundles of them and sold them to larger institutions.

"Mortgage brokers had to be able to sell their mortgages to someone. They could only produce what those above them in the distribution chain wanted to buy. In other words, they could only respond to demand, not create it themselves. Who wanted these dicey loans? The data shows that the principal buyers were insured banks, government sponsored enterprises (GSEs) such as Fannie Mae and Freddie Mac, and the FHA—all government agencies or private companies forced to comply with government mandates about mortgage lending. When Fannie and Freddie were finally taken over by the government in 2008, more than 10 million subprime and other weak loans were either on their books or were in mortgage-backed securities they had guaranteed. An additional 4.5 million were guaranteed by the FHA and sold through Ginnie Mae before 2008, and a further 2.5 million loans were made under the rubric of the Community Reinvestment Act (CRA), which required insured banks to provide mortgage credit to home buyers who were at or below 80% of median income. Thus, almost two-thirds of all the bad mortgages in our financial system, many of which are now defaulting at unprecedented rates, were bought by government agencies or required by government regulations."

This requires a little mental gymnastics to understand correctly, because it is not intuitive. Normally, we think of demand coming from the consumer, but for firms that only lend it is the other way around. Their operating cash comes from businesses UP the chain, not loan interest. Therefore, they are really selling in reverse, to the banks, and their "product" is the value added by pulling a group of mortgage borrowers into a securitizable group.

The point: if wonky demand for bad mortgages was not created by the federal government, these companies would not have been able to sell their product. They would have either had to stick to traditional, safe mortgages, or else they would not have existed in the first place. Companies like Countrywide were probably originated because banks wanted to subcontract a layer or two of risk from the bad mortgages the federal government was forcing them to take. A kind of, "we know when the Countrywides fail that we have a sales cycle or two to purge our own toxic assets and lobby our congresspeople for a bailout." It worked for most of the big banks, and who can blame them for doing it? They were damned if they did, damned if they didn't.

A possible general algorithm for quantum computers?

Exciting news comes from MIT via Next Big Future: a general algorithm for quantum computers which solves linear systems of equations. This is big news, since recent years have shown significant progress in overcoming technical challenges of quantum computing, but the lack of a general linear algorithm that was demonstrably faster than digital computers threatened to confine QC to a small set of somewhat esoteric problems like code-breaking and database search. Digital algorithms for linear systems take on the order of N steps to solve N equations, where this algorithm takes on the order of log(N). Not a big difference for small matrices, but enormously better as the matrices grow in size. A general linear algorithm that is faster than digital algorithms for the same thing is promising, because linear systems underlay virtually everything modern computers do well.

MIT researchers present a new algorithm that could bring the same type of efficiency to systems of linear equations — whose solution is crucial to image processing, video processing, signal processing, robot control, weather modeling, genetic analysis and population analysis, to name just a few applications.

A gotcha, of course, does exist. Namely, once the result is determined it is in a quantum superposition, so it takes order-N steps to read it out. Two possible solutions present themselves to my uneducated mind: the read-out step may be something that can be done massively in parallel; i.e. by entangling the "result" qubits and sending them to N readers which can be purpose-built cheaply to just be able to query a set number of qubits in a set way. The second way is working with the problem rather than around it. Many or most matrices a digital computer sees are either highly sparse (consisting overwhelmingly of zeros), and/or most of the answers are trivial. This can be exploited using digital computers in parallel. For instance, a finite element analysis might be run on a digital computer with a few million nodes, and simultaneously on a quantum computer with trillions of nodes. Then the digital computer could direct the quantum computer to only query the nodes with high stress concentrations to bring them to the higher quantum resolution. Similarly, systems can be set up so that the pertinent variables are in a known location. A market forecast, for instance, might take into account trillions of variables to forecast a stock price in the last row of the results vector.