Friday Night in LA

Friday night, I was in Los Angeles for one day, at a conference. And the conference was over by lunchtime. So what to do with myself, in LA on a Friday night? A DJ night at a dance club? An underground art gallery opening? An improv standup show by Kevin Smith? A performance of “The Curious Incident of the Dog in the Night-Time” entirely in Spanish?

Oh, sweet, an astronomy lecture and stargazing at Caltech!

In the end I didn’t even go to the stargazing. After the lecture (about the ALMA 66-antenna array, sharper than Hubble and in mm-wave!), four of the grad students and postdocs set up shop at the front of the room and ran a “stump the experts” style Q+A.

It was pretty fun.

A few pictures here.


It was Saturday night. We could have gone to a movie, but hey, we live in Manhattan! Anyone can go to a movie! Let’s go to this thing.

FUSIONS dissects through music, clowning, and video, the impact of the technology on the nowadays’ society. The humorous and absurdist dysfunctional daily escape of the characters into the realm of reality TV shows, movies, the internet and social media, place their lives into a virtual reality, raising questions about the future of the human relationships.

I expected a combination of modern dance, atonal music, and some weird distorted video projections. Perfect for 10 PM on a Saturday night.

Instead, it was more like Sophocles. Characters with occupations but not names stood alone before the audience, delivering monologues on the nature of life as an actor, nerd, clown, or, um, woman, basically. The woman is distinguished as being “obsessed with reality TV”, which is sort of a bizarre idea at the end of 2016. What’s a TV?

The clown is an actual clown, who does a whole bunch of very impressive clowny tricks. Also everyone goes by their real names, or rather, the characters all take the actors’ names.

After the monologues, a very simple, surreal story develops involving a love triangle of sorts. The clown is the straight man, representing the audience in his desire to muck around and generally not have anything to do with these other characters. The others are pencil-thin single-issue archetypes, more concept than character.

The Actor spends most of the play telling the audience they should go see a movie instead of a play, and enumerating all the various ways that watching and acting in movies is superior to watching and acting in plays.

In this case I suppose he may have been right.

Mystic Pizza

So there was this movie called Mystic Pizza, in 1988. These days it’s mostly famous for having had Julia Roberts in it before she was famous. Also, it turns out it’s a real place, sort of. The screenwriter had pizza at the real Mystic Pizza and was inspired to write a story and set it there. Then they made a film set near the actual restaurant to shoot the movie, and then the restaurant completely renovated itself to look exactly like the film set, and doubled in size to handle all the extra traffic.

Anyway, movie or not, Mystic Pizza is reasonably located to be a lunch stop on the way from Boston to New York, so that’s what we did. Turns out, if you can manage to find parking, it’s a darn good pizza at a fairly reasonable price, plus a full menu of local Italian joint stuff. Even on a peak travel weekend the wait for a table was short, maybe 10 minutes, and service was reasonable.

They’re really obsessed with the movie. The place is covered in stills from the movie, posters from the movie, and monitors showing the movie on a loop. They also wear and sell t-shirts that were invented for the actors to wear as uniforms in the movie.

I burned my mouth, which I guess is an endorsement of a sort. So if you go, pace yourself.

The Clock

I watched The Clock at the Boston Museum of Fine Arts from about 4:45 to 6:30 on Friday. My thoughts on The Clock:

The experiment works in part because scenes with clocks in them are usually frenetic. In a movie, the presence of a clock usually means someone is in a rush, and so most of the sequences convey urgency.

It’s often hard to spot the clock in each scene. In less exciting sequences, this serves as a game to pass the time. In many cases the clock in question is never in focus, or is moving too fast for the viewer to notice. The editors must have done careful freeze-frames and zoomed in on wristwatches to work out the indicated time.

The selected films are mostly in English, with a fair number in French and very few in any other languages. This feels fairly arbitrary to me.

Scenes from multiple films are often mixed within each segment. It seems like the editors adopted a relaxed rule, maybe something like: “if a clock appeared in an original, then a one minute window around the moment of appearance is fair game to include during that minute of The Clock, spliced together with other clips in any order”.

The editing makes heavy use of L cuts and audio crossfades to make the fairly random assortment of sources feel more cohesive.

I swear I saw a young Michael Cain at least twice in two different roles.

Some of the sources were distinctly low-fidelity, often due to framerate matching issues. I think this might be the first production I’ve seen that would really have benefited from a full Variable Frame Rate render and display pipeline.

I started to wonder about connections to deep learning. Could we train an image captioning network to identify images of clocks and watches, then run it on a massive video corpus to generate The Clock automatically?

Or, could we construct a spatial analogue to The Clock’s time-for-time conceit? How about a service that notifies you of a film clip shot at your current location? With a large GPS-tagged corpus (or a location-finder neural network) it might be possible to do this with pretty broad coverage.

Fantastic Movies and How to Make Them

I’ve read the first three Harry Potter books so far, just enough to see that there are some weighty themes lurking beneath the surface. Then on Friday I went to see Fantastic Beasts, the new movie that kicks off another wave of the franchise.

Considering only the setup, Beasts could be taken as a similarly childlike tale. An accident-prone zoologist and his bumbling sidekick get themselves into a whole lot of trouble on a trip to New York City, and only manage to escape thanks to a little help from his lovable animal friends. There’s plenty of slapstick comedy, dressed up in the very best modern CGI, and even a perfectly chaste love interest. Kids movie.

Except … there’s more going on. Our British visitor encounters an American (magical) government that

  • Is structured as a gigantic secretive bureaucracy
  • Forbids miscegenation (with non-magical people)
  • Routinely implements the death penalty without effective due-process protection
  • Has been infiltrated by crypto-Nazis

It’s hard not to hear our British screenwriter’s voice in our British hero’s judgmental astonishment, and it’s hard not to see American society and government past and present being politely skewered by proxy.

I think this combination of light comedy and deeper commentary is a winning formula.

A/B governance: a manifesto

At their convention this summer, after a year of primary wrangling and many more years of labor activism, the Democratic party officially adopted a platform that calls for a Federal minimum wage of $15 per hour. That’s more than double the current Federal minimum wage, $7.25 per hour. They base their number on the principle that a single parent of two children working full time should not be below the federal poverty line.

In contrast, the Republicans have consistently opposed raising the minimum wage, implicitly supporting it at its current level, set by the George W. Bush administration. They often argue that raising the minimum wage will eliminate many jobs whose marginal profit to the employer is lower than the minimum.

In essence, the parties appear to agree that there should be a minimum wage, and agree about most aspects of labor law (e.g. 40 hour week, overtime pay at a 50% premium). They just disagree about the minimum wage number, by roughly a factor of two. Both sides make principled, philosophical, and emotional arguments in favor of their position. Neither proposes any data. What actually happens depends on which party gets 51% control of the legislatures and executive. In mathematics we would call this a step function: policy doesn’t change, and your vote doesn’t matter, until a crucial point, where one last vote crosses a threshold and produces a discontinuous change.

This is a silly way to run a democracy. It’s an even sillier way to run an economy. I think we can do better.

We have two big problems to solve: we make important decisions without data, and we make them discontinuously. This is like trying to drive while blindfolded, in a car with a lightswitch instead of an accelerator. Solving either problem independently is somewhat unrewarding; the system is still broken. Solving them together might be easier.

Let’s focus on the data problem first. How can we get real evidence of what the effect of a change to the minimum wage would be? Many economists have tried, for example, to estimate how many people would lose their jobs as a result of an increase to the minimum wage. The results are all over the map; they don’t even agree whether the effect exists.

The lack of agreement in economic studies is no surprise; they have almost nothing to go on. The best data available for this kind of measurement is in studies like Card and Krueger’s, which looked at 410 fast food restaurants near the New Jersey-Pennsylvania border, when New Jersey raised the minimum wage. That’s a worthwhile study, but it only applies to one industry, in one location, at one time, and only if you are willing to accept a sweeping set of assumptions that there’s nothing else really “different” between one side of the border and the other.

The statistical power of a study with, in effect, two data points, is not very high. Getting more statistical power is hard. In serious sciences, there’s only one universally trusted way to get clean data about the effect of some intervention: a randomized controlled trial. If economics has a bad reputation for reliability, it’s mostly because economists don’t do them. Economists can’t raise the minimum wage for half the businesses at random, and then observe everybody’s tax records to see how the two groups fared.

But the government can.

A randomized trial of this kind, known in silicon valley as an “A/B test”, seems a little unlikely when comparing the two parties’ minimum wage proposals. Surely business owners would not stand for their minimum wage varying by a factor of two, controlled by a coin toss. Luckily, there’s no need. Gigantic randomized trials of this kind would be capable of detecting extremely small effects. For example, there are 28 million small businesses in the US. A randomized trial across these businesses would be able to detect effects of the trial smaller than 1/5000th the natural variation between them. That’s less than 0.02%

Suppose you believe that doubling the minimum wage would result in small businesses falling ten percentiles in growth, relative to their current distribution. With an A/B test of 28 million businesses, you could detect that effect by raising the minimum wage 1 cent per hour for half of them, and lowering it by 1 cent for the other half.

I know people might find any randomization of the law distasteful, but we’re talking about 1 cent per hour. For tiny fluctuations like this, I think that it’s politically viable.

Of course, this is not the only possible randomization scheme. An economist might want to randomize at a coarser grain, like entire zipcodes, sacrificing statistical power in order to study larger scale dynamics. Conversely, we could imagine randomizing at a finer scale, maybe down to individual employees, maybe subdivided in time, to gain tremendous power to resolve microscopic effects.

Remarkably, you can run many such studies simultaneously without creating interference. Specifically, they are independent to first order, and the first-order approximation is extremely good when considering these kinds of small perturbations.

Out of all these countless possible studies, how do we decide which ones to do? The current political process will always choose None of the Above; whichever party is in power would rather pursue their agenda than collect data that might refute it. What would an experimentation-focused process look like?

In typical legislative systems, >=50% of legislators are empowered to take actions that alter the lives of 100% of the population, including the power to run a randomized trial with half the country in the control group (no change) and half in the treatment group (under change). What if X% of the legislature could pass randomized-trial laws whose treatment group is X% of the population, selected at random? Even individual legislators could run “small” trials, to test their own hypotheses about the consequences of government actions.

As much as I love randomized trials, the idea of having my life upended by random chance and one crazy legislator sounds awful, and insupportable. To avoid this situation, we can also limit the “strength” of trials. To define this, first consider that even the slimmest legislative majority can pass bills of “full strength”, whose magnitude is limited only by the constitution in effect. Proportionally, we could allow trials representing X% of the legislature (X < 50) to run trials at 2X% of full strength. How do we establish full strength? In principle, one can imagine a constitutional court, presented with a legislative template with blanks for various parameters, establishing the constitutional limits on those parameters. This certainly seems unlikely in the context of contemporary forms of governance, so let's looks for an easier alternative, at least as a first step. Suppose a Y% majority, Y > 50, passes some piece of new legislation that changes some value from A to B. Maybe it’s the number of weeks of unemployment insurance eligibility. The minority (X% = 100% – Y%) opposes, and the minority believes that a randomized trial is worthwhile. Both the “before” and “after” states are presumptively constitutional, so this sets a lower bound on “full strength”. Then with our proposed equation, the minority can run a trial that keeps X% of the population closer to the old plan. This treatment group, in the strongest allowed trial, would be eligible for B – (B – A) * 2X% weeks of unemployment, or B – (B – A) * X / 50.

There are two obvious extreme cases. One is the small minority, X -> 0. Naturally, in this case, the number of people affected goes to zero, as does the magnitude of the effect.

The other special case is X -> 50, approaching an equally split legislature. In this limit, the minority can run a trial affecting 50% of the population, and people assigned to the trial group will continue to experience value A, unchanged from before the bill passed. This means we have satisfied our continuity goal!

In case that isn’t clear, let me break it down. Suppose the two parties have two preferred states of affairs, and the legislature is always nearly evenly split. Under current systems, whichever party gets >50% of the legislature enacts their agenda, and we toggle back and forth between the two states whenever control switches between the parties. Under this system, the majority enacts its agenda, but the minority can roll it back for half the people, resulting in half the population experiencing version A, and half experiencing version B. This is true regardless of which party holds the majority!

I think my favorite thing about this arrangement is that, when public opinion is split on what policy is best, the resulting trial has the strongest treatment and the largest sample size, resulting in the greatest possible statistical power. That means that instead of an endless deadlock, we will quickly have the best possible data to tell us about the relative effects of each party’s policy.

Obviously not all policies are subject to this kind of trial. I’m not suggesting we randomize our next declaration of war, or constitutional amendment. Still, almost any domestic policy that includes a number can reasonably be treated this way, from tax rates to statutes of limitations.

There are many questions left to answer. How do we prevent legislators from “re-rolling” until a specific person ends up in the treatment group? How do we prevent legislators from creating a multiplicity of redundant trials that add up to excessive strength? How do we maintain budgetary balance, coupling taxation and spending while allowing this kind of randomization? How do we ensure that trials have sufficient statistical power? I don’t know the answers, but I have a feeling that with a bit of help from game theory, we could construct a set of bylaws that would achieve the goal while limiting abuses.

Note that I said “bylaws”. Everything here can be done by most any legislature on its own initiative, just by altering its own bylaws. A constitutional amendment might help, but it shouldn’t be necessary.

Of course, this raises the final question: would any legislative body ever actually do this? As I noted before: whoever has power now is usually happy with the structural status quo.

That might be broadly true, but in this case I think there’s reason for hope. This proposal does reduce the absolute power of the majority, but it does so by increasing the autonomous power of each individual legislator, including members of the majority. If the majority legislators are all truly of one mind, then perhaps there is no advantage, but that’s never true. Fissures and factions are always visible within the majority. A randomized-trial bylaw would empower those factions to show off their own proposals, and arm them with the data to convince everyone else.

In a sense, I think this proposal is plausible because it appeals to arrogance. Everyone believes that their policy proposals are the right ones, and their opponents are wrong. That means both sides ought to favor a trial, for it will surely prove them right.


There are so many things to say, but first we have to talk about the election. Ugh.

At their best, Trump and Pence look a lot like Bush and Cheney to me. Both pairs were

  • A governor and an utterly corrupt and self-interested businessman
  • A president with no interest in actually governing, and a Vice President who loves power
  • A president who inherited his position from his father, and a self-made Vice President
  • Obsessed with preventing women from aborting pregnancies
  • Ostensibly isolationist but also ostentatiously militaristic

Just like in 2000, the Democratic candidate won the popular vote but lost the Electoral College, leaving both the Presidency and Congress in the hands of the Republicans. (Then in 2002 the Republicans took the Senate as well.)

It’s possible that Trump is what he appears to be. If so, it’s the end of the Republic. He has, for example, promised changes to libel laws that would terminate freedom of the press, promised immigration policies that would end separation of church and state, and exhibited a penchant for personal retribution that would amount to the end of rule of law.

But let’s suppose instead that Trump ends up being an absentee Republican airhead like Bush, except disturbed instead of “compassionate”. In that case, I expect a rerun of Bush’s first term, 2001-2005. So let’s review. Purely from memory, I recall

  • Terrorist attacks. Yes, I blame the Bush administration at least partly for 9/11 because their isolationist ignorance of international relations and their unseriousness about governing resulted in overlooking a direct warning in an intelligence briefing
  • A useless war that claimed the lives of thousands of Americans and many tens of thousands of Iraqis, cost trillions of dollars, destabilized the entire Middle East, and led to the rise of ISIS
  • A stock market crash and economic recession
  • Tax cuts for the wealthy (without a corresponding spending reduction) resulting in the greatest federal budget deficits of my lifetime, and debt that I will be paying off for the rest of my life
  • Restrictions on embryonic stem cell research that set back biomedical sciences, and especially therapies for diseases like Parkinson’s, by years
  • Abandonment of the Kyoto Protocol, resulting in acceleration of global warming
  • Unchecked growth in the cost of healthcare and higher education
  • No Child Left Behind, which reduced funding for the most under-resourced schools
  • Failure to regulate financial institutions, which would produce the crash of 2008

Just checked … oh yeah, there’s also the USAPATRIOT act, which dramatically weakened our major constitutional protections against a tyrannical government, suspension of habeas corpus, illegal torture at US prisons and at secret “black sites”… I’m sure there’s more that I’m forgetting.

In short, 2001-2005 was one of the worst periods in American history. I’m not much for predicting the future, but I’ve lived long enough now that I know what to expect.

News and View

The Clinton campaign quietly announced today that their election night event will be held at the Javits Center, New York’s largest convention center.

I can see the Javits Center out my window. I am literally looking right at it right now. I can even see the TV screens inside, through the glass walls, which means I’ll probably be able to see the speeches on those screens without leaving my apartment.

New York City!

Like information, but less informative