Vote Machine Learning For President!

You have to hand it to American democracy. After many years of political back-and-forth, we have finally reached a consensus: government blows. Government can’t do anything, least of all build an HTML website with a form on it to collect health information.

It’s popular these days to trash policy and policymakers, and certainly the politicians who watch over them. In contrast to the heap of ineptitude that is Washington, Silicon Valley is instead an oasis of un-bureaucracy, a paragon of potential. If only we could get more of our technologies into government to just get shit done, all – literally all! – of our problems would be carried away like our laundry with Washio.

That’s why I am voting Mel! for President.

Mel!’s birth name was Machine Learning, but that didn’t test well with political focus groups because Americans apparently don’t like machines or learning. Government of the machines, by the machines, for the machines isn’t exactly a soaring rallying cry, but everyone likes Mel!

Seriously folks, Mel! is fantastic. He takes random forest walks with his child processes. He even knows how to make decisions about trees since he pays attention during those walks. He hangs out with his nearest neighbors as an upstanding member of his community. He also ensures that the kernel density of his popcorn is just right at his family reunions.

Mel! is everything we want and need in a candidate.

Before I continue lauding Mel!, we probably should take time to talk about his platform (he is truly Silicon Valley’s candidate – a platform!). Basically, it can be described as a simple IFTTT algorithm of if X then ML where X is anything. If problem with website, then machine learning. If complicated policy, then machine learning. If Russia invades Y, then machine learning. Finally, a policy that can fit on a bumper sticker of your autonomous car.

Now, this may sound like magic, because well, it is. But don’t let magic frighten you. The idea of machine learning is actually pretty simple, which is why Mel! is going to do great with voters. Programmers can build statistical models (often using Bayesian statistics) that maps past data to predicted outcomes.

One popular example of these models is spam filters. Previous emails marked spam can be analyzed to create a statistical model of what spam looks like (basically campaign fundraising emails except for Mel!’s). When a new message arrives, we can compute it through the model to determine the likelihood it is spam, and if it is above a given threshold, we place it in a special junk mail folder.

Machine learning is extraordinarily flexible, which is why its potential is so exciting and its uses are so varied. At some point, it’s all just statistics, applied to any number of problems where repetition and past data are key elements.

Mel! isn’t coming out of nowhere. Government seems the perfect place for these sorts of tools, and indeed, we are already seeing numerous applications of machine learning today to public policy problems. Traffic mitigation is one area that has already seen several diverse implementations, from the license plate readers at the Golden Gate Bridge tollbooth to traffic mitigation algorithms that can change the intervals of red lights to best move traffic.

What Mel! aims to do is to use ML everywhere in government. Take immigration, an area of policy that has reached a sort of asymptote of absurdity these days. We could replace all of our messy quotas and waiting lists with a big black box of machine learning magic, and admit people instead based on some objective function such as potential for entrepreneurship. Mel! intends to symbolize this shift by replacing Lady Liberty’s torch with a golden MacBook as soon as shipping times improve to one week.

Mel! also wants to use ML with parole boards. Why are we releasing criminals when we could have a computer determine likelihood of recidivism. We already know that parole judges are completely arbitrary based on time since last food intake, so why not just replace them with a better decision-making algorithm? Mel! is tough on crime.

In fact, Mel! intends to go deeper. Admissions to college – why do we go through such a massive rigamarole every year stressing everyone out? Why don’t we just replace the whole system with an ML algorithm that will determine the fate of students in a neutral and fair way such as “meritocracy.”

(Um, Mel! this is getting a bit unsettling).

Like Inception, Mel! insists we must go deeper. In these times of austerity, government resources need to be placed with those most capable of taking advantage of them. Using ML algorithms, we could ensure that the “right” people get good healthcare, the best schools, and the easiest access to limited resources.

Mel!! (yelling at Mel! is hard since his name is yelled by definition).

Damn it, another candidate for president has dived head-first into the crazy side of the pool. And I liked Mel! He may have been naive sometimes, but he seemed decent when I first got to know him.

The challenge of course with machine learning is that machines aren’t substitutes for politics. Instead, they merely move the politics around. If we change our immigration system to use machine learning instead of quotas, that simplifies the system, but then the politics simply moves to the input data. What should we include as part of our model? Country of origin? Height? Birth order? Our choices will determine who gets in, and so the politics will simply rear its head in another place.

Machine learning has a place in policy – there are sites where optimization is important including transportation and other more operational parts of government. But from where I stand, I don’t want to ever give Mel! as much power as he so desires. Government may be messy, but it is our mess, and it needs to stay that way.