Can a Master Algorithm become the Next US President?

I just read “The Master Algorithm: How the Quest for the Ultimate Learning Machine Will Remake Our World” by Pedro Domingos. Fair warning: it’s not for the faint-hearted. Domingos is a leading researcher in the area of machine learning and he’s a zealot for his subject. It’s pretty technical so the book is kind of heavy going. On the other hand it’s probably the state of the art, at least for a layman like me.

But the book raises some interesting issues such as, when will we achieve the Singularity popularized by Ray Kurzweil – when will computers overtake humans in intelligence? What are the techniques we will use? How much are we already using them already in everyday life? With self-driving cars on the horizon it’s a timely question.

I’ve already tackled one aspect of this issue in a previous post “Can Siri Become a Great Leader? – The Coming Out of Synthetic Psychology”. My thesis was that there is no reason why you couldn’t create a synthetic leader based on the approach of reverse engineering leadership and personality assessments into a virtual agent.

But on reflection I wasn’t thinking of the more difficult questions of leadership, just the garden-variety leadership issues that all leaders in the business and the military face. I wasn’t really thinking of political leaders though. With the next election inexorably drawing nearer the natural question is, could we replace a president by an AI – artificial intelligence – aka a Master Algorithm?

Most people would regard this as being silly or even facile but here’s a question to some of them. Would many Democrats feel that GW Bush and previous conservative presidents did such a bad job that a machine could have done better? Probably I would guess. And if you asked the same question of Republicans about Obama, I would wager that most of them would say yes too. There you have it; it’s a question that’s at least worth considering.

Let’s start off with the problem. Naturally most of the problems that politicians have to address are intrinsically insoluble. They all know that and so do we so we both play a game in which we pretend they are soluble and then we all dance round the subject which is why our levels of political discourse are so egregiously infantile.

So it should be easy to replicate this low level of discourse. In principle all you have to do is to feed into the AI the main tropes used by each side and then the anti-tropes used to counter them. That’s certainly not difficult. So in general all you have to do to get an AI President is to match every situation to his side of the political spectrum and then pull up the appropriate trope. No problems there.

How about making the politician look human? Well there are lots of politicians that have problems with authenticity starting with Hilary Clinton and Jeb Bush so the bar isn’t too high. We have a history of competitions for the best AI program to win the Turing Test (in which a human judge must correctly judge which responses are from a computer or a human) and the state of the art is progressing steadily. Just layer on top of that a reverse-engineered narcissist – we are overwhelmed with that sort of knowledge – and you get a politician. Piece of cake.

But a politician has to meet and greet people and kiss babies. How good would an AI be at recognizing faces and responding to them? Uh oh, this is bad news for the humanists out there. It’s now widely accepted that computers outperform humans at facial recognition. So an automated politician can recognize more people than you or me, and it never forgets a name. Once it knows the name it also knows more than you or me and can utter the correct pleasing but anodyne responses. That would make them better on the hustings than many a real person. So score another one for the Master Algorithm here.

How about developing strategies to counter the sort of sticky situations that arise every day, like Syria, crossing gender lines, income inequality, government spending and personal email servers? Again all you have to do is pick the category, feed in the party line, add the trope and then the policy. Foreign, social, health, tax, redistribution etc. Policies and strategies can all be easily classified in this way.

And guess what? The resulting policies and strategies might even beat the current ones and probably couldn’t get worse anyway. Even if your AI’s policies don’t work, that’s actually just what happens in real life so your AI will be right on the money!

There’s a relatively trivial question of what machine learning techniques you use. Read Domingo’s book to crib up on that and hire some of the developers from IBM’s Watson and Deep Blue (winner of Jeopardy you might recall). Voila, you’ve actually got far more material than you need and certainly more knowledge than Donald Trump (OK I know that’s an unrealistically low bar)) or the other Presidential candidates.

According to the third of Arthur C. Clarke’s three laws, “Any sufficiently advanced technology is indistinguishable from magic.” Even 5 years ago, a self-driving car would have been firmly placed in the magic column. Three years ago, editing the human genome in real-time would have been viewed as playing God.

Now they are already here; we’re whizzing past them so quickly that they are already disappearing in the rear-vision mirror of our Teslas and the high res photos of GoPro-equipped drones.

Our politicians haven’t earned their keep. It looks like it’s a systemic issue. As they say, the definition of insanity is doing the same thing over and over again and expecting a different result, just like we are doing now.

It’s time for a radical change. Maybe a Master Algorithm is a good place to start over.

Can politicians be automated?