History of the Capital AI & Market Failures in the Attention Economy

The article that this post is about came to me as a Twitter recommendation from both Tim O'Reilly and Charles Stross. That was more than enough to pique my interest. O'Reilly said: "This is far and away the best summary of the argument for capitalism as an #AI with a flawed objective function  that I’ve ever read. I love it."

He's right. 

History of the Capital AI & Market Failures in the Attention Economy is one of those articles that will likely be read and discussed for at least the rest of this year. It's long, opinionated, and well researched.

This is a lengthy discussion, so I’ll begin with a summary.

The basic idea in this post is to consider capitalism as a highly efficient objective function (or “AI”) with its parameters optimized for the satisfaction of our short term desires rather than our long term interests.

Paranoia about runaway feedback loops – in consumer capitalism, artificial intelligence, mass media, ‘Wrestlemania politics,’ etc – ultimately stems from the inscrutability of the emergent behavior of these complex systems to the individual actors and observers operating within them.

Rather than responding with Luddite / anarchist nihilism, we should remember that technological and social systems like these have dramatically reduced our exposure to the unpredictability of the natural world and greatly improved living conditions on a number of dimensions over the past few centuries.

At the same time, we should not ignore warning signs of a dystopian future, nor should we hope that a ‘personnel change’ of institutional leaders will solve our problems.

It's a long article. If you don't have time to read the whole thing, skip directly the author's Practical Recommendations. This is one of them:

(5) Set boundary parameters around economic growth and disparity. Instead of incessantly bickering about the dystopias that result from either raw economic growth without redistribution (which inevitably leads to extreme disparity) or the state enforcement of absolute equality (which leads to either stagnation or totalitarianism), we should define policies that moderate between these extremes and land us in explicit, acceptable ranges. We could, for example, set parameters such that whenever the economy is growing as a whole, we have a redistribution slider which ensures inequality decreases at the same time (see the 𝜏 parameter in this post).

But it's better to take the time and read the whole thing. It's worth it.