New technologies and the sorting process of the market

Andrew Orlowski points out that Artificial Intelligence is not, in fact, all that good.

For years policymakers have believed that machine learning – the foundation of the current wave of AI hype – is fundamental to advances in robotics that will herald a “Fourth Industrial Revolution” – but the two don’t really go well together at all, or at least not using currently fashionable methods.

“Data-hungry, idealised algorithms simply fail,” leading researcher Filip Piękniewski told me. “The reality is we don’t know how to build control systems for machines that would be anywhere as robust as our own brains but we just love to fool ourselves into thinking we can do that,” The veteran robotics expert and former head of MIT’s AI lab Rodney Brooks was even more scathing about OpenAI’s defence that the “market wasn’t ready” for such wonders as the Cube-dropping Cube solver. “Perhaps the ‘problem’ is that the market is mature and understands what is valuable and not.”

The logical suggestion then follows, perhaps politicians should stop investing in or directing AI in the search for a Hail Mary Pass around current problems.

This does not mean we should all be ignoring AI of course. It just means that we’ve already got a method of sorting through the varied claims - the market.

A new technology, whatever it is, will work on some human problems and not on others. The Grand Task is to sort through those problems we wish to solve, test whether this new tech aids or not, then kill off the failures and do more of the successes. Which is exactly what free market capitalism does - it is, rather, the saving grace of the system. It also contains its own feedback loops - profit is the signal that a problem is being solved, as people wish to have it solved, losses are the signifier of a dead end.

Thus it is exactly when a new technology appears over the horizon that we must use that free market capitalism to test it. As, for example, the financial markets have been. High Frequency Trading, or perhaps arbitrage using algorithms to the extent the two differ, is exactly that AI. Crunch the numbers, read the patterns and without knowing anything about why they are trading upon the back of the pattern recognition. This works, even to the point that the original excess profits have now been competed down to perhaps below the cost of capital. All we’re left with is more efficient capital markets which are cheaper for the retail investor to use.

Shame, eh?

Or, as we’ve pointed out before, it’s exactly when we don’t know that we need to use markets to find out.