AI Regulation: what will really happen?

 

In the previous post, I joked about concerned liberals writing anguished prose about vague AI-related dangers: robust regulation is certainly required!

Hmm.

'So what would you do?' is a fair question. I answer thus: meta-regulation.

Two questions: What problem are we addressing? Under what conditions do we regulate?

We are certainly at the start of a multi-year AI revolution, driven by deep learning systems using artificial neural networks. In truth, what we have today are the electronic analogue of a disembodied, dessicated human neocortex. GPT systems induce concepts at increasingly abstract levels and produce relevant-ish output.

This is a platform technology. Think early demonstration steam engines, or petrol engines or electric motors. There is a limited market for concepts-in/concepts-out: although for academic intellectuals it must seem as if these systems are undermining their whole way of life.

In economics, we learn that we regulate when we have a market failure. Externalities occur when providers don't properly internalise costs or benefits; monopolies undersupply; some human market relations (drugs? prostitution?) are held to violate social and/or biological norms. We use the power of the state to redress the balance through regulation. 

What are the market failures attributable to the new AI? Apart for mass-production of text/images (where such falsifications may already constitute possible crimes) we don't have too many examples as yet: just suspicions that with cognitive platforms problems could spring up anywhere.

This may remind you that with smart people who've gone to the bad, crimes can spring up anywhere. Law books consequently bulge.

We're not going to know for ages if 'AI' is the granularity where we need to legislate (I really doubt it). I do understand, however, that politicians need to be seen to be doing something given the general air of panic and hysteria. 

They should take their time and prevaricate. The threat of regulation should be used to prod over-mighty tech companies to consult their corporate consciences a little more. 

The answer is meta-regulation.