The extrinsic problems of AI

Two of the biggest downsides of AI are these:

First, given how energy is undercosted and has catastrophic climate externalities, any new invention that uses energy is problematic. (And the dataset gathering part not only leaks poison to the climate, it also offloads costs to innocent server admins getting scraped. No matter how "undercosted" energy is, that all adds up when you're being robbed. I personally haven't had any problems with being scraped, but I've been hearing how these issues have been more and more overwhelming and disruptive and taxing on my fellow admins.)

Second, given how we have an increasing wealth gap driven by ownership of means of production, any new invention that’s a powerful means of production risks further centralizing and concentrating that ownership and further driving the growing wealth gaps.

We need to patch these two bugs in market capitalism (or replace capitalism entirely) before we use AI.

But we need to fix those two issues even if we never use AI. The problems not only are wholly extrinsic to AI, they predated AI to an already urgent level.

Now, I’m saying extrinsic rather than fully decoupled. Given the runaway and self-amplifying looping qualities of these bugs in capitalism, and in AI itself, we risk disastrous singularities and tipping points and haywire-going.

Machine Learning—good and bad arguments against

Energy is undercosted

Externalities

Exploitation

Noise positivity

Please stop externalizing your costs directly into my face

The quaint wealth gaps of history


Source