Checks & Balances

So Sam Altman, the CEO of OpenAI is "...completely unbothered" that he might be less intelligent than his AI. As Guy Keleny puts it most succinctly in his letter to today's FT: 'AI is either stupider than people or smarter. If stupider it is useless, if smarter terribly dangerous. Where's the upside?'. I tend to agree, particularly with regard to the race to develop a General AI model, which is fraught with all manner of potential demons, practical, moral and political. But as to very vertical, use-case-specific models? I feel that much useful work can be done here [blog posts passim].

The main issue surely has to be the sheer expenditure of electrical energy that AI, along with crypto and the cloud, consumes? There has to be an upper limit on it, for practical as well as environmental reasons, one would reasonably think. The law of diminishing returns has to influence things in the long run, with long-term profits eventually being outstripped by the running costs of these behemoths. And as the sole motivation behind these developments is pecuniary rather than altruistic, once profits dwindle, so does motivation, unless one or both of two things happen:

Either the programming fraternity gets better at writing efficient, and more importantly, effective code and/or quantum computing passes from chimera to actual; or those limits will inevitably be met. Whether the decision to bale out as an industry comes before its environmental effects are considered important enough to consider is moot, but there you go, that's capitalism for you. As I've said, there are many less-generalised, focussed uses of AI already in effective use, and it would seem more prudent to concentrate on the achievably useful than chasing the impossible dream of replacing human intelligence [and to what possible end, other than venality?] with machines...