Ask HN: Do sociotechnical pressures select for beneficial or harmful AI systems?

The full question I'm wondering about is as follows:

Do sociotechnical selection pressures reliably favor ML systems that (a) increase their own future deployment probability and (b) reshape institutions/data pipelines to entrench that probability, even without explicit 'survive' objectives?

I've gathered some links exploring this and tangential ideas here: https://studium.dev/drafts/f1 - I'd love to find more reading material