This is a great, concise (5 min!) video by @emilymbender
First off, she proposes to replace the term AI with "automation" because this allows raising pertinent questions much more clearly:
- What is being automated?
- By whom and why?
- Who benefits?
- Who is being harmed?
- How well does it work?
- Who is accountable?
- What laws/regulations apply to what is being automated?
(the last point refers to what @tante & @CyberneticForests recently pointed out; don't regulate tech, regulate behavior)
Then she proposes to distinguish different types of automation:
- consequential decisions (e.g. in the courts, banks, or HR departments)
- classification (face-recognition, slotting people into groups)
- recommender systems (automated newsfeeds)
- access to human labor (uber, mechanical turk etc)
- translation of information from one format to another (text-to-image, language translation)
- content generation (chatGPT)
They have in common that they are built on training data and the ability to reproduce certain patterns in that data (including all the biases in that data).
https://www.youtube.com/watch?v=eK0md9tQ1KY