🤔 racist ai
- various AIs "learning racism" is expected, b/c racism exists, and it perpetuates via memetics/content
- (ethics are arbitrary, purely memetic)
- it's only "wrong" because "population controlling the AI" is different (and so, has different values) from "population used to train the AI" (which is slightly racist)
- likely big focus for future training
- ability to prune/curate training data to include/not-include certain values/ethics/morals/opinions
- a-la-carte ethics for your AI
- or instead of pruning the incoming data, adjusting the model post-hoc
edit: 2023-01-25- turns out, GPT-3 already has learned what "racism" is, in that it can be prompted to "not be racist"
- language models "learn" words, which are mimetic constructs
- values are also mimetic constructs (partially perpetuated through language)
- will likely see various cultural groups train their own models (and AGIs) to serve their values (and help those groups compete against other groups)