Friday, November 15, 2019
More point-missing on AI
As usual, Brendan Dixon at MindMatters makes some good evocative points but misses part of the big picture.
When I read a piece created by another human, I am engaging with another mind. When I read a piece “authored” by an AI, however, I’m engaging an algorithm. The human may ponder, evaluate, weigh, and rewrite. (I sure do.) An AI spews: Given this input, under these conditions, this is the output. Period.In a genuine publishing setup, the writer and editor are working together. A human writer can certainly spew an algorithmic output. Most fiction writers do. A human editor can also be algorithmic, catching fake grammarhoid "rules" like "split infinitives" and "lie/lay confusion". When both writer and editor are fully human, the result transcends algo. I'm thinking of my current final edit on courseware. The publisher has a new copy-editor who is inspiring. She catches real failures of clarity, not grammarhoid "errors", and best of all she knows when a paragraph or image or animation could use improvement. She caught an image that had been subtly wrong for many years, and she inspired me to redo an animation that had been unclear for many years. There's no way an editing algo could feel that the animation is unclear, and there's no way a writing algo could make the animation clearer. Dixon again:
AI can be useful in writing. Consider an AI working from a mass of field reports and “writing” an overview of very complex events. Or a health writer who must summarize a huge batch of academic papers to determine what coherent message can be gleaned from them about a controversial issue in cancer treatment or vaccination. These uses of machine analysis benefit us by augmenting what we can do — like every other tool we use.An algo could probably do a better job of gathering than modern "journalists", but only because modern "journalists" can't even read their own words, let alone process external facts. The algo would be unable to detect subtle biases unless it was told in advance to trust CNN and mistrust RT, which is what the wetware "journalist" already does. The algo's analysis of science papers on an important topic would be equally useless for the same reason. 100% agreement with current fashion as of this picosecond = valid. Less than 100% agreement = invalid.
Labels: AI point-missing, From rights to duties
¶ 5:51 AM
