What's this about?
"Humans are animals that are very hard to persuade. You can only try things yourself, see the results, and recognise your past mistakes. Perhaps this is because the brain is an advanced model that cannot be fine-tuned, only reinforced."
Is the analogy any good?
Better than most. Supervised fine-tuning shows a model correct examples and trains it to reproduce them — which is what persuasion attempts to do, and it works about as poorly on people as being told the right answer works for genuinely learning anything. Reinforcement learning requires the system to act, receive a consequence, and update from that. The analogy breaks down in the obvious place (people do learn from testimony, or civilisation would not work) but it names something real about which beliefs are actually revisable.
Why is it worth noticing?
- It is a rare case of AI vocabulary genuinely clarifying something about people rather than the reverse.
- It also has a practical reading: if you want someone to change their mind, arrange for them to experience the consequence.
- Small engagement, included because the observation is better than its reach.