All the talk about inscrutability of convolutional neural nets and AI that can explain itself has me wondering: what are non-trivial (and used in real systems) examples of the opposite kind of algorithm: highly scrutable and self-explanatory?
For example greedy graph search.
Conversation
Replying to
People who talk about the inscrutability of neural networks are often complaining about the lack of mathematical proofs for optimality, convergence etc.
If you bring them, it doesn't mean they will suddenly become scrutable :) any more than to 0.00000001% of mathematicians..
3
Replying to
No, that’s not what regulators and politicians are talking about. They want human justifiability of algorithmic decisions, not convergence proofs or guarantees. That’s a sideshow of minority math-supremacists within tech that is largely unimportant right now
1
2
Show replies
Replying to
generalized linear models are not self-explanatory but are highly scrutable






