I feel like the delta between human cryptanalysis and AI cryptanalysis will be greater for symmetric cryptography, no? Like an attack that actually breaks ECDLP would probably be comprehensible to humans but an attack that actually breaks AES probably wouldn't be.
