I wouldn't be too confident in that. Actuaries are not able to implement the models they'd like to. Regulators don't like complex models and they don't like models with disparate impact. Or models that make some risks (accurately) expensive.

I'm a developer now but by last few years in actuary work were mostly spent helping insurance companies deal with regulation. We did some modeling with with machine learning but ultimately the models that got approved had to be simple (compared to machine learning). My job before that was a lot about internal negotiations with other departments.

Better models were not the main bottleneck.

Absolutely, great points.

I think there could be scope to use ML on the pricing side however, while leaving the reserving models explainable to regulators.

You'll be hard-pressed to get business people to use AI for actuarial work.

If the intent is to codify rules in an AI that performs actuarial work, then it is far less error-prone to simply write a program that uses the rules directly. AI is not the proper solution, if I understand from your post what you are trying to solve.

May I ask who will lose his professional credentials, when your AI does not follow established practises and insurance coverage goes belly-up?

I am reminded of https://xkcd.com/1831/