Csaba Szepesvari: waiting for full AI understanding is futile, deployment testing is the real brake

CsabaSzepesvari · x · 2026-09-21

RL researcher Csaba Szepesvári weighs in on whether AI development should slow until we understand the "physics" behind models:

Knowing with 100% certainty that a model works (= true understanding) is likely impossible, in AI or anywhere else, so waiting for that won't help.
We already have explanations at different abstraction levels; the real risk stems from unpredictability of unknown unknowns.
Proper lab testing and deployment procedures are themselves a meaningful way to slow down—more practical than halting research.

More from AGI Musings