The obvious solution to me is sponsorblock switching to sampling pixels out of each frame, like that project that encoded data into video streams (yet resilient to compression), there are algorithms that could fingerprint any ad with an extremely high degree of accuracy. It'd be more complex than the current implementation, but it'd also be more resilient. I'd settle for it hiding the video and suppressing the audio for the ads duration, possibly displaying a countdown timer, vs actually watching the ad. Then Youtube would get paid, but have no way of knowing you haven't seen the ad, and the metrics around their ad effectiveness would ultimately suffer, so users still win.

You could even go so far as to have the client cache the video, several minutes in advance, dropping all the ad frames, so it's a seamless experience for the user. I got money, but will spend 10x as much ensuring Google gets less from me. It ain't about money. It's about sending a message!