A system doesn’t need to be wrong to be dangerous. If a decision is unappealable, unauditable, and irreversible, alignment stops being meaningful - regardless of accuracy or intent. Which decisions are we currently automating that must retain a human veto? #AI #AIEthics

The real risk in advanced automation isn’t that machines become smarter than humans. It’s that decisions get made faster than humans can question, appeal, or reverse them - until responsibility still exists in name, but no longer in practice. I tried to write this concern down

Which types of real-world decisions should NEVER become effectively unappealable due to AI automation, no matter the efficiency gains? Feel free to reply with others - or argue none should be off-limits. #AIethics #AIGovernance

This is what it feels like now

David We have entered the Singularity

Editing real photos of real people without consent is a different harm category than generating synthetic images. Safety is about preventing that outcome, not arguing who else might theoretically do something similar.

The BBC reporting makes something clear. If reporting the harm spreads the harm, this is a design failure, not a moderation problem. My note on why certain AI capabilities should not exist at all. substack.com/@kornucopian/n…