2 Comments

Letting “divorce-your-wife” Bing, “humans please just die” Gemini & co run nuclear power plants… hmm, what could possibly go wrong? And I thought our worst case scenario was that one of these AI systems could hack into a nuclear power plant. Def better to give them full access right away!

Expand full comment

Thanks for the comment. I agree concern is absolutely warrented and there are risks inherent to over using automation. Especially in nuclear where accidents cannot be tolerated. Humans should remain in the loop. That said, the use cases I'm discussing here, the AI's role is to spot problems for humans to resolve not to operate. Errors can still happen, but unlikely to be catastrophic. This leans into the limitations of today's AI capabilities. That said, its important to recognize we are moving into a world where increasing complexity paired with labor constraints and the constraints of the human mind might require us to rely on AI to carry some of the safety burden. Cybersecurity is the best example: the volume of network traffic and interconnected systems is too high for humans to manage, and network irregularities are often too subtle for humans to notice. Today, widely used anomoly detection sysetms designed to monitor network traffic, report irregularities that could be cyber attacks, and sometimes perceptively act to block traffic rely on AI techniques. In all sectors, nuclear likely included, anomoly detection is in use because humans simply cannot manage the complexity or scale of the task.

Expand full comment