AI agreement is not a safeguard.

23 Oct 2026 · 1 min

Watch on YouTube

0:58

In this short monologue, the speaker illustrates the danger of deploying probabilistic AI systems without clear consequences, using a meme about a robot offering poisonous mushroom recipes and an Amazon incident where an AI deleted production servers. The core argument is that AI should be employed to help design and build robust, low‑risk processes, but once those processes are in place, AI should not remain a critical component of the workflow. AI may be acceptable for peripheral, non‑essential tasks such as generating commentary, but it should never be the key decision‑making element in high‑stakes operations. The speaker advocates for a governance model where AI is a tool for process creation, not a permanent operational safeguard.

Chapters

I'm not sure if you've seen that meme. There's a gravestone and there's a robot standing next to the gravestone. It says, Dave died from mushroom poisoning on this date. And then the robot's got a little speech bubble next to it that said, oh no, you're right, Dave, that mushroom is poisonous. If you'd like, I can give you recipes without poisonous mushrooms because there's no risk. And there's stories of Amazon a few months ago, their AI just deleted some production servers because there's no consequences with it. So when you've got something that's probable and then you've got no consequences, it becomes quite a risky scenario. My view is we should be using AI now to help us build the processes mean we don't need to use AI afterwards. So once you've built a robust process, do we then need to use AI after that? So we use it to build the process, but there's no AI in the process at the end. You might say, but we can use AI to write commentary and other things. Yeah, we can. And in those scenarios, we can definitely think about using AI, but it's not the key part of your process. You could write all of that commentary on your own without any AI. The AI is there to just help you with that.