The question is, where do you strike it? And when you're talking about alignment, How does that not become a strategic disadvantage and Achilles heel for the UK or any country which gets the balance wrong?

Well, if you got the balance wrong, it would be a strategic disadvantage.

That's the easy answer.

In the same way that the strange Lovian logic of the Cold War was that you, on the one hand, had to make these horrific, mind-bendingly awful calculations about how many casualties you could inflict, how many people might survive the nuclear holocaust, what was the relative advantage.

in the hope that your game theoretic framework was the same one being applied on the other side, and therefore they would conclude that it wasn't worth it.

So you were doing this cost-benefit calculation.

I think the argument that we're making in the Weirdlovian thing is similar.

How do you feel about human beings being behind the loop rather than in the loop? And is it worth distinguishing those two things?

And that's a, that's a US system, Project Maven?

Project Maven is a, is a US system that was born out of the Afghan and Iraq conflicts.

I remember in two thousand eleven, the US reported that they had thirty-seven years' worth of video footage recorded in a single year in the Afghan conflict, but that volume of data can only be realistically processed using artificial intelligence and algorithms.

And then there's a rigorous process of going from that to understanding, okay, well, if we strike it, is it gonna be compliant with international humanitarian laws? Then you have to figure out proportionality, like, okay, if I'm gonna strike this target, is it going to be proportionate relative to the military advantages that it gains, and is it necessary? And then eventually you've got to package that all up for a lawyer, a commander, panel of others around, and they are gonna make a decision to approve or not that target.

So AI is gonna be using to sort and make sense of information at, at pretty much every stage, uh, planning and th- and then post-strike in particular.

So, so, so I, I guess I'm, I'm asking how much extra human work that produces for someone like you to be sure that the algorithm has summarized that correctly.

Well, I, I would likely have a much clearer idea of the level of false positives and false negatives that the machine produces than I do my analyst because it's a bit like I, I won't use medicine where I would be out of my depth, but I will use taxis 'cause we all use those.

We hold Waymo to a much higher standard than we do Uber drivers, right? Like, the, the [chuckles] number of people that are killed by, uh, Uber drivers is almost certainly higher than Waymo.