Grant Harvey

Yeah, I can imagine there's going to be all sorts of complexities around unexpected events because you're always training for educators, right?

You're always training.

If the car's driving from A in a straight line to B and there's no traffic, it's easy.

It's going to be great.

But it's when someone does something unexpected or there's only a person or whatever, then you've got to build the data set in order for the model to behave in the way that you want it to for performance reasons, for legal reasons, for whatever.

And again, that's a challenge, I think.

Model usage in heavily regulated industries is going to be a big challenge because there's going to be concepts of liability.

And that's why, again, human oversight.

It might be inefficient, but it'll be the kind of thing where there's legal robustness to it so people are comfortable with it.

Otherwise, you're like, oh, yeah, I made this medical decision based on what another harmonistic model told me.

And who's liable if something goes wrong?

Yeah.

That's a huge unsolved problem, I feel like.

Who's the fall operator?

Is it the person who, who, who designed it?

Like it's people like, yeah, people want to see someone.

Is it me taking a nap in the back seat?

You know, and then if that, if that, yeah, if that's the, like, just imagine as soon as that's the case,

then suddenly self-driving cars, they're not dead because they're not dead.

It's not a sort of binary thing, but suddenly they can't self-drive.