Hello, SFIA audio listeners.

In this month's Nebula exclusive, big alien theory,

we're asking the reason alien civilizations might be rare is because most aliens are huge.

To hear it and every episode early and ad-free, plus hours of bonus content,

check out go.nebula.tv slash Isaac Arthur and use my code, Isaac Arthur.

From robot genies twisting wishes to paperclip maximizers eating the world,

the real danger of AI isn't just a machine rebellion, its obedience without understanding.

Artificial intelligence is a curious paradox.

Is it a technology built to follow instructions, to do what we tell it to do, and yet

is surrounded by a cloud of worry that it might not do what we really mean.

The fear is not just that it will ignore us, break its chains, but that it will obey us

too literally, and in that gap between what we want and what we ask

lies the problem of AI alignment.

We stand at the threshold of a world increasingly shaped by machine intelligence,

from recommendation algorithms shaping what news we see, to large language models writing

essays, to our time as vehicles navigating our roads, the

reacher AI grows every day.

The question, can we make AI safe, is not an abstract one for future civilizations,

it's a pressing question for our own era.

So today, we'll take a close look at what researchers mean by alignment,

why it matters so much, and how we might approach the problem. Along the way, we'll detour into

as most famous three laws of robotics, thought experiments like paperclip maximizers,

...