Karen Hao

And part of the reason was there were different ideas around how to recreate human intelligence because there were different ideas of what human intelligence is.

One branch was humans are smart because we know things, so let's create machines where we symbolically encode knowledge

into the machine.

And that's how we're going to replicate human intelligence

And the other branch was humans are smart because we can learn things.

And so let's create machines that learn.

And this is the concept of machine learning.

And in the initial days or decades of AI research, it was really focused on this symbolic branch.

And that really wasn't panning out.

And that's why the funding dried up.

And pretty much every modern AI system that we see today, including ChatGPT and all other large language models, now derive from the second branch.

the machine learning branch.

Exactly.

And the idea was, well, we learn from experiences, but machines learn from data.

So let's create computational systems that calculate patterns within data and derive some kind of understanding of

quote unquote, about the rules of the world through the data that we feed these machines.

And the reason why it took off is twofold.

One is that

It's an effective, you know, when you throw a bunch of data at a computer, it does find patterns.

And just that act in and of itself is highly useful.