David Autor

Because it can learn inductively from unstructured information and make inferences and recognize patterns and see things that we don't necessarily see, it allows it to function in settings where we don't really have good tools.

So much of the work that could in theory be automated because it follows simple rules and procedures has been automated.

So most of the work that we do doesn't look like that anymore.

Most of our work is actually not simple execution of repetitive actions.

That certainly was the case 100 years ago.

That was the case in early factory work.

That was the case in a lot of office work.

Most of the work that people do now requires decision making.

And we haven't had tools that are good at working in the kind of messy environment of weighing competing objectives.

And what AI is potentially useful for is supporting that type of work.

In many, many cases, AI is better as a collaborator.

And so increasingly, the work that we do where decision making is important.

The stakes are very high in medicine, in law, in construction, in child care, in skilled repair.

So having tools that help us do that well would be great.

Radiology is a good example.

It's now a very widely discussed one.

Jeffrey Hinton, who's the inventor of neural nets, one of the inventors and won the Nobel Prize for that, said about 10 years ago, oh, within five years, it's perfectly obvious we won't need radiologists anymore.

Machines will just be better at this than people.

And there is now a ton of AI in radiology.

And it's a very good tool.