UVA Today

In a recent entry to Darden’s “Ideas to Action” blog, Korinek said he is “taking time to reflect on life after cognitive automation has made me redundant as an economist.”

The bold statement reflects the seriousness with which Korinek takes AI development. UVA Today caught up with Korinek to learn more.

Q. How certain are you that advancements to AI could lead to the elimination of your job? When could that happen?

A. Do I view it as an absolutely certainty? Probably not. But honestly, I do view it as a distinct possibility that this may happen, even within the current decade. I would not have said that if we had spoken a year ago, but the progress over the past 12 months has just been so rapid.

Q. What has caused this rapid progression?

A. A few years ago, everybody was convinced that physical jobs would be automated first and cognitive jobs would be much, much slower to be automated. But in some ways, the recent developments with these language models – like ChatGPT – have really turned that take on its head. And it seems now that cognitive automation is proceeding much faster than physical automation.

I would say that there aren’t necessarily any cognitive tasks that we can perform that couldn’t also be performed by sufficiently advanced computers. The question is ... would you be willing to have, for example, an AI lawmaker, an AI judge?

Q. Do you think someday it could get to that point? What are the limits to this when it comes to computers replacing specific jobs?

A. Let me distinguish between whether AI systems can display the capability and whether it’s desirable for us to actually let AI perform those tasks. I do believe they can acquire the capabilities to, for example, evaluate legal texts, to negotiate law, to evaluate complex legal situations like what judges do. I don’t think our society, at least in the near term, is going to be willing to completely hand those tasks over to an AI system. But I believe what we will see is AI systems will provide more and more assistance in those tasks.

And if you have an assistant that becomes really good at what it’s doing, you start to trust them. When you have automated systems that, for example, are really good at writing laws, there’s going to be a tendency to take more and more of what they’re giving us.

Q. How can AI impact journalism?

A. Right now, a system could write an article based off our conversation, but you still have to do quite a bit of editing. In a year or two, the editing is going to be less. In three or four years, even less so. And at some point, the article the system produces is going to be so good that you’re going to say, “Well, why should I spend my time editing this?”

At this point, the system is by, no means, perfect. It would still take you and me to go over that text afterward and say, “Here, it did really well and here it screwed up and hallucinated,” but those moments of hallucination are getting fewer and fewer. These systems are getting better.

Another thing that ChatGPT can do quite well is brainstorm three or four titles for an article based on an interview. It comes up with a few ideas and then you can pick and choose and maybe combine. It’s really already quite capable already of doing that.

Q. If cognitive workers are increasingly at risk of being of automated, what can they do to prepare for a potential change?

A. In the near term – the next couple of years – I would advise people to familiarize themselves with these systems, because they can use them to their advantage and become more efficient in the work that they are doing. Those that use these systems are going to have a leg up on those who don’t.

In the longer term, we have to remind ourselves that work is a means to an end. In the past, work has always been valued by society because it was the basis of our survival. Before the Industrial Revolution, we worked to create our daily bread. At the start of the Industrial Revolution, some of us started to work in industry, and then in services.  

But work is not an end in itself. If advanced machines can perform our work better, then it would be a waste of our time to continue to work. If everything is being automated, those of us who still like to write articles or research papers because they enjoy it, they can still do that. No one’s going to prevent them from enjoying it.

Q. But could there still be a way for those folks to make money doing that?

A. This is indeed the main concern. If machines can automate only part of what we are doing, then the remaining parts become more valuable, and our incomes go up – as has always been the case since the Industrial Revolution.

But if things change and machines can automate literally everything, then the value of human labor would be greatly diminished. Why would we continue to pay white-collar workers large salaries if machines can do an equal or better job at a fraction of the cost?

But there is no need to be filled with dread quite yet. If machines really become capable of doing everything that we can do, then economic growth would also take off. We would experience an era of abundance like humanity has never seen before. Even if we are not working, we could all be far better off than today if we manage to share just a small sliver of the windfall that will be delivered – for example, if we manage to distribute a fraction of ownership in the most advanced AI systems across society or to provide a universal income to everyone.

Korinek has recently explored these policy options in a research report entitled “Preparing for the (Non-Existent?) Future of Work” that was published by Brookings.