Why AI is not coming for our jobs — yet
Jan 18, 2023

Why AI is not coming for our jobs — yet

Computer science professor Mark Finlayson says "generative artificial intelligence" tools like ChatGPT can create coherent material and may produce new jobs for people in creative occupations.

Now that so-called generative artificial intelligence models, such as DALL-E and ChatGPT, can create impressive visuals and formulate complex responses, will human artists, writers, radio hosts, and all sorts of creative and knowledge-based jobs, go extinct?

Mark Finlayson, an associate professor of computer science at Florida International University, offered his perspective on this zillion-dollar question in a recent essay for The Conversation.

Finlayson believes that these tools are likely to change creative work, but not always for the worse. He told Marketplace’s Meghan McCarty Carino that he expects a lot of disruption as some people prosper in future work environments and others fail to adapt.

The following is an edited transcript of their discussion.

Mark Finlayson: It may seem like these generative AI tools are pretty spectacularly “smart,” but what you have to remember is these systems have basically gone out and memorized the internet at some level. They’ve read trillions of words of internet content, all generated by people. So it’s actually just copying and pasting and recombining in very sophisticated ways things that it’s already seen out on the internet.

Meghan McCarty Carino: Yeah, I may have missed out here, but I feel like I have yet to see ChatGPT be funny.

Finlayson: Well, I mean, no, I’ve seen it tell jokes. There’s a lot of jokes written down on the internet. And it can go out there and combine and recombine them. It can tell stories, fairly simple stories. I mean, the thing that I noticed when I was interacting with it is that the text is fairly generic, very high-level. And this is something that is sort of a consequence of the way the machines are trained, that it’s just looking at everything that’s been written down on the internet, which is a whole lot of stuff of high quality and low quality, right? Much of it quite low-quality. And it’s just reproducing that. So it doesn’t have an abstract sense of what is right or wrong, what is correct or incorrect, factual or not. So you know, the machines, if it’s seen a bunch of factual stuff, it can reproduce factual texts. But if it’s read a bunch of nonfactual stuff about some random topic, it’s, again, very hard to determine which bits are suspect in training data. If you don’t have the knowledge to assess that yourself, then you don’t know that you’re being led down the garden path. So I think, you know, that that ability of people to, you know, have that connection to the real world, to be able to assess what’s factual, what’s not, what’s correct, what’s not, that’s going to be continually and increasingly important in the world. So this process of, you know, generating raw generic, high-level text is probably a task that we’re not going to need people to do.

McCarty Carino: What kinds of new jobs might this technology help create?

Finlayson: If I knew the answer to that question, I could make a lot of money. But what we might see is content creators, or people who are creative knowledge workers, they’re going to take the generic model provided by the company, they’re going to license it, and then they’re going to what’s called fine-tune it. So they’re going to add their own stuff to it, and they’re going to adjust the parameters in a way such that the model can do something that other people’s models can’t. And coupled with the person’s subject matter expertise in a particular domain, they’re going to be able to offer a service which the general model can’t provide. So that would be an example of a job that doesn’t exist right now. But we will probably see this coming up in the next year or two as people will take these models and adapt them to their own niche markets, and then offer that service on the marketplace.

McCarty Carino: This idea that, you know, the robots are coming for our jobs, it’s obviously an old refrain. What can we learn about how other new technologies have, you know, changed the labor force, changed jobs that might inform how we think about these AI developments?

Finlayson: Yeah, you know, I grew up in Michigan in the 1980s, and there everybody was constantly talking about, oh, the robots are coming for the factory jobs. And you know, we’re not going to have any factory jobs anymore. Well, today, there still are factory jobs, even though there are robots in factories. It’s just the job in the factory has changed. Those people get paid more, they’re much more highly skilled, they have much more transferable skills than they did before. So I think this is a good example of how these tools are going to get integrated into the workforce.

McCarty Carino: But just as in the case of manufacturing, I mean, not every worker is going to be able to be retrained to be the boss of a robot, right?

Finlayson: Yes, I think there’s going to be a lot of disruption. There are some people that are going to be able to adapt, but there are going to be people who can’t adapt. And that’s a policy, I think, as a society, we really need to be grappling with much more energetically than we are. I mean, some people are talking about it. But I feel like we’re just a little bit like deer in the headlights right now, where we can see this coming, but we’re not doing anything that’s going to really make a big difference to the people who are going to be on the losing end of that equation.

McCarty Carino: How are you using these tools? And what are you most excited about in terms of what they could mean for the future?

Finlayson: Well, I do research in artificial intelligence. And so I’m, you know, trying to develop technologies that push forward our ability, the ability of machines to understand natural language, and until ChatGPT came around, these language models weren’t actually able to generate stories in any meaningful way. I mean, they could generate something which kind of looked like a story, but it wasn’t very coherent and it wasn’t very compelling. But now ChatGPT seems to be able to spin up or spit out basic folktale-like structures pretty easily, and so, you know, it’ll be interesting to see how that plays into my work in terms of getting machines to understand stories rather than just generate them.

We’ll include that piece from The Conversation where Mark Finlayson goes into more detail on these issues at our website.

That article also includes commentary from other experts. Like Lynne Parker of the University of Tennessee, who notes that these generative AI tools raise big legal questions about plagiarism. As Finlayson said, these tools are trained on existing work on the internet — some of which is copyrighted.

So what counts as fair use, what’s transformative and what’s straight up copyright infringement are issues we’ll dig a little deeper into in future episodes. Parker also points out that although these types of AI democratize access to creative tools, they could accelerate the loss of skills that might be pretty important, like writing.

That’s something Daniel Herman is thinking a lot about. He’s a high school English teacher in Berkeley, California, who we had on the show last month.

He fed ChatGPT some of the essay prompts he gives to his students and was both impressed and disconcerted by the results.

But rather than reflexively thinking of this technology as a bad thing, Herman compared it to the calculator. Being able to multiply 987 by 987 in your head is probably not a necessary life skill, he said, and he’s thinking about what type of writing and reasoning skills students will really need in the future and how tools like generative AI might fit in.

Now, it’s easier for younger people to adjust to a disruptive technology, of course. Folks who have already made a long career based on skill sets that become obsolete might have a harder time, according to research by Northwestern University that examined which workers suffer most when new technology arrives.

They found wages for workers 45 to 55 years old grew more slowly than the wages of their younger peers in similar jobs, presumably because they had less familiarity with, and less time to pivot to, the new ways of working.

The future of this podcast starts with you.

Every day, the “Marketplace Tech” team demystifies the digital economy with stories that explore more than just Big Tech. We’re committed to covering topics that matter to you and the world around us, diving deep into how technology intersects with climate change, inequity, and disinformation.

As part of a nonprofit newsroom, we’re counting on listeners like you to keep this public service paywall-free and available to all.

Support “Marketplace Tech” in any amount today and become a partner in our mission.

The team

Daisy Palacios Senior Producer
Daniel Shin Producer
Jesús Alvarado Associate Producer