One of the more hopeful scenarios for how artificial intelligence could affect jobs is that it would take over more of the boring grunt work and free up humans for loftier pursuits.

Mondelez, the company behind many of America’s favorite snacks, like Oreo cookies, Sour Patch Kids candy and Ritz crackers, is trying to do just that — using AI to speed up innovation for food scientists and give their taste buds a break.

Marketplace’s Meghan McCarty Carino spoke with Wall Street Journal reporter Isabelle Bousquette about how AI is changing the snack game. The following is an edited transcript of their conversation.

Isabelle Bousquette: When they are, you know, creating these new recipes, it used to be a process that was, you know, fairly trial-and-error based, and the process is still fairly similar now with this AI tool, except what the AI is doing is it’s giving the scientists the ability to actually go in and tick the box for the flavor characteristics that they’re looking for. They can tick a box if they want it to, you know, taste more buttery or more oily or eggy, or have more chocolate chips, or be rounder, less round, or, you know, it’s really endless, like the amount of characteristics they can tick there, and they don’t all apply to, like, every product they create. And they’re still kind of taste tested. There’s a lot of human in the loop, but it’s basically making that whole process, like, four to five times faster.

Meghan McCarty Carino: So how has it kind of changed the workflow of food scientists working in this?

Bousquette: It’s still developing, so it won’t necessarily be used in every product that comes to market. I think they have a rule that if they’re changing fewer than three or four ingredients, it’s maybe not worth [it] to go through the tool. You know, there’s several people involved in the process, and they’ll sort of work with brand stewards to make sure that they’re not, you know, pushing the remit beyond the core of, you know, say, what an Oreo should be. But yeah, essentially it’s, you know, just making the process a lot faster and, you know, just getting them those recipes, that they used to have to put in the manual work to do, in a more automated way.

McCarty Carino: Yeah, tell me more about some of the limits for this tool and some of the maybe challenges that have been encountered.

Bousquette: Yeah, yeah. I mean, one of the things is like they have to give the tool a lot of their own data, of which they have a lot because they’re a massive company. You know, [in] earlier iterations of the tool where maybe they didn’t give it enough data, it would suggest things that were, you know, way out there, like a cookie that was, like, almost entirely made of baking soda because it just didn’t really understand what a cookie was. I mean, they’ve spent several years basically training and priming this tool to, you know, really understand, like, the attributes of all their different products and what they should all be. And again, they’re still double-checking everything for sure.

McCarty Carino: And in this case, it sounds like, you know, maybe freeing some workers up from the kind of onerous responsibility of so much taste testing that not all of them relish.

Bousquette: Yeah, yeah, absolutely. I mean, I think the fact that this is going to accelerate time to market and basically reduce the number of, you know, iterative tastings that have to happen for any given product was a relief to some of the staff there that basically sometimes have to do tastings, you know, up to four to five days a week, which I thought sounded fun, but I guess they were like, when you work in a company like Mondelez, the amount of tastings is not fun. It’s hard to give, you know, tasting feedback on products where the differences are sort of minute, and when you’re doing it so much, I think sometimes even your feedback, you just kind of lose a sense of what you’re even thinking. So, yeah, I think fewer tastings per product [is] definitely a relief to some.

McCarty Carino: How much is Mondelez sort of acknowledging or promoting the use of AI in its products? Obviously, they had you come report on it.

Bousquette: Yeah, no, it’s funny. I mean, and again, like, they’ve been working on this tool for five years, and it’s been used in, you know, 70 different products at this point. But as a consumer, you wouldn’t know the fact that Mondelez is using it. It’s helped them respond to consumer demands faster. And the last couple years have been kind of challenging for the consumer packaged goods snacking sector, consumer wallets tightening. They’re, you know, scaling back on treats and so, you know, maybe it’s a situation where they would rather go for, you know, a packet of minis or a packet of thins, rather than sort of, you know, the traditional big package. And Mondelez is in a position where they can sort of respond to that quickly. So I think that’s something that’s helped them, even if the consumers don’t actually know, like, oh, this was made with AI, how cool.