ROBERT SIEGEL, HOST:
From NPR News, this is ALL THINGS CONSIDERED. I'm Robert Siegel.
Scientists at IBM have already built a computer that can beat humans on "Jeopardy!"And now, they are sharpening their intellectual knives to make a computer that might someday challenge competitors on "Iron Chef." NPR's Joe Palca gives us a taste of the future.
JOE PALCA, BYLINE: The idea isn't really to win another TV competition. It's to explore the new field of computer science called computational creativity. Lav Varshney is a computer scientist at IBM Research.
LAV VARSHNEY: The whole goal in computational creativity is to come up with new things that have never been seen before.
PALCA: And not just new things. The computer is trying to create things that are useful or aesthetically pleasing.
VARSHNEY: And in particular, we were thinking about culinary creativity because food is so visceral, right? Everyone eats. It helps define our culture.
PALCA: Varshney says there are two steps to making a computer that can be creative in the kitchen. The first is to give the computer access to a huge number of recipes that are already popular, what Varshney calls the computer's inspiration database. Then, the computer rearranges the recipes, adds new ingredients and generates millions of new ideas for recipes.
VARSHNEY: And then the second step is actually to take those millions ideas and actually find the best ones.
PALCA: That's the tricky bit, since taste tend to be idiosyncratic. Some people actually like pimento cheese sandwiches or pig brains in milk gravy. Varshney says he and his colleagues are relying on a concept from culinary circles called the flavor pairing hypothesis.
VARSHNEY: The basic idea is that two ingredients that share a lot of flavor compounds will go together well in Western cuisine.
PALCA: Varshney says he gives the computer information about the chemical components of various foods and let's the computer generate possible combinations. That's led to predictions of things going together that you might not expect.
VARSHNEY: Things like chocolate and blue cheese, we predict, will go together well or Jamaican rum and blue cheese.
PALCA: I found a recipe for a cake with chocolate and blue cheese on the Internet, so somebody seems to think that pairing works. Now, if you're thinking there's just no way a computer will ever be as creative a chef as Julia Child or James Beard, computer scientist Celine Latulipe says you're not alone.
CELINE LATULIPE: There are still many people who say, I don't think computers can be creative. And if a computer created this, then I don't think it's creative just because a computer created it.
PALCA: Latulipe is at the University of North Carolina, Charlotte. She's pretty convinced computers can be creative and her research is about how you would assess that creativity. She has developed a set of criteria - criteria that could be applied to Varshney's computer to see if it's being creative. For example, there's a quality she calls fluency.
LATULIPE: The number of ideas created. So that would be the number of different recipes that this thing comes up with.
PALCA: And then how original are the recipes.
LATULIPE: Are these actually new and different.
PALCA: And also quality.
LATULIPE: Are they actually good and relevant and useful.
PALCA: The IBM scientists aren't ready to put their computer to those tests yet. They're still in the early stages of developing it. But Lav Varshney says the computer has already come up with some interesting recipes. For example, it proposed a kind of mash-up between an Indian curry and a Spanish paella.
VARSHNEY: It had turmeric and some other Indian spices and potatoes, pork and beef. And then it had kind of a mango rum topping.
PALCA: OK. Sounds intriguing. But how did it taste?
VARSHNEY: I'm actually vegetarian, so I didn't eat that one. But the team did, so they thought it was pretty good.
PALCA: I'll buy that. It actually does sound pretty good. Joe Palca, NPR News, Washington.
NPR transcripts are created on a rush deadline by Verb8tm, Inc., an NPR contractor, and produced using a proprietary transcription process developed with NPR. This text may not be in its final form and may be updated or revised in the future. Accuracy and availability may vary. The authoritative record of NPR’s programming is the audio record.