Computers have tremendous capacities for storing information and performing numerical calculations—far superior to those of any human. Yet, when it comes to other capabilities, including creativity. computers are woefully inferior to people. But a young IBM Research scientist, Lav Varshney, believes that before too long computers will indeed be creative.
His work concerning the sense of taste and food recipes, which we featured in our IBM Next 5 in 5 predictions last week, was highlighted on National Public Radio’s All Things Considered broadcast on Christmas Day.
Varshney and his colleagues are developing a kind of computer intelligence, which they call computational creativity, that’s similar in some ways to IBM Watson, the machine that beat former human grand-champions on the Jeopardy! TV quiz show.
Watson was really good at ingesting vast amounts of information across a wide variety of subjects, understanding the rules of Jeopardy!, and coming up with a set of answers in which it had some level of confidence.
The technology that Varshney and his colleagues are working on draws on a large inspiration database that includes food recipes, molecular information about the ingredients in food and psychophysical insights about people’s likes and dislikes when it comes to chowing down. The goal is to formulate new recipes that are both tasty and good for you.
The secret sauce in the team’s technology is based in part on Bayesian inference, a statistical method for analyzing evidence about the world based on the degree of confidence the machine acquires. The Watson scientists used this technique, as well. What’s new in computational creativity is that the system doesn’t just reason about existing recipes that are likely to be both healthy and tasty. Rather, it actually creates completely new recipes from the molecule up based on the knowledge it acquires.
Because people like variety in their diets and are drawn to surprising things, Varshney and his colleagues use a technique that evaluates new recipes not only on the basis of healthfulness and flavor, but also on how different the experience is from what people have known before–its novelty. They hope to be able to expand people’s perceptions about what tastes good.
In one of their recent experiments, they came up with a new dish–a grilled eggplant-ginger sandwich drizzled with blueberry sauce–which tasted quite good, according to Varshney. Who would have guessed?
The point here isn’t to replace human creativity with machine creativity. Rather, it’s to give people additional tools that help them make good dietary decisions or try things they might not have considered before. “This is an extra thing, a supplement,” says Varshney. “We’re not pushing humans out, but expanding the frontiers of human experience.”
Varshney published this blog post about his work on the IBM Research blog.