Food Networks and Creatively Computed Cooking

Not only that data analysis has made further inroads into our understanding of cooking, cooking is apparently changing with computing – or at least, that is what IBM has been using its Watson for.

IBM Watson Food Truck
IBM Watson Food Truck (img © IBM, Quelle)

But, let’s rewind a little:

The idea that different cuisines are different in (large) part because they employ different combinations of different ingredients which account for their different flavors has been going around for quite a while. It was suggested by Elizabeth Rozin under the term of “flavor principle” back in 1973.

More recently, there has been a related idea which assumes that foods are typically combined the way they are because they taste good together thanks to shared chemical compounds that give them their flavor(s). Recipes were analyzed and networks of flavors built from that, presented in “Flavor network and the principles of food pairing“, published in 2011.

IBM has just recently let its artificially intelligent computer system Watson loose on a database of foods, flavors, and cuisines. Thus, they created a system capable of “taking” a main ingredient, being told what style of cooking the resulting dish should be like (or instead telling what style or styles served as inspiration, and creating fusion cuisine), and suggesting ingredients from there, based on how that particular cuisine combines flavors, what flavor compounds are contained in the ingredients and taste well together (or contrast the way some cuisines prefer it), and with a measure of the novelty factor of these combinations thrown in for judging the likely surprise effect.

It’s technological co-creativity at its most haute, for the chefs of the Institute of Culinary Education who they teamed up with still had to decide which suggestion to pick and how to prepare it.

I wonder about a few things.

For one, this could allow for some testing about flavor principles, which seems not to have been done because everyone’s so in awe about the creativity in the computer. What combinations would the system suggest, though, if it’s told to create something belonging to just one cuisine and having just about no surprise effect (but perhaps a maximum pleasantness)? Would that be something recognized as a typical combination of the cuisine in question?
(It would pretty much have to be since it would be straight out of the recipe databases – but who knows? By the sound of the system description, it could just as well spit out many, many more combinations than are actually used.)

Also, I would love to know if there are any ingredients/flavors that none of the cooking styles analyzed would combine, but that’s more of an aside, thinking of how humans seem to eat just about anything.

The main point that got me to thinking about flavor networks, though, has of course been the chilli.

Thing is, I just recently looked at the article about flavor networks again, including the supplementary material, and there is a role that the chilli is playing.

In fact, if I’m not mistaken, it seems to be suggested that chilli is rather more authentic/typical an ingredient of East Asian cooking, according to that research, than even soy sauce. (And it’s equally important in Latin American cooking.) Interesting result, that. For one because it wouldn’t be entirely surprising – and secondly, because it doesn’t seem like that should be right, given that there are regional variations in East Asian cooking, and only some of them utilize chilli a lot, others hardly at all.

Furthermore, it was interesting to find that there was some recognition of the different flavors in different chilli. The differentiation used in that research article, however, was between such kinds of capsicum as Thai chilli and tabasco, cayenne and green bell pepper, but it missed green chilli and habanero and doesn’t make it clear what exactly – other than the name – the difference they see between e.g. Thai chilli and tabasco would be. (Tabasco is rather easy to imagine, I’ll admit, but Thai chilli doesn’t necessarily say much – and others say rather less again.)

Ultimately, of course, the question is one of taste and tradition: This feat of cognitive computing seems remarkably adept at finding new and creative combinations – but who’d want new things all the time, let alone have to figure out how to best prepare and present them?

Leave a Reply