SxSW 2017: Lose weight with AI

Since about half of all presentations at SXSW have the magic word AI in their title, it was inevitable that I would end up with a presentation on AI. “You are what AI cooks: Personalisation of Taste”. A handful of Japanese people (and three food bloggers) sat in a small room ready for Yoshiki Ishikawa. A professor from the School of Health Science, Faculty of Medicine at the University of Japan. For fifteen years he investigated weight management in humans, but came to the conclusion that mankind is simply unable to manage his weight. Something that is painfully evident here in America.  Together with a few other researchers and a chef, he embarked upon another mission:  to develop an AI that knows better than yourself what you find tasty (and what’s good for you).

A computer that cooks for you is nothing new.  In 1969, Honeywell Robotics had already come up with their own computerised kitchen. In those days, if you had ten thousand dollars and some serious engineering skills, then this computer was able to tell you what ingredients go well with your beef and broccoli.  Ultimately, not one single machine was sold, as it was mainly a marketing stunt.  The computer existed and worked, but the fact that you had to enter the ingredients directly into binary code and the response of the Honeywell computer had to be read as a series of flickering lights, did not make for an easy sales pitch. Incidentally, the ten thousand dollars gave you the right to make use of a two-week user’s course.

Last year, 47 years later, Moley Robotics introduced a real kitchen robot.  This year, the world’s first robotic kitchen can actually be purchased. The amount on the price tag is seven times more and the execution is a lot more ambitious than that of Honeywell. You select a recipe listed on the touchscreen and the two robot arms prepare a full meal behind a glass wall.  The movements were taught by using images from real chefs.

Animated GIF  - Find & Share on GIPHY
Yoshiki’s story isn’t about the robot itself, but about which intelligence such a robot needs.  The first step was to create a flavour network in which (almost) all of the ingredients in the world have been mapped out. Which ingredients complement each other, strengthen each other, and which food group do they belong to? For example, did you know that a combination of milk and beer works well together?  No, really.  According to Yoshiki, this combination tastes like coffee, and I’ll take his word for it. Incidentally, coffee also goes really well with peanut butter and beef.

Once the computer has understood our tastes and the combination of ingredients, any and all possible recipes found by the team (including going through the entire recipe website Yummly) are put into a three-dimensional matrix: The Food Galaxy. Using mathematical formulas, which were all nicely incorporated into the powerpoint, an enormous navigable network of thousands of recipes, recipe styles, and how they relate to one another is created.

You can also add a recipe that you’ve often used or you can add all the ingredients in your filled-up lunchbox.  The system calculates where you are within The Food Galaxy.  You can also specify which part of the world you want to explore – or according to the computer – which part you should explore in order to eat healthier and tastier.  If you move from America to Japan, The Food Galaxy determines which ingredients you can replace with those you used often in the States.  Type in a typical French meal and let The Food Galaxy translate the same type of meal Japanese style.  Yoshiki Ishikawa calls this computational creativity, the ultimate goal of AI. A learning system which comes up with creative solutions better than we could ourselves.

Currently, this is a fairly manageable concept, but Yoshiki believes that eventually computers will know better than ourselves what is good and tasty for us.  At the moment, the team is developing an AI that observes what we eat and learns what we think is tasty, but also what we should be eating.  Combine this with a robot like Moley’s Robotic Kitchen and we’ll never have to cook again.  Not only because we don’t have time for it or maybe because we don’t like cooking, but because our robotic chef can simply do it better and knows.

Three years ago, there was a food truck at SXSW in which Chef Watson, the super computer from IBM, created recipes based on the favourite ingredients of the visitors.  In 2012, Professor Varshney, the supervisor of the Chef Watson Project, predicted that computers would be able to determine what our preferences were, better than we could ourselves.  Yoshiki took on this challenge and the first step has been made. His next step is clear: the saying “you are what you eat” will change into “you are what AI cooks for you”.