On a sunny day in the summer of 2015, I found myself standing in the basement of an office building in Silicon Valley. The room was packed full of appliances, as if a commercial kitchen had been blown apart. Several stacks of wall ovens huddled together on dollies. Tangles of electrical cords popped up behind a broad stainless-steel counter. Rags and sponges dried slowly on the edge of a giant sink, over which a pre-rinse nozzle drooped like a willow branch. Nearby, a computer monitor showed Python developer tools next to a spreadsheet dotted with red and green cells. In the middle of the room was an improvised prep station on which there was an odd little machine. It looked like a tiny microwave designed to warm up a single cupcake. I later learned it’s a moisture analyser: the researchers use it to see how much juice is left in their samples of lamb, swordfish, chicken, pork and beef. I’m writing down everything I see: 10:45 am. The kitchen is full of smoke.
That basement room was a test kitchen, where two chefs and several researchers had been busy cooking a whole farm’s worth of meat in a set of prototype ovens. The company’s vision was to invent a cooking machine that could whip up Michelin-grade dishes at the press of a button, and the crew was tasting the food and filling out ‘sensory inventory’ forms that recorded qualities like colour, smell, temperature, texture and taste.
As the design lead for the oven’s touchscreen interface, I was there to learn from the chefs in order to build something that worked for them and for the ‘home chef’ or end user. The tension between their culinary intuition and the ovens’ logic was immediately apparent: the kitchen was full of smoke because one chef had attempted to cook three racks of lamb directly under a broiler to get the colour you’d get by ‘smacking it with a propane torch’. This instinct didn’t translate to a language the oven understood. I wrote down Smacking? and kept watching. The other chef, tapping away on another of the ovens’ screens, muttered: ‘These things are worthless. I need to be able to use my chef spidey senses.’
The chefs were hired to help program the ovens to follow their recipes. But they also had to pretend they were normal people cooking dinner for their families in order to see if the ovens could replicate expert techniques. On one hand, the chefs needed an interface with enough controls to make manual adjustments, adding to the oven’s corpus of training data. On the other hand, end users would need simplicity and the impression of smarts.
The company building the ovens viewed human intervention in the recipes as a liability. Having too many options would invite users to fiddle with the recipes, which would cut against the ovens’ smarts and risk producing a bad result.
|
Know Your MBA Exams |
|
