Dammit Jim! I'm a Doctor, not a Chef!


Humankind’s pursuit of technological advancement and Artificial Intelligence (AI) has taken yet another bold step as an eight year old super computer - IBM’s Watson - has started focusing on food and recipes.

Watson was Originally built to play in the American quiz show “Jeopardy!” where contestants are read answers by the host and have to reply in the form of a question. In much the same way that it’s predecessor Big Blue, dispensed with the former human World Chess Champion, Watson succeeded in beating two former Jeopardy! champions to scoop the $1 million prize.

Nowadays Watson lives in the “cloud” and no longer resides inside a single physical box. But, it’s fascinating that “he” or “it” appears to have the ability to learn, to evaluate hypothesis, generate ideas and understand language. It’s these abilities which IBM sees as allowing Watson to provide linguistic answers to questions which require understanding unstructured information.

Before we look into Watson’s foodie skills and credentials - it’s worth taking a quick side-track into the mysterious world of Artificial Intelligence and language and in particular the Turing Test. The test was introduced by Alan Turing in 1950 with the words "I propose to consider the question, 'Can machines think?'" Because "thinking" is difficult to define - Turing devised a test using language as the key indicator. Unlike say, a game of chess, language is significantly more complex. Whereas Chess has a finite number of moves in a particular game or position - language involves layers of meaning resulting in almost infinite possibilities. In England for example - and this may well be a peculiarly British traite - one person may see the other looking-up at the sky and ask “will it?” to which the other will invariably understand the question to mean ‘will it rain’? In some Native American tribal languages the word “tomorrow” refers to the the next time it will rain. René Descartes suggested that language defines our reality and so it goes on. The point being that language really is jolly complicated and serves as an excellent indicator of our ability to understand and reason.

So, the basics of the Turing Test involve human examiners having a conversation over a computer terminal where they have to determine whether the entity on the other side of the screen is another human or a piece of artificial software. Until now the Turing Test has not been conclusively and properly beaten. Earlier this year there was a news item about an examiner having been fooled - but it transpires he was fooled to the extent of a somewhat petulant 13 year old avoiding any questions and so not really exhibiting the linguistic abilities the test is designed to reveal.

We don’t yet have the sort of Universal Translator envisaged in Star Trek - or even an Earthly equivalent - because despite the best efforts of Google, Skype/Microsoft and huge advances in voice recognition and automated text translation - we haven’t cracked language yet. Remember, Google isn’t actually in the “search” or “media” business - they are in the AI business. Our human ability to use language remains one of the defining elements of our intelligence. In other words, when the Turing Test has been properly and completely beaten - and that day is edging ever nearer - we will have AI in a few years but that day is not here yet...at least it doesn’t appear to be.

Now, back to food. For the past two years Watson has been taught some of the basics of classical well established recipes, along with a plethora of food combinations and cooking systems. Naturally there’s also a fair amount of science and chemistry involved too - so molecular cuisine and gastronomy is no longer reserved for the likes of Ferran Adrià or Heston Blumenthal.

Some examples of unusual menu items and unorthodox combinations were presented at IBM’s Zurich headquarters as “Chef” Watson demonstrated an early version of the cooking system which is due to be refined in the next few months. (In egg terms that must make it a beta version!)

For starters, a beefsteak tomato taco, including a splash of cider vinegar, vanilla and a couple of twists of lemon peel. This was followed by broccoli soup with slices of fresh mango, a dash of cider and a small amount of orange juice. The main course was a chickpea ragout with Middle Eastern flavours and the unusual addition of plums and olives, accompanied with a bottle of Bengal butter-nut barbecue sauce. Dessert was green-tea pudding with whiskey, tapioca and cardamom.

The Times interviewed the the renowned Chef Michel Roux Jr of the excellent Gavroche restaurant, who reportedly commented “the soup sounds awful, but my palate tells me the green-tea pudding would work.” The great chef went on to say he didn't think computers would take over in the kitchen “because you can’t, in all honesty, beat a human sense of taste”.

According to IBM - Watson’s creative culinary process involves a large database of well established recipes combined with formulae used to calculate a pleasingness scale of sensations relating to smell and taste - which is the intriguingly named field of “hedonic psychophysics”. It makes sense considering that our experiences of most pleasures or unpleasantness are due to psycho-physical factors combined with factors to do with our internal state - that is the hedonic elements. All of this is a somewhat highfalutin way of explaining the well established example that we generally find the pleasantness of sugar increases when we are more hungry but decreases during satiety. The “pleasantness” of a particular ingredient or sensation, comes down to extremely precise chemical properties including atom counts.

Beyond hedonic psychophysics, Watson groups ingredients based on their compatibility. Research suggests that when two compounds are mixed and smelt, there’s a linear relationship to between the pleasantness value of the individual ingredients and the olfactory pleasure in giving the mixture a good sniff. It’s well established that smell is massive part of the eating experience and so the Watson’s approach is equally concerned with smell.

Steven Abrams - one of Watson’s developers - explained that when programmed to use a specific ingredient to invent a recipe, and after examining billions of potential combinations, Watson creates short list of around 100 suggestions ranging from “unique” to “classic”.
“If your goal is to see whether Watson can come up with something completely disgusting, I can promise you that the answer is ‘yes’.”

At this point it’s interesting to refer back to a piece we published last month On The Language of Food, where we noted that the language used to describe ingredients or to review restaurants, revealed much about society, our economy, subconscious and perceived status - not to mention the establishments being reviewed. In this respect, Watson may well create some novel and wonderful taste combinations, but it will have to compete with the “eternal golden braid” of human memory which accompanies the culinary traditions around the world and our perception of various ingredients and dishes.

It also seems that Watson may be entering into a realm not dissimilar to art and music - whereby they may be linked on a mathematical basis, they remain creative endeavours of the soul which speak a language beyond words. In this respect, I remain a huge fan of Douglas Hofstadter’s excellent book “Gödel, Escher, Bach: An Eternal Golden Braid”, also known as GEB,  with the tagline "a metaphorical fugue on minds and machines in the spirit of Lewis Carroll”. Perhaps we may need a new edition renamed ““Gödel, Escher, Bach, Escoffier..” GEBE.

In 2001 A Space Odyssey all the talk of AI gave one the heebie jeebies - but seeing as we’re now talking about the GEBEs in a completely different light - perhaps AI will simply make us hungry for more delicious taste sensations, and on that note it’s time to try some mango pieces in my spinach soup.

You Might Also Like