IBM's "Cognitive Cooking" Food Truck

I'm not ashamed to admit that "Wait, Wait... Don't Tell Me!" is one of my main sources of breaking news, and that's where I first heard that Watson, IBM's own Jeopardy champ, is running a food truck at South by Southwest. Of course, I had to look into it...

A joint venture between IBM and the Institute of Culinary Education, the food truck is an exercise in what IBM (rather bloodlessly) calls "cognitive cooking" -- a street-food demonstration of the practical applications of their "cognitive computing" system, aka Watson. Would you like to read an advertorial about it in Slate? Here you go. And here's IBM's promotional website about the cognitive cooking project. 

This is how you use it. You have to input three things: the main ingredient, the cuisine (eg, Indian, Azerbaijani, Canary Islander...), and the type of dish (eg, burrito, bisque, sandwich). (At SXSW, the type of dish was left up to a Twitter vote, and I suppose the other variables were supplied by IBM.)  Watson then reviews the vast universe of possible combinations, modeling the flavor chemistry of each component and its interaction with other flavor compounds, as well as the potential taste appeal of the final dish and how novel the combination is. It outputs a set of recipes comprising 12 to 14 ingredients, each with a rating based on its assessment of flavor interactions, likeability, and surprise. Just like on "Chopped," you're judged not only on taste but also on "creativity." The goal is to come up with something that's both "weird" and "good."     

[An aside: What is it about the times we live in that makes cross-cultural comminglings the apogee of "weird" cooking? "Indian turmeric paella," are the first words out of the advertorial's mouth. "Peruvian poutine," "Swiss-Thai asparagus quiche," "Austrian chocolate burrito" are all dishes featured in the cognitive cooking recipe archive. Are these combinations really so strange, or unimaginable without cosmopolitan Watson to liberate us from our parochial attachment to thoroughbred cuisines? This is not, I think, simply a retread of the 90s vogue for "fusion," which sought a diplomatic accommodation between US appetites and "exotic" (usually Asian) ingredients and techniques. All the borders have come down; materials and methods can be freely recombined without tariffs or translations; culture is just another seasoning. Should we call this "world markets cuisine," globalism's dinner plate, neoliberal gourmandise?]     

IBM's challenge is to prove to all of us that Watson isn't just some better sort of Google, a more refined filter for sorting relevant from irrelevant, signal from noise. What IBM wants to demonstrate is that Watson can provide creative or unprecedented solutions, things that don't just work right but also "feel right." As the Slate advertorial puts it, "A system that can generate new things the world has never seen before is a significant step in cognitive computing."

This is actually a rather tall order, especially as IBM is always careful to insist that "cognitive computing" is not a replacement for human creativity (the brain is "the most creative computer of all," in their words) but a tool to enhance it. The decision to use food -- and, specifically, the creation of unusual flavor combinations -- as a debut showcase for this technology is thus very deliberate, and taps into a longer history. Sure, the marketing team has festooned this with all the right merit-badges -- hipster foodies and their food trucks, Twitter crowdsourcing, SXSW, "the cloud" -- to gain likes and influence retweets in those zones of social media where knowing what's "trending" counts as connoisseurship. But the problem of meshing these two kinds of information about flavor -- what IBM refers to as "chemoinformatics" (ie, its chemical behavior) and "hedonic psychophysics" (ie, our sensory experience of it)  -- is something that has daunted the flavor industry since, at least, the mid-twentieth-century.

I've just been reading the proceedings of the 1961 Flavor Chemistry Symposium, hosted by Campbell's Soup at their old HQ in Camden, New Jersey. This was one of the very first scientific conferences devoted to this chemical subfield. (The Society of Flavor Chemists, the first professional organization, had been inaugurated less than a decade earlier; the American Chemical Society wouldn't create a flavor chemistry division until six years later.) The papers from this conference makes it clear how rapid progress has been in the field: more and more, the molecular structure of flavor compounds, their chemical precursors and interactions with other molecules during cooking and preparation, how they degrade, what influences them, and so on, are being quantified, verified, understood. As Carl Krieger, the director of Basic Research & Product Development at Campbell's remarks at the kick-off of the conference, there was a new "realization that the mysteries of flavor can be solved."

Except. Except that "the physiology and psychology of taste, odor, and flavor" are still vast unknowns. Krieger ventures that only by making positive identifications of flavor chemicals "will it be possible to describe flavors in universally meaningful terms" (ie, by their chemical names) rather than the subjective terms of experience -- "metallic," "stale," "rancid," -- "which, I must confess, seem to me to be pure gibberish." Thankfully, Krieger concludes, their conference will not focus on perception of flavors, but their chemistry - "something that I believe all of us feel is more amenable to direct experimental study." 

Okay, that's all well and good for Krieger to say, but knowing what the flavor compounds are doesn't answer the million-dollar question: "Will people like it?" That's a big missing piece of the puzzle -- the gap between the chemoinformatics, so to speak, and the hedonic psychophysics. Flavor companies -- and the US government, especially the army -- labored to make flavor evaluation "objective," to standardize descriptive vocabularies, to train tasters and impanel consumers to supply their opinions before a product hits the market. But these studies always involved human beings, unruly instruments on their best days, and their subjective responses are, by definition, not generalizable -- do not produce the "universally meaningful terms" that Krieger claimed chemistry did.

And this, fundamentally, is what IBM claims is different about its "cognitive computing" model, and what it's trying to show with this food truck project. We're quite used to claims like "chefs can only consider combinations of two or three ingredients at a time; computers can contemplate quintillions" -- yes, computers can outfox even the foxiest human thinkers. This system doesn't just crunch numbers, it makes judgments about subjective sensations. As the IBM advertorial tells us, it "understands why thousands of different recipes are appealing, what people prefer." Here's the crux of the claim: "It understands, learns, and considers not just big data but also human perception."

These two things -- big data, human perception -- continue to be held at arm's length from each other. But isn't the promise of this technology, in fact, that it successfully converts human perceptions into data, data that the machine-system can "consider" and that are susceptible to the same tools and techniques that guide the collation and analysis of other forms of 'big data'? The dream realized here is that we will finally be able to bring subjective experience into the same table that we use to calculate agricultural yields or profit margins.

What is supposed to make Watson different, I think, is that it claims to formalize the bodies of knowledge that have so far resisted formalization. Things like intuition. Experience. What we in the STS biz call "tacit knowledge" -- the kinds of things you learn by practice, by doing -- like how to make fine adjustments to instruments, or to hone a curve on the form of a chaise lounge, or to add a new ingredient to a recipe. Not just the look of things, but what we felt at what we saw. But Watson enters a crowded field, because our "personal technologies" increasingly aspire to recognize and cater to our subjective preferences. Like when Netflix deduces your taste in movies, not merely spitting out a list of other black comedies, but synthetically tailoring for you an array of "Dark GLBT Comedies with a Strong Female Lead." Or the new music data venture that scans Twitter for early "flickers of excitement about a fledgling band," "the kinds of signs music scouts have always sought." The Watson system isn't just about helping General Foods design new crazy flavors of potato chips; IBM promises that the applications for cognitive computing are in all fields that rely on "design and discovery." This isn't a technology that competes with Google; it's technology that competes with technicians and so-called knowledge-workers -- designers, flavorists, A&R divisions, R&D folks -- highly skilled workers whose refined, intuitive knowledge of their fields are supplemented (or supplanted) by "cognitive computing."

But fear not! Our cherished celebrity chefs won't be driven to extinction by our new networked overlords. "Cognitive computing is a sous-chef working alongside seasoned professional chefs." Right, it's not Emeril's job that's at stake, but those of his unnamed assistants, who will surely still be required to slice and dice -- Watson, after all, doesn't have hands to get dirty -- but perhaps less entrusted with the fine adjustments and refinements, with the knowledge side of technical work. (Similar, for instance, to what Deborah Fitzgerald calls the "deskilling" of farmers after the introduction of genetically modified hybrid corn.) Or maybe not. Maybe systems like this really do foster innovation, break down the barriers that have hitherto prevented us from dreaming up a Swiss-Thai quiche, an Indian paella.  

I should wrap this up on a less lugubrious note. So I'll add that, the consensus on the internet seems to be that Watson's food was pretty good and somewhat novel, though some were disappointed that it was prepared by humans and not robots. Brillat-Savarin said it, and I believe it: "The discovery of a new dish, which excites our appetite and prolongs our pleasure, does more for human happiness than the discovery of a star." The question, I suppose, is how you define "new," and what you mean by "discovery."