By Stove, I Think He’s Got It: IBM’s Watson in the Kitchen

IBM has been working on expanding the capabilities of its supercomputer, nicknamed Watson for some time now. One of the major innovation is the Cooking with Watson App that was released to the public last month. The app features the ability to input up to four ingredients and then Watson takes it from there. The app – IBM Chef Watson ( https://www.ibmchefwatson.com ) – was beta-tested by the Bon Appetit team, after their database of 10,000+ recipes were added to the Watson database.

One of the benefits of using Watson is that the supercomputer can analyze all recipes containing the requested ingredients (including quantities) and generate a brand new recipe based on its conclusions.

Perusing the created recipes shows a diverse range of flavors and food combinations. Once the recipe is generated, Watson makes any disclaimers up front about any ingredients it thinks will work well, but that it is unsure of the quantities. It asks for feedback on those parts of the recipe it is not sure are correct. For example, a tomato tart recipe notes that “Chef Watson is pretty sure that orange zest will taste good in this dish, but needs your help in figuring out the details”, hoping that the recipe crafter has insight that can help Watson make better procedural and flavor choices in the future.

IBM’s Watson is not only making inroads in recipe creation, it is also using its computing muscle to power Watson Explorer, an enterprise solution that allows a company to connect data points to make more informed decisions. By connecting the company’s internal data stores with the internet’s vast stores of data, the system allows companies to make informed decisions on a much reduced timeline.

But let’s get back to recipe creation! For this exercise, I perused my pantry and refrigerator for ingredients that I thought would be weird flavor combinations.

My first attempt uses tahini sauce, maple syrup, cucumber, and olives, and here is the outcome:

tahini

 

tahini 2

 

 

 

 

 

 

 

 

 

The recipe also notes the inspiration for the recipe as well, in this case Pineapple-Glazed Chicken with Jalapeno Salsa. This recipe seems a little strange in its combination of ingredients (such as cranberry and horseradish), and the availability of ingredients (such as blood orange and cranberry) should be driven towards a seasonal subcategory. The app also has the ability to modify a recipe based on style of cuisine. For example, if I select the style “Africa”, the recipe changes significantly with the addition of vegetable broth, sparkling wine, and balsamic vinegar. I have to admit this confused me, since none of these ingredients are African in origin.

Other options allow you to search by dish name, and in many instances this returned no result, which is most likely due to the varied nature of the ingredients I selected.

In all fairness to Watson, I tried it again, but this time I chose blackberry, sage, crème fraiche, and mustard as my ingredients. Here is what Watson came up with:

 

blackberry

 

What I found most interesting is that the recipe once generated allows you to change the dropdowns to explore alternative ingredients. For example, if you don’t have blackberries on hand, then you could substitute kaffir lime, lime, lemon, sweetened flake coconut, grapefruit, or red grapes.

Part of the R&D development process is often taking existing flavors and reimagining them into something new entirely and for this purpose this could be an easy way to play with flavor combinations. While this may seem feasible in the future, the robustness of the program only improves with more trials and the addition of a greater breadth of recipes.

At this time my overall impression is that this app will work well for the home cook and possibly in a professional restaurant setting where chefs are looking to branch out from current recipes. Where this technology would be beneficial to the R&D community would be its ability to interface within the parameters of the company’s needs. This would mean a robust interface that compiles the data within the company database and integrates it with internet based search parameters. This is the shape of future R&D development. Will you be ready when it happens?

 

Comments are closed.