I'm excited to buy a copy. It looks beautiful, and in addition to the interesting recipes there looks to be a decent discussion of how the computation works (at a high-level, conceptual level; not code).
I have read the book and tried Watson food a few times. The book does not hide the fact that the Watson recipes have been curated/modified by a real Chef (the book usually tells you about the major modifications). So you could ask: how is it really different from randomly generated stuff? My experience is that there is something really distinctive about Watson food: a very unusual association of a handful of tastes that a regular chef wouldn't have thought about but that works surprisingly well.
Utter marketing fluff. Watson cannot do simple arithmetic, and by extension certainly does not know how to cook, nor can it learn to understand any cooking technique. I know folks want to expect a lot from Watson but this is simply not realistic.
I think the interesting part here is that it seems Watson is making recipe suggestions based on Machine Learning (read: statistical) analysis of other recipes.
I do this in my cookingspace.com web app: I use the statistics from about 100K recipes to know which ingredients tend to be most often used together, and I use these relationships to suggest extra ingredients for recipes.
With regard to proportion of ingredients (e.g., acid vs. oil, salt vs. sugar, regular seasonings), surface area, heat transfer over time, and the like, there is some basic math intrinsic to cooking.
All accounts seem to allude to Watson's ability to infer natural language and problem solve, including arithmetic. I have not found anything explicitly saying it can not perform simple arithmetic.
Read the article/press release carefully and note the donation of Watson resources to RPI is meant as a testbed for improving its capabilities. Nowhere will you find an explicit demonstration by IBM of Watson doing math. I've actually worked with Watson, the question about math comes up quite often, and people have a lot of great expectations around the general notion of AI and problem-solving, even wondering if Watson can answer philosophical questions. But, today, Watson is not built to solve these kinds of problems.
Watson's core capability is natural language processing--to discern the "what" of an English question, and to match it with germane texts in its corpus.
This isn't true. Of course there's a lot of math under the hood for Watson and in particular for Chef Watson. Yes, various parts of Watson also have the ability to do arithmetic too - although it's not designed to be something that you can ask to solve your differential equations homework.
What we have done (as I work on Watson and specifically on Chef Watson), is give Watson the ability to parse natural language recipes and information about ingredients from a variety of different sources. We've also hooked it up with nutrition information and various food databases - including some that break down the individual chemicals in each ingredient and identify the way in which different ingredients and components relate to one another.
Take a food combination chart, feed it into an algorithm with a randomizer and grouping parameters, have a chef pick out some good ones, and voila, a computer-generated cookbook to use as publicity for your computer and cooking school.
16 comments
[ 2.5 ms ] story [ 42.8 ms ] threadPersonally, I'm looking forward to seeing how some of these taste.
Can you provide some sources? I assumed that if Watson could win at Jeopardy, it would be able to do simple math problems.
Edit: I am not sure what you mean about Watson's inability to do arithmetic.
"The two major goals for IBM and RPI are to improve Watson's mathematical ability and to aid its recognition and interpretation of new words."
http://www.gizmag.com/ibm-watson-supercomputer-rpi/26038/
All accounts seem to allude to Watson's ability to infer natural language and problem solve, including arithmetic. I have not found anything explicitly saying it can not perform simple arithmetic.
Watson's core capability is natural language processing--to discern the "what" of an English question, and to match it with germane texts in its corpus.
What we have done (as I work on Watson and specifically on Chef Watson), is give Watson the ability to parse natural language recipes and information about ingredients from a variety of different sources. We've also hooked it up with nutrition information and various food databases - including some that break down the individual chemicals in each ingredient and identify the way in which different ingredients and components relate to one another.
http://commons.wikimedia.org/wiki/File:Food_Combining_Chart....
Apparently cooking with AI has some successes in its history.