A computer learns about ingredients and recipes What else would you put on a peanut butter sandwich?

Recommendation engines are everywhere. They let Netflix suggest shows you might want to watch. They let Spotify build you a personalised playlist of music you will probably like. They turn your smartphone into a source of endless hilarity and mirth. And, of course, there’s IBM’s Watson, recommending all sorts of “interesting” new recipes. As part of his PhD project on machine learning, Jaan Altosaar decided to use a new mathematical technique to build his own recipe recommendation engine.

The technique is similar to the kind of natural language processing that powers predictive text on a phone, and one of the attractions of using food instead of English is that there are only 2000–3000 ingredients to worry about, instead of more than 150,000 words.

The results so far are fun and intriguing, and can only get better.

Notes

  1. Jaan Altosaar published an article about his work that gives an explanation of how it all works. It also allows you to investigate the food map and use some of the other tools he built.
  2. A scientific report that may have inspired Jaan (and possibly Watson) to take up the challenge is Flavor network and the principles of food pairing.
  3. That paper offers great explanations for why some novel food pairings work, including Heston Blumenthal’s iconic white chocolate and caviar, published in 2002.
  4. The madcap adventures of Chef Watson are everywhere on the internet. The recommendatiuon engine that is me suggests a report from Caitlin Dewey which includes a recipe for the ubiquitous Austrian chocolate burritos (but no explanation of what makes them Austrian. Just the apricot purée?).
  5. The banner shows a small part of the food map, with East Asian ingredients tightly clustered while North American ingredients are all over the shop.

huffduffer icon   Huffduff it

Good industrial food

Fabrizia lanza evan kleimanLast night I was lucky enough to go to a conversation between Fabrizia Lanza and Evan Kleiman. Fabrizia is taking over from her mother’s world-famous cooking school in Sicily, most notably with an amazing programme called Cook the Farm. And Evan, among her many other achievements, hosts a show called Good Food on KCRW. They were talking about Food and Agriculture in Italy.

The discussion was very interesting, thoughtful and thought-provoking, and here’s the main thought it provoked in me?

Where’s the romance in milking 300 sheep by hand, twice a day?

Fabrizia made much of her local shepherd, who indeed milks 300 sheep by hand each morning, makes cheese, takes the sheep out to pasture, where they spend six hours or so whatever the weather, brings them back, milks them again, makes another batch of cheese and does it all over again tomorrow.

And he has to milk them by hand, because “how else will he know whether they are happy?”

The whole shepherd thing was connected to determining a fair price for food and embedded in a much wider discussion of sustainability. Then, a little later in the conversation, Fabrizia Lanza talked about the idea of creating “a good Monsanto”. She fully recognizes the need for a food industry alongside the romantic, the artisanal and the authentic, but an industry that has values other than pure profit.

So why shouldn’t her shepherd use milking machines? The use of a milking machine does not in and of itself mean that the practice is in any way inhumane or bad for the animals. You can choose whether to rush, treating the sheep as no more than milk-producing machines. Or you can choose to take things a little more slowly, assuring yourself that each ewe is as happy as can be. The milking machine is not the problem; the way it is used may be.

Maybe the shepherd can’t afford such a thing, probably he wouldn’t want to get into debt. My point is only that he doesn’t have to milk by hand to produce an extremely fine product. If the community truly wanted to support him, they could help him to milk by machine. Maybe he could repay a loan with that fine cheese.

This is all part of a much larger problem, which is what Fabrizia was alluding to when she spoke of good industrial food. In many cases, as Rachel Laudan has pointed out on the podcast, Industrial is best. She spoke about salt and sugar, but there are other examples too, including the gentler pressing of grapes that some people say makes more delicious wines, and a high-tech brewery turning out very fine beer. Indeed, Fabrizia Lanza acknowledged that much of the olive oil in Sicily used to be rancid, and now it isn’t. She didn’t say whether this might at least partially be the result of modern pressing equipment. I suspect it is.

And that’s the point, for me. No industrial machinery automatically results in an inferior product. It is the entire system in which it is embedded and used that determines the quality of the food we eat.

Eat This Newsletter 050

6 March 2017

  1. My main reason to recommend Rachel Laudan’s A kipper for breakfast is not to disagree with her preferences (although I certainly do), but to use it to recommend Alan Davison’s throughly delightful book A kipper with my tea: selected food essays. If you don’t know it, you should. And if you do, you should reacquaint yourself with it, just as I wish I could reacquaint myself with the delights of a kipper.
  2. How did the elite East coast urbanites respond to An English Sheep Farmer’s View of Rural America? And rural America?
  3. Does it always take a foreigner to interpret what the locals take for granted? A blog post Following persimmon around the world confirms my prejudice.
  4. I enjoyed this open letter from Emely Vargas, asking why her mother doesn’t teach her brother to cook. Would it be too much to expect a reply? Or to hear her brother’s point of view?
  5. And let me toot my own horn, again. In the wake of last week’s podcast on the cost of a nutritious diet, I took a look at a couple of studies of fat taxes and thin subsidies to ask “What if the differences in how people buy food depend more on whether they are obese or normal weight than whether they are poor or rich?”

Fat taxes and thin subsidies

Following up on the cost of a nutritious diet, a couple of studies: obese people have different spending patterns, and whether subsidised community supported agriculture improves quality of diet.

I think it was Professor David Zilberman of UC Berkeley and his colleagues who first coined the phrase Fat taxes, thin subsidies in a 2004 paper that calculated the potential health benefits of subsidies on certain classes of fruits and vegetables. Those calculations showed that “[e]stimates of the cost per statistical life saved through such subsidies compare favorably with existing U.S. government programs.” The phrase now crops up on both sides of the debate.

The latest meta-analysis, for example, says thin subsidies and fat taxes both do what they’re intended to do, increase the consumption of healthy foods and decrease the consumption of unhealthy foods.

Nevertheless, some say that fat taxes and thin subsidies remain a bad idea because they favour wealthier consumers. Rich people already buy healthier food, so thin subsidies just reduce their food bill. Furthermore, rich people wouldn’t contribute much to the fat tax. Poorer people are buying unhealthy food, so increasing their costs with fat taxes does them no good unless they also change their purchasing habits. One way to do that, of course, is to provide education about nutrition, and there’s a prevalent view that some of the fat tax revenue could be used to provide that education. It seems a bit rum, though, to expect the poor to finance their own education through the fat tax.

Anyway, most of the research to date has looked at how poorer and richer people would respond to these various economic incentives. What if the differences in how people buy food depend more on whether they are obese or normal weight than whether they are poor or rich?

Rich fat Russians

In Russia, at least, it does. Matthias Staudigel, of the Institute of Agricultural Policy and Market Research at the University of Giessen in Germany asked How do obese people afford to be obese? Consumption strategies of Russian households in a paper published in 2012. Russia is a good place to ask the question for a couple of reaons. First, the economy fluctuated fairly wildly during the transition and through the rouble crisis of 1998. Secondly, the Russian Longitudinal Monitoring Survey combines detailed economic data with measures of height and weight for household members. So there are good measures of what household members weigh and how they spend on food under varying economic conditions.

In principle, consumers could respond to higher prices (or lower incomes) in three broad ways. They could allocate a different part of their spending to food, they could buy different amounts of specific foods and they could buy foods of different quality.

Faced with higher prices, obese households (but not those who are merely overweight or normal) tend to buy the same amount of food, but of lower quality. Differences in the amount they spend and in the amount they buy are lower or non existent. Faced with a fat tax, obese consumers (in Russia) would probably just shift to cheaper versions of the same foods. As Staudigel concludes, “policies aiming to reduce obesity should consider deviations in consumption behavior of normal and obese consumers in terms of quality”.

Poor obese Americans

In Russia, obese households are richer than overweight and normal households. They spend more in total and on food. In the US, the reverse is true. Obesity and overweight is associated more with poverty.

In the US, households headed by an obese person have a different spending pattern than those headed by an overweight or normal weight person. They spend more, percentage wise, on cheese, processed meat and meat, poultry and seafood and less on fruit, vegetables, juice, milk and yogurt and cereal and snacks. Obese shoppers allocate most of their additional spending to processed meals, processed meat and sweets. To compensate, they allocate less to fruits and vegetables.

Shoppers also differ in how they respond to price changes. In general, obese shoppers are less responsive to price changes than overweight and normal shoppers. So fat taxes, at least on the things obese shoppers buy, are not going to change their behaviour as much as might be expected.

Get real

Against the theoretical models of the economists, it’s nice to come across news of a study that plans to look at how people will respond IRL to at least one thin subsidy. The study is called Community Supported Agriculture Cost-offset Intervention to Prevent Childhood Obesity and Strengthen Local Agricultural Economies, which despite being a giant mouthful says exactly what they plan to do.

Community supported agriculture (CSA) takes many forms, but essentially participants pay a farmer upfront for a season’s worth of vegetables and fruit. The farmers get a financial guarantee and some stability, and the community gets fruit and vegetables at significantly lower cost. For a variety of reasons – such as the upfront cost of a share, inconvenient pick-up points and not knowing what to do with some of the produce – poorer communities are less likely to take part in CSA schemes, even though they probably have most to gain.

Subsidised CSA schemes could offer several benefits: better quality diets for poor people, reduced obesity, especially for children, and support for thriving local agriculture. All of that is yet to come; the article describes the initial research to design the study and how exactly they are going to encourage poorer families to participate and make use of the produce they find in their CSA boxes. It is worth bookmarking now as a repository of evidence on the many linkages among CSAs, fruits and vegetables, economics and health.

How much does a nutritious diet cost? Depends what you mean by "nutritious"

Recently I’ve been involved in a couple of online discussions about the cost of a nutritious diet. The crucial issue is why poor people in rich countries seem to have such unhealthy diets. One argument is about the cost of food. Another is about everything other than cost: knowledge, equipment, time, conditions.

My own opinion is that given all those other things, the externalities, a nutritious diet is actually not that expensive. But that’s just an opinion, so I went looking for information, and found it in a paper entitled Using the Thrifty Food Plan to Assess the Cost of a Nutritious Diet, published in the Journal of Consumer Affairs in 2009. The very first sentence of that paper is:

How much does a nutritious diet cost?

Parke Wilde, author of that paper, is an agricultural economist at the Friedman School of Nutrition Science and Policy at Tufts University in Boston, and I really enjoyed talking to him for the podcast.

Our discussion was limited to the United States and to the peculiarities of SNAP, but it is clear that other rich countries are also grappling with the difficulties of deciding what constitutes a nutritious diet and how to ensure that poor people can afford to eat it. Parke alluded to some of the crucial policy decisions, such as not allowing people to spend their SNAP benefits on things like alcohol while at the same time giving them the freedom to buy sugar-sweetened sodas if they so choose. There are incentives to buy healthier food, but even those come up against an important fact.

Unhealthy calories are much cheaper than healthy calories.[1]

One study, which I found here, looked at exactly what you can buy for a dollar. If you’re looking for a snack, a dollar’s worth of cookies will buy you 1200 calories. The same dollar on carrots gets you only 250 calories. Thirsty? A dollar of orange juice is worth 170 calories, while a dollar of sugar-sweetened soda is 875 calories. Of course calories aren’t everything. But in the short term, if you’re hungry, calories are everything.

A different version of the same sort of story is the “healthy food is more expensive” trope, which crops up regularly. In the UK, a 2014 study claimed that on average 1000 calories of “healthy” food cost about 3 times more than 1000 calories of “unhealthy” food. This kind of study – and others like it – uses government agency definitions of healthy and unhealthy but doesn’t actually take nutrition into account. The cheapest category in that study was starchy carbohydrates – which the UK’s Food Standards Agency places just behind fruit and vegetables in terms of “healthiness”. You could eat a lot of “bread, rice, potatoes and pasta” for calories and a bit of fruit and vegetables for nutrients and end up with a pretty nutritious diet.

Would you want to, though?

That’s the crux of the matter. A nutritious diet can be had for reasonably little money (given that you know how to cook and have the facilities) but you might not enjoy it all that much.

Notes

  1. Using the Thrifty Food Plan to Assess the Cost of a Nutritious Diet
  2. If you really want to get some insight into the tricky triple balancing act of foods, nutrients and cash, Parke Wilde and his collaborators have made their model available for you to play with.
  3. Parke Wilde’s website. A really interesting recent piece of his looked at proposals to amend SNAP to restrict access to sugar-sweetened beverages.
  4. The USDA’s monthly Cost of Food reports are online
  5. The banner photograph is from Millie Copper who tried to buy a healthy diet based on the Thrifty Food Plan costs for a couple of weeks.
  6. Thanks to Christopher Gifford, a great Patreon, for inspiring this episode.
  7. Peel an onion, and stick five or six whole cloves in it. Put it in a saucepan with about 500 gm of brown or green lentils. Add a stock cube or a spoonful of miso paste if you have some, but if you haven’t, that’s OK. Cover with water and bring to a boil. Simmer for 30–45 minutes. [2] Serve with lots of bread and a nice green salad.

huffduffer icon   Huffduff it


  1. I’ve actually written at length about this before, but cannot for the life of me find it now.  ↩

  2. You can just boil for about 5 minutes and then swaddle the whole thing in cushions or blankets, or a ready made haybox if you have one.  ↩