“All natural food” as an constrained optimisation problem
A look at all natural foods through the lenses of Bayesianism, optimisation, and friendly utility functions.
How should we consider foods that claim to be “all natural”? Or, since that claim is a cheap signal, foods that have few ingredients, all of them easy to recognise and all “natural”? Or “GM free”?
From the logical point of view, the case is clear: valuing these foods is nothing more that the appeal to nature fallacy. Natural products include many pernicious things (such as tobacco, hemlock, belladonna, countless parasites, etc...). And the difference between natural and not-natural isn’t obvious: synthetic vitamin C is identical to the “natural” molecule, and gene modifications are just advanced forms of selective breeding.
But we’re not just logicians, we’re Bayesians. So let’s make a few prior assumptions:
There are far more possible products in the universe that are bad to eat than are good.
Products that humans have been consuming for generations are much more likely to be good to eat that than a random product.
Now let’s see the food industry as optimising along a few axis:
Cost. This should be low.
Immediate consumer satisfaction (including taste, appearance, texture, general well-being for a week or so). This should be high.
Long term damage to the consumer’s health. This should be low.
The last point is the weakest one, though. Long term health impacts are fiendishly hard to measure (by anyone), and consumers’ memories are short. The main things that the food industry wants to avoid are blaring headlines like “salmon saturated with mercury”, “sugar linked to cancer”, or “food industry coverup”. Even there, journalists can create scandals out of very little (or ignore major ones if something more juicy comes along), so the pressure to actually prioritise long term consumer health is very weak. So most optimisation pressure is along the first two axis.
So the food industry will strongly push to decrease cost and increase satisfaction, while mostly taking a random walk on long term consumer health. Given the assumptions above, this means that we’d expect the long term health impacts to worsen (because there are far more negative products than good one). This somewhat similar to the importance of programming everything of value into a friendly AI, lest the things not programmed get squeezed out.
Now, what does “all natural” mean in this setting? It means that the natural food industry is facing a massively constrained optimisation problem. They are extremely limited in what they can do (compared with the rest of the industry), and it mostly involves shuffling around with products we suspect to be benign or positive. Similarly with GM modifications: selective breeding is much slower and uncertain, so the optimisation pressure is less: they literally can’t change things as much or as fast. One supporting argument for this is that all natural or GM-free products tend to be more expensive or less satisfying than others, demonstrating less optimisation pressure.
This is entirely not a result I expected to find. Pushing it to the extreme, it would seem that the most traditional and unchanged food (after removing stuff we know to be bad) is likely to be best, as long as people don’t get too inventive with them.
Either that, or it’s just plain old market segmentation and price discrimination which extracts more money from the more wealthy people who would buy such foods.
Slight correction. It’s not more wealthy people, it’s people willing to pay more, although they tend to be more wealthy as well.
Fair trade can also cause problems for the farmers because of price discrimination. If the manufacturer wants to price-discriminate, they only want to produce a limited amount of fair trade product. So the farmers produce fair trade raw material, but they can’t sell all of it as fair trade, because the manufacturers only certify a limited quantity of it. The farmers who paid the increased cost of making fair trade material, but could not sell it as fair trade, lose out.
BTW, “fair trade” products at my supermarket (not exactly an upscale kind of place) tend to be as cheap as other products of similar quality, or even cheaper.
That depends on the magnitude of the impact. See e.g. mercury or radioactive elements (there were popular health tonics with radium in the early-XX century US).
Not quite, there are clear thresholds effects here. The prominence of personal-injury litigation (in the US) also plays a role. Look at what happened to the asbestos industry.
Of course, but so does the “regular” food industry as well. Adding random stuff to food isn’t really how it works.
One can also make the argument that the “natural food” industry should just deliver the natural food from the fields to the table and not mess around with it in the process.
What does “should” mean in that sentence? That the government agencies that regulate what qualifies as natural food should forbid the “natural food” industry from doing any optimisation?
It means that I would prefer this to be the case and may roll my eyes at the “optimization”. Tremble at the power of my eyeroll! :-P
It has taken us decades to learn how to measure the damage for those impacts. Just look a smoking and lung cancer. The tobacco industry spent many years denying the effect and nobody in medicine got a nobel prize for the smoking and lung cancer link.
Who would they have given it to? When I ask myself for ‘the man responsible for showing smoking & lung cancer are linked’, nothing comes to mind, but I do remember that the claim had a long history going back to the Nazis among others and was the result of a long succession of correlational studies and animal experiments. Who in particular, who was still living when the connection became undeniable (no post-humous awards), deserves the Nobel for all that?
That fact that it took decades to prove the effect is my point. If it’s really hard to show negative health with the kind of impacts that smoking has, we have to conclude that we are bad at detecting negative health effects.
To be honest, my reaction to that claim is that we’re not bad at detecting negative health effects, we’re bad at ethics: it wouldn’t be that hard to detect the negative health effect experimentally (what would it take, maybe a thousand subjects and a decade?), it’s just gosh, that would be immoral—much better to stick to moral observational studies and let millions of people keep dying for decades to come!
Most humans don’t live in an environment that’s easy to control experimentally. Getting compliance, or adherence with dietary interventions isn’t easy.
What kind of setup would you suggest?
Recruit non-smokers. Randomize, then, large payments to the recruited smoker/non-smoker setups with provision of free cigarettes to the former group, with detection of compliance in both groups through cotinine measurements of hair. You’ll get some attrition between groups, but I don’t think a whole lot, and the effect is large enough that a bit of bias won’t defeat the experiment.
That’s where the addictiveness comes into play!
Really, detecting smoking’s problems is easy. There’s easy measures of compliance, the necessary experiment is small, the intervention is self-reinforcing...
It’s easy if you don’t have evil ‘ethics’ and can shut up and calculate and actually do the experiment.
If you’re going to shut up and calculate, you need to calculate not only the benefit and harm from the experiment, you need to calculate the benefit and harm from weakening the ethics code. You can’t weaken ethics just for one experiment; if you weaken it for experiments where the benefit outweighs the harm, you’ll also weaken it for experiments where the harm outweighs the benefit but the people performing the experiment value harm to some people more than to others, and for experiments where the experimenters just don’t know how to calculate, and for experimenters where an unspoken goal of the experiment is to actually cause harm, and....
It’s like asking if we should allow courts to use evidence gained from an illegal search. If you “shut up and calculate” the costs of freeing a criminal versus finding a criminal guilty, you’ll determine that it’s always good to use the evidence. But if you allow the use of evidence gained from illegal searches, you’ll incentivize illegal searches, and that incentivization must be included in any calculation.
I could even phrase this in the language of precommitting: you precommit to always following the ethics code because following an ethics code is, on the average, advantageous but you don’t get the advantages unless you’re the type of person who’s willing to follow it even in disadvantageous situations.
Smoking is a situation where any remotely accurate model of the situation says that the value of information is extremely high inasmuch as smoking was one of the most popular activities in the world at the time, was expected to continue be so, the correlates (if causal) translated to enormous loss of life, and since it took decades to do so, the correlative evidence was obviously too weak & unpersuasive to motivate people to quit or overcome societal presumptions/precommitments against heavy regulation & bans.
Against this obvious point, you set—as even more desirable—preserving a bunch of made-up incoherent* rules or ‘ethics’ put into place post-WWII as a response to the Nazi concentration camp tortures and things like the Tuskegee syphilis experiment, none of which activities would have passed previously-accepted norms of conduct or even scientific reporting (eg. were any of Mengeles’s ‘experiments’ done adequately and could reach any level of scientific validity? the only halfway useful ones I know of were the altitude experiments), and were done, surprise surprise, under conditions of secrecy and the auspices of totalitarian dictatorships and the closest thing to that in the West, militaries. But you know what, having IRBs and ‘ethics’ didn’t stop stuff like MKULTRA post-WWII, did it? That’s because secrecy is the father of abuse, not any ‘lack of research ethics’. You’re solving the wrong problem. The problem is not that scientists in their ordinary course of conduct investigating things such as smoking are monsters. The problem is that governments and militaries and corporations and elites are monsters who, if you let them, will happily do monstrous things and attract and enable and cover up monsters like Mengele and try to ignore things like ‘downwinders’.
* I find ‘informed consent’ particularly hilarious, having participated in a few experiments at college. At no point was I ever ‘informed’ in any meaningful way about risks with, for example, a specific probability like ‘based on a meta-analysis of previous experiments, we think there’s a <5% chance you’ll have a rash or worse’. No wonder the bioethics journals continue debating how many angels can dance on the head of a pin—I mean, how informed consent applies to Third World clinical trials, old people, reuse of data for additional analyses, etc etc.
So yes, I am perfectly happy to bite your bullet and advocate for undermining your ‘ethics’ and things like IRBs. They are worthless. They do nothing but impede science, and trade off endless sins of omission in exchange for (possibly) preventing some sins of commission. This is nothing like courts, and precommitments do nothing here but harm.
The willingness to settle for trash correlations massively harms science and society (one word, which should silence all doubters of this: “diet”), and to the extent that a smoking experiment weakens this norm and helps people consider the long run & inherent uncertainty of even the best correlative evidence, I regard it as an unalloyed good above and beyond the issue of tobacco.
I hope that one day, the question people ask will never be, ‘could it possibly be moral to run a randomized experiment on X?’, but rather, ‘could it possibly be moral to not run a randomized experiment on X?’ (or better yet, they don’t ask the question at all since it is taken for granted that, not being superstitious barbarians, of course a randomized experiment has been run).
Is it not allowed to run randomized controlled trials assigning diets to people? I’m pretty sure I’ve read about such trials. Do ethics boards forbid assigning diets they (without good evidence) believe to be harmful?
Like any area, the ethics boards hold experiments to much higher standards than things like surveys; I don’t think it’s exceptional in this regard except to the extent that the area has irresponsibly taken its dubious results as gospel and tried to remake society. (I criticize a lot of psychology for bad research practices, but at least with most of it, people don’t try to reorganize their lives and diets based on the latest survey.)
On top of that, diet research focuses heavily on junk correlations because it’s unusually hard to run RCTs on diet. Unfortunately, they ignore that correlations are far less informative compared to causations than correlation research is easy to run compared to RCTs. We’d be better off if most diet research had never been done, ethics ignored, and the funding used for a few large RCTs instead. That’d’ve avoided the farcical history of diet advice like salt, fat, etc.
That’s only an implicit answer, and I want to be sure I understand correctly. Do ethics boards forbid trials with diet interventions? Or is the problem only that diet researchers do the wrong things and then oversell their results?
They and the general culture of ‘ethics’ and overrating professional expertise and correlative results forbid trials on the margin.
I don’t see any ‘only’ about the matter.
(Retracted—Note to self: Read the grandparent before commenting)
And I’ll repeat something I don’t think I emphasized enough: you can’t weaken ethics just for one experiment. Weakening ethics codes for experiments whose benefits outweigh the harm to others also weakens ethics for experiments of other types. That’s how human beings behave in the real world. It’s no use pointing to an abusive experiment and saying “that would have been banned anyway, even without your code”. First of all, it obviously was not banned already; Tuskegee did pass enough previously accepted norms of conduct for it to actually happen. Second, modern ethics codes make a much better Schelling point; if you instead say that it’s okay to hurt people or encourage people to hurt themselves but it’s not okay to do so in bad cases, it’s much easier to rationalize away a bad experiment than if you say “no, period”. It’s easy to say in hindsight “it was banned by an existing ethical code anyway”, but that’s not what they thought at the time. See http://lesswrong.com/lw/ase/schelling_fences_on_slippery_slopes/
Illegal searches are not a generic statement about “courts”.
We don’t use evidence from illegal searches because although using the evidence would cause more benefit than harm in the particular case we want to use it in, using such evidence has the effect of encouraging illegal searches, not all of which are going to be in similar cases.
Likewise, you follow the ethics code because although breaking the code would cause more benefit than harm in the particular case you want to break it in, breaking the code has the effect of encouraging more breaking of the code, not all of which will be in smiliar cases.
Precommitments benefit you here. You precommit to following the ethical code even if it is harmful in the specific case you care about, because being a person willing to follow such a code has the advantage of encouraging other people to follow the same code, and most of that will be beneficial, while being a person who is willing to make exceptions to the code won’t do that.
I have never argued that standards should only be weakened for one experiment. My argument here is for a whole-sale universal shift to a different standard.
Tuskegee was a secret, as I’ve already said, so how did it pass norms of conduct? Secret scandals do not show anything about accepted norms of conduct, or rather, they show the opposite of what you want it to show: that they couldn’t get away with it publicly and had to keep it a secret. No one went to the newspapers at the start and said ‘we’re going to kill a bunch of blacks with syphilis’ and the newspapers printed that and everyone was ‘well ok that’s within accepted norms of conduct’ and then later changed their minds.
Let’s say that complicated arbitrary systems of ‘consent’ and ‘benficience’ which cannot be defined clearly and lead predictably to many kinds of deeply suboptimal outcomes are, in fact, a Schelling point. So? A Schelling point is not a magic wand which justifies every sort of status quo; why should one think that the violations halted by the existence of a Schelling point outweigh instances like smoking where the enforcement of medical ethics leads, by the most conservative estimates, to millions of excess deaths?
And which encourage overbearing tyrannies which themselves cause massive death and disutility. Here’s an example of where the slippery slope bottoms out at something bad. But what is there for principles like randomized trials as the default?
What are these huge, well-established, overriding threats? Who is the Stalin or Mao of randomized trials?
Bayesian, not logician. More freedom means more optimisation power.
There are some things with clear health benefits (eg preservatives to make sure the food doesn’t rot, pasteurisation) but I agree with your point, if health was the only consideration.
It is not obvious that the expected negative utility of e.g. GM-food due to unexpected bad effects on health is higher than the expected positivity utility from e.g. lower cost. Note that GM-food is not a random “product in the universe” but a slight modification of something known to be ok + non-vanishing ability to predict the consequences of this modification. Moreover, it might be possible to use the theory we already have to optimize towards improved long-term health and it might be such modification will more than cancel out the expected negative utility from unpredicted effects.
Who’s “we”? As Stuart describes, the incentives for the industry to do this aren’t really there. They have much more incentives to manipulate crops to produce their own pesticides.
Maybe, I was just pointing out the problem is not inherent in GM-food. A similar issue exists for “all natural” food: the industry might still optimize towards unhealthy cheap food. The real source of the problem is on the side of demand. In a perfect world, there would be a reliable source of information about health and people would pay attention to it.
As Stuart argues, they have less means to optimize in that way.
Right. Not enough customers are running scientific studies to test how mortality rates change when the digest different food.
We don’t live in a perfect world and have to make choices based on the world in which we are living.
This is a good point.
Not sure I understand what you’re saying. Are you being ironic?
Of course. But we weren’t discussing a specific choice, as far as I can tell.
Yes.
I do think we are discussing the choice to buy “natural food” and prefer it over genetically modified food.
In this case, an in-depth discussion would require analyzing the potential health hazards given what we know about the genetic modifications involved and the regulations in place, versus the cost difference and some way to compare the two.
And we should probably start by defining “genetically modified”.
Under some definitions almost all commercial crops and farm animals are genetically modified. There are no wild cows and Golden Delicious apples don’t grow on trees in forests.
EU laws have well defined definitions.
I came across two definitions:
“Genetically modified organisms (GMOs) are organisms, such as plants and animals, whose genetic characteristics are being modified artificially in order to give them a new property.”
That’s extremely broad and vague (and incorrectly uses present progressive rather than past tense)
“organisms in which the genetic material (DNA) has been altered in a way that does not occur naturally by mating and/or natural recombination”.
A bit more precise, but still not well-defined. Is artificial insemination GM? What about irradiation to increase mutation rate? And technically, the DNA of the organism isn’t altered; it has the same genes it was born with.
Definitions are always a tricky business. As far as I understand the current EU definition of genetically modified allows irraditation to increase mutation rate. I don’t know whether the organic food regulations do.
Recently I posted something about the EU commision wanting to say that coloring eggs red or blue makes them not fall under the organic standard. It’s not possible to read a single paragraph about the definition to really tell you what falls under it but you have bureaucrats who do mind all the little details.
Which is kind of the point. How can we have a meaningful discussion about something, when it takes pages to define it? And if the definition is so complicated, then it really isn’t legitimate to ascribe companies’ aversion to complying as simply not wanting to inform customers.
By actually thinking about the issue? Just look at how many pages Eliezer needed to explain his idea of “truth”.
That’s not what I do. Companies lobbies on multiple fronts at not wanting to inform customers.
I don’t think that companies would have an issue to explain on a web page in a few pages what their production process entails. I can’t even easily find out whether or not beef I buy in my supermarket comes from grass-fed cows or whether it doesn’t.
Companies do have lawyers that can read a bunch of pages and then apply the correct label.
I practice we often don’t know much about the genetic modifications involved. If you buy some foot item in the US as a customer that contains some genetic manipulations there no way for you to know exactly what they modified.
Even if you would know which genes they exchanged, we are at a point where we don’t even know what all the genes in the human body do.
Ps: and even if true, there’s an un priced externality there...
I don’t understand why it’s an externality given that nothing is stopping the supplier from providing information about the food.
You’re right, “externalities” is incorrect. It’s more about information and ignorance. The point being that there are hidden factors meaning we can’t just compare demand, supply and prices without adjusting for it.
Then why do the suppliers are intent on lobbying that they have to provide the least amount of information that’s possible?
Supplying useless information still has a cost just because supplying any sort of information is not free.
Some consumers are irrational and supplying some information plays into the biases of the consumers
Some information the government may want manufacturers to supply for political reasons unrelated to the safety of the product. Imagine a law which required that manufacturers state whether the product is made using labor from illegal Mexican immigrants. This can overlap with #3.
The manufacturer may fear that consumers will interpret required labelling in the context of non-required labelling—for instance, if consumers are used to safety warnings, they may be more likely to interpret some kinds of required labels as safety warnings.
It’s a bad thing, but still not an externality.
Agreed. But I was surprised that “all natural” would have any health benefits at all.
Define “massively”?
To the best of my (admittedly relatively ignorant) knowledge:
The worst generally acknowledged culprits for dietary promotion of disease are sugars and trans fats. Sugars are easy to add in arbitrary quantities in “all natural” form.
The worst remaining likely culprits for dietary promotion of disease are salt, nitrates/nitrites, and excesses of many other fats (likely everything except omega-3 and monounsaturated fatty acids in the average modern diet?). These are all easy to add via “all natural” ingredients.
The worst still-controversial suspected culprits for dietary promotion of disease are high-carb diets and high dietary cholesterol; both of these are easy to add via “all natural” sources.
We obviously don’t know nearly as much as we should about nutrition as we should. The big Mediterranean diet study showing 50+% reduction in cardiovascular incidents was only a decade ago, and the study narrowing most of that down to nuts and olive oil was just a year ago! But it seems that we know enough to say that there are ways to make food much more immediately satisfying and much less healthy without using any ingredients that wouldn’t have been available throughout human history; all we have to do is use too much of some of them. All-natural diets scored a big victory in the case of trans fats, and may avoid similar artificial food mistakes in the future too, but the worst natural ingredients seem to be much worse for you than the best artificial ingredients, so simply avoiding the latter isn’t a panacea.
I’m surprised by the lack of research on organic foods and health, and it seems like it wouldn’t be too hard for a talented researcher to compare the health and mortality of people who consume organic vs. inorganic diets, after controlling for differences between the two groups, such total nutrient consumption, exercise, premorbid conditions prior to organic consumption, etc. Modified food may or may not have adverse effects beyond different nutrient contents (which so far is debatable), but I’m surprised at the amount of people who have jumped on this bandwagon with scant supporting evidence.
There is also the possibility that people will eat worse when consuming organic. I suspect that an inorganic diet composed of fish, fruits and vegetables, legumes, lean dairy, and nuts will be far healthier than an organic diet composed of fried chips, fatty artisan cheeses, chocolate bars, and low fiber carbs. Go to Trader Joe’s or Whole Foods and watch how many carts are filled with the things you shouldn’t eat. In fact, it seems the all-natural industry follows #1 (as far as they can) and #2 quite well, and if organic retailers are a proxy, they are about as good at ignoring #3 as the rest of the industry.
Controlling doesn’t get rid of all the confounders (easiest one: people who eat organic care more about what they eat, almost by definition—how do you control for that?), and long term studies are very hard to do.
A place to start is to feed two groups of animals foods, one eating organic and the other eating inorganic, with identical or near-identical nutrient compositions, and see how they respond over time. Linking dietary effects between animal and human models has been done in the past, so it isn’t too far-fetched. It won’t be perfect, since the animals won’t be humans, but it is certainly better than the paucity of data available, and assuming that organic = good with scarce evidence (see below).
http://ajcn.nutrition.org/content/92/1/203.short
Here is one meta-study. Here is another one
Are we trying to find out if organic foods are more nutritious, or if organic foods offer health benefits beyond nutrition? (or to reverse that, do inorganic foods offer adverse effects beyond nutrition) Remember I said , ” Modified food may or may not have adverse effects beyond different nutrient contents (which so far is debatable),” The authors conclude in your 2nd link that they agree the evidence on the benefits of organic foods is scant at the moment.
“Organic” and “non-modified” are very different things.
“Organic” means that the food producer has received a particular kind of certification for his production. By the way, in this context the opposite of “organic” is “conventional”, not “inorganic”.
“Non-modified” has a less well-defined meaning, but generally it means food as it comes from the farm, not from a factory.
There is lots of “organic modified” and “conventional non-modified” food.
Misnomer noted. So, is there evidence that conventional foods (or foods that are not organic) have adverse effects beyond possible nutritional differences, when compared to organic foods, and genetically modified vs. not modified? (and by not modified I mean not genetically modified, if the context preceding the words didn’t make those words crystal clear) I am of course open to the possibility, but I would like to see harder evidence before paying a premium.
Not to my knowledge.
In reality there no reason to assume that the health difference between organic and conventional bananas is the same as the health difference between organic and conventional butter.
I think modeling this in supply-side terms is the wrong way to go. There are strong purity heuristics floating around in what marketing flaks somewhat unimaginatively call the LOHAS market segment. It seems more parsimonious to assume that “all natural” and related terms are attempts to meet those heuristics than that they’re an end product of some kind of long-term optimization process.
Particularly since a lot of marketing terms in that general space are meaningless or nearly so. If you want to buy eggs at Whole Foods, for example, you’re likely to be confronted with a bewildering array of marketing claims; if you actually do the research, though, it turns out that most of the phrases in question have no well-defined meaning, and some (e.g. “vegetarian feed”) are likely to indicate less “natural” conditions than the alternative.
Yes, and why do you think we evolved purity heuristics about food?
In general, foodborne disease. In this case, though? Mostly the appeal to nature, plus a dash of cultural romanticism.
That’s why I focused on the harder-to-fake signal of having only recognisable ingredients, or no GMOs. Other strange restrictions (eg Kosher) might also work.
This has a low false negative rate, which is great from a health perspective, but be aware that you will get a lot of false positives using the ‘recognizable ingredients’ heuristic. Technical names can be fairly intimidating, even for ingredients that you’ll find in a home garden. If you notice the same chemical appearing in many places (i.e. Tryptophan), then it might be worth looking it up with a focus in natural occurrences; a mental bank of ‘safe’ technical-sounding ingredients might increase your dietary range pretty substantially without increasing health risks much at all.
Yep!
In that case, why are they using the technical name and not the name of the garden plant?
Because the plant itself is not an ingredient, in most cases. But many of the artificially synthesized compounds are identical at the molecular level to those found in garden plants.
This joke is an informative way of putting it.
Yes, but that means they’re not including any of the other things that would be in the plants. Also just compounds in plants tend to be embedded in complex structures, that the artificially synthesized compounds aren’t.
Ok, so if I were to give you, say all the ingredients listed for an egg, could you make one for me?
Almost certainly not. Could you explain why you think this is a meaningful criterion for health?
I don’t know of any evidence supporting the idea that a chemical ingredient becomes more or less healthy if consumed in the presence of other chemical ingredients (except in terms of long-term nutrient deficiencies). Similarly with the arrangement of these ingredients into any larger structure. Sugars can be simple or complex, and these do have dramatic differences in health outcomes, but this is a difference at the molecular level and has little to do with whether they are embedded in a particular pattern. The human digestive system works by breaking down these arrangements, starting with basic chewing, and so I would be surprised to see structural associations as having much in the way of consequence.
(One possible exception in the form of dietary fiber, but as the indigestible portion of our food, it’s a bit of a special case.)
Have you looked at these issues at all? It is fairly common that the presence of one substance makes another more bio-available. Pellagra being the most famous example.
.