Your points are well taken from the perspective of an individual considering their own diet. But what about, say, the head of a large public school cafeteria? What if said person-in-charge is busy dealing with other aspects of their job and would like to automatically generate menus? (Or at least auto-generate a few suggestions.) I would argue that in this scenario, every approach *is* optimization. In other words, any algorithm for generating candidate menus is operating according to some (possibly implicit) objective function. However, I don't mean to claim that this objective function is *right* or even that some "correct" objective exists. Unlike algorithms that we design, reality is vague, indeterminate, and not well defined!
But I live in Berkeley where the free school lunch menus were designed in collaboration with Alice Waters.
More seriously, though, I don't think it's useful to conceive this as optimization. I think we can attempt to understand decision making without optimization (see today's post!)
I'm looking forward to learning more about naturalistic pattern recognition and decision making! I see that if the goal is to *understand* human decisions, optimization may not be a useful framework. By the same token, optimization may not even be great for *imitating* human decisions.
But I'm thinking of usefulness in a different way: once I've decided to offload something to an algorithm, I claim that optimization is a useful lens for *understanding that algorithm* (perhaps even for designing it). Do you disagree with that?
No, I don't disagree. Because algorithms are always optimizing something.
But I also don't think people use "algorithms" in the sense of undergrad CS. My read of Naturalistic Decision Making is that humans don't use "algorithmic procedures" when they decide things. That is, no one has good ways of simulating Naturalistic Decision Making with computers.
Then it sounds like both of us agree with "Everything [that we make computers do] is optimization". The problem is that not everything is computation! One last question here: do you think that designing automated decision-making systems that make decisions more like people is 1) possible or 2) necessary for them to be useful?
It depends on what you mean by useful, but I don't think it's necessary. But our discourse would be better off if we stopped pretending like these two kinds of decision making are the same.
Of course, many would further claim that the optimal diet is not a static thing, but a rhythm. It has elements of periodization (seasonality, culture) in it. This would be good for the mind and the body. The body is very adaptive and being healthy requires exercising the adaptation mechanisms regularly (NNT has some paragraphs along those lines in Antifragile IIRC). One example would be the recurring zero diet, fasting, culturally imposed. Diet is (was) cultural. Bro science went to town on fasting recently, as expected. Maybe there is some stochastic LP to handle all of the above?
I'm going to stop writing about my self-experimentation at some point, but I flirted with intermittent fasting and all I learned was that everything I ate after dinner was garbage.
Hey Ben, I'm not sure the statement, "Everyone wants to find the best diet that keeps them healthy and thin" is true. For example, I'm lifting weights to get stronger, knowing that I am definitely not getting thinner. To stick with your framework, it might be difficult to optimized (your "best") for both healthy and thin. If people want to be thin, they may want to be as thin as possible without dipping below whatever their criteria for "healthy" is. In that case you'd want to minimize the calories, nutrients, etc., where "healthy" is another constraint defined by some kind of measurable metrics--not my area of expertise, but maybe things like resting heart rate, blood pressure, body composition.
There also seems to be empirical evidence that body builders are actually very good at this, optimizing both their bulking and cutting phases, which have different criteria for "best" (first, bigger/stronger, then thinner, more defined.) You don't get a body builder physique without incredible focus on your food intake.
Bringing spices into the equations may be absurd for body builders--these are guys drinking chicken breast shakes--but maybe not for "quality of life" optimizers.
I mistyped. I wanted to write "healthy and fit." I'm going to edit it now, but your point stands. Everyone has different goals in mind with their diet. "Healthy" and "fit" are also hard to quantify.
And I love that you bring up bodybuilders. They are indeed odd cases. The really obsessive ones will just eat chicken, broccoli, and rice for a whole prep. That's commitment. The energetic basics behind their diets do seem pretty simple: if you want to bulk, you add 500 calories to your diet. If you want to cut, you remove 500 calories. You eat a bit higher percentage of protein during the cut because you don't want to lose muscle mass (though the science here is sketchy as best).
It's also hard to use body builders as examples for the rest of us, because the pros are all on tons of drugs.
Your points are well taken from the perspective of an individual considering their own diet. But what about, say, the head of a large public school cafeteria? What if said person-in-charge is busy dealing with other aspects of their job and would like to automatically generate menus? (Or at least auto-generate a few suggestions.) I would argue that in this scenario, every approach *is* optimization. In other words, any algorithm for generating candidate menus is operating according to some (possibly implicit) objective function. However, I don't mean to claim that this objective function is *right* or even that some "correct" objective exists. Unlike algorithms that we design, reality is vague, indeterminate, and not well defined!
But I live in Berkeley where the free school lunch menus were designed in collaboration with Alice Waters.
More seriously, though, I don't think it's useful to conceive this as optimization. I think we can attempt to understand decision making without optimization (see today's post!)
I'm looking forward to learning more about naturalistic pattern recognition and decision making! I see that if the goal is to *understand* human decisions, optimization may not be a useful framework. By the same token, optimization may not even be great for *imitating* human decisions.
But I'm thinking of usefulness in a different way: once I've decided to offload something to an algorithm, I claim that optimization is a useful lens for *understanding that algorithm* (perhaps even for designing it). Do you disagree with that?
No, I don't disagree. Because algorithms are always optimizing something.
But I also don't think people use "algorithms" in the sense of undergrad CS. My read of Naturalistic Decision Making is that humans don't use "algorithmic procedures" when they decide things. That is, no one has good ways of simulating Naturalistic Decision Making with computers.
Then it sounds like both of us agree with "Everything [that we make computers do] is optimization". The problem is that not everything is computation! One last question here: do you think that designing automated decision-making systems that make decisions more like people is 1) possible or 2) necessary for them to be useful?
It depends on what you mean by useful, but I don't think it's necessary. But our discourse would be better off if we stopped pretending like these two kinds of decision making are the same.
Of course, many would further claim that the optimal diet is not a static thing, but a rhythm. It has elements of periodization (seasonality, culture) in it. This would be good for the mind and the body. The body is very adaptive and being healthy requires exercising the adaptation mechanisms regularly (NNT has some paragraphs along those lines in Antifragile IIRC). One example would be the recurring zero diet, fasting, culturally imposed. Diet is (was) cultural. Bro science went to town on fasting recently, as expected. Maybe there is some stochastic LP to handle all of the above?
I'm going to stop writing about my self-experimentation at some point, but I flirted with intermittent fasting and all I learned was that everything I ate after dinner was garbage.
Hey Ben, I'm not sure the statement, "Everyone wants to find the best diet that keeps them healthy and thin" is true. For example, I'm lifting weights to get stronger, knowing that I am definitely not getting thinner. To stick with your framework, it might be difficult to optimized (your "best") for both healthy and thin. If people want to be thin, they may want to be as thin as possible without dipping below whatever their criteria for "healthy" is. In that case you'd want to minimize the calories, nutrients, etc., where "healthy" is another constraint defined by some kind of measurable metrics--not my area of expertise, but maybe things like resting heart rate, blood pressure, body composition.
There also seems to be empirical evidence that body builders are actually very good at this, optimizing both their bulking and cutting phases, which have different criteria for "best" (first, bigger/stronger, then thinner, more defined.) You don't get a body builder physique without incredible focus on your food intake.
Bringing spices into the equations may be absurd for body builders--these are guys drinking chicken breast shakes--but maybe not for "quality of life" optimizers.
Interesting approach.
Hi Joshua,
I mistyped. I wanted to write "healthy and fit." I'm going to edit it now, but your point stands. Everyone has different goals in mind with their diet. "Healthy" and "fit" are also hard to quantify.
And I love that you bring up bodybuilders. They are indeed odd cases. The really obsessive ones will just eat chicken, broccoli, and rice for a whole prep. That's commitment. The energetic basics behind their diets do seem pretty simple: if you want to bulk, you add 500 calories to your diet. If you want to cut, you remove 500 calories. You eat a bit higher percentage of protein during the cut because you don't want to lose muscle mass (though the science here is sketchy as best).
It's also hard to use body builders as examples for the rest of us, because the pros are all on tons of drugs.
Ah, yes, forgot about the drugs!