I think I understood what you’re saying the first time around and again I agree that your account of things is certainly possible, but even in your case the tumor or fat mass has to be sustained by calorie consumption from the outside. The increase in calorie intake we’ve seen is about the amount we would have expected if someone had just told us that people are 15% fatter on average without any increase in exertion or heat dissipation to compensate for that.
Another way to say this is that I’m just making a claim about the conditional expectation E[average body mass increase | average calorie consumption increase = 20%] and pointing out this expectation is not far off from the actual observed body mass increase we see. In that sense the increase in calorie intake can account for the increase in body mass in correlational terms. The question of why both of these variables went up, though, is difficult and their level is obviously not determined by this argument.
Externally, why people feel the need to keep eating until they become obese is not a question with a clear (to me) answer. It could be because there is some process that’s operating in the body that’s taking priority over other activities and hoarding a lot of the energy intake to produce fat cells, which I think is your story. However, it could also be that some part of the brain is malfunctioning and leading the lipostat to be poorly calibrated. It could also be that the food you’re eating messes with the natural feedback loops in your body that are supposed to make you feel full when you’ve eaten enough.
I think this question is interesting and your account is logically possible. I just think that
I don’t have any special insight into which account is right, and
I don’t think your particular explanation is favored that much over other competing explanations.
I don’t think the overfeeding experiments provide strong evidence for your scenario, though I agree that they should be a Bayesian update in favor of accounts in that broad neighborhood. What would convince me is experiments which involve smaller increases in calorie intake but sustained over much longer periods of time, on the order of a year or so. If such experiments failed to find an effect that would be a strong update for me towards your view. Right now I buy the cancer analogy on conceptual grounds but I don’t think we have enough evidence to conclude obesity is like cancer in this regard, though it very well could be the case.
To give an analogy of my own, the overfeeding studies look similar to attempts to settle disputes about which programming languages are best for productivity by asking undergraduates to complete some simple tasks in them over the course of a few weeks. What really matters is how you do when you’re working with big and complex software in the real world that has to be developed and maintained by large teams with turnover for years, sometimes decades; but that obviously doesn’t lend itself to a simple experimental design so people still keep arguing about it.
“Pregnancy” probably isn’t a thing. “Pregnant” people eat around 500 more calories per day. This is sufficient to explain all the weight gain from “pregnancy” without supposing anything other than thermodynamics at work—anyone who eats an extra 500 calories per day will probably gain that much weight over the course of 40 weeks.
I think Ege’s alleging that SMTM presented two causal graphs:
calories → ? → weight gain
calories → weight gain
Ege’s saying that 2 is simpler and sufficient, so we don’t need to posit a ? in the middle.
You’re pointing out that we still need to address a third causal graph:
? → calories → weight gain
Edit: And maybe that there’s also a scenario where ?, calorie intake, and weight gain are all in some complex interrelationship. Maybe contaminants cause more fat deposition and less energy and more hunger, thereby increasing weight gain per calorie, increased calorie intake, and increased contaminant intake via food. Or something.
Ege’s agreeing with you, but wants to emphasize that this is compatible with criticism of SMTM’s alleged emphasis on graph 1.
Note: I say “alleged” only because I’m sidestepping evaluating the truth of Ege’s claim. Just trying to clarify what it is (AFAICT).
I don’t agree with this presentation of what I’m saying.
I’m not terribly sure what SMTM means when they say “the increase in calorie intake is small”, but all possible interpretations of their claim seem wrong. For instance, one plausible interpretation in causal graph lingo is “if you applied the do operator on calorie intake and raised it by 20%, we would have seen an increase in body mass that’s significantly smaller than what we’ve actually seen”. I think this claim is wrong, basically for long-run energy balance reasons.
I’m not saying anything else about the structure of the causal graphs, which could be arbitrarily complicated and involve arbitrarily many nodes and dependencies. I’m just saying that if you apply the do operator on calorie intake and raise it by 20% then you’d get an increase in mean body mass that’s about as big as what we’ve seen.
Thanks for clarifying that I misrepresented your view. Based on your response here, you’re pointing out that there’s a strong correlation between increased caloric intake at the population level, and increased obesity. You are also saying that the explanations you’ve read from SMTM for why this correlation exists seem wrong, and also that they underestimate the magnitude or importance of the caloric intake.
WRT Eliezer’s arguments, you seem to be agreeing that there may be some underlying force(s) causing that increased caloric intake. However, you are very uncertain about which, if any, of the hypothesized forces(s) are the true causes of increased caloric intake.
Eliezer and others seem to be perhaps mistakenly interpreting you as denying the existence of, or “need for,” a deeper explanation for increased caloric intake and consequent weight gain. You are confused about why they are making this mistake.
I’m not sure who is “to blame” for the miscommunication but I suspect I simply was not clear enough in my top comment. Now it’s likely too late to clear up the issue for most readers as they won’t be following the developments in this thread.
I’m just saying that if you apply the do operator on calorie intake and raise it by 20% then you’d get an increase in mean body mass that’s about as big as what we’ve seen.
This is “assuming there’s no link “increased calorie intake → increased energy expenditure”″, right? I think one of the things Eliezer is saying is that there seems to have been such a link in the past and now there isn’t / it’s much weaker.
That’s not quite true—there is at least the naive link that a higher equilibrium body mass leads you to expend more energy in daily activities even if you exercise the same amount as before. In my very naive model I assume these are directly proportional, but Natalia cites some better research that does a log-linear regression of calorie expenditure on equilibrium (I think? I didn’t check this part) body mass which seems to be more accurate empirically.
I think it’s unclear whether we had the link you mention in the past, too. We definitely had a correlational link: people who did hard labor and ended up exercising a lot every day took in much more calories, as we would expect, and they were generally not obese. However, I think my argument would work just as well in the past if you just applied the do operator on calorie intake per day and looked at the causal impact on equilibrium body mass, as I don’t think there’s evidence that there’s a big downstream link from calorie intake per day to exercise.
I don’t understand why you’re “retrying”. I already agree with your point and you not saying “yes, you already agree with me” is quite confusing to me. As I say in my comment:
Externally, why people feel the need to keep eating until they become obese is not a question with a clear (to me) answer. It could be because there is some process that’s operating in the body that’s taking priority over other activities and hoarding a lot of the energy intake to produce fat cells, which I think is your story.
Do you think this characterization of your position is unfair or wrong? If so, why?
As far as I can see the only object-level point I disagree with you about is that I don’t think the evidence for obesity being like cancer or pregnancy is as strong as you seem to think it is. It’s definitely possible for it to be like that but I would bet against it at even odds. I explain this here:
I don’t think the overfeeding experiments provide strong evidence for your scenario, though I agree that they should be a Bayesian update in favor of accounts in that broad neighborhood. What would convince me is experiments which involve smaller increases in calorie intake but sustained over much longer periods of time, on the order of a year or so. If such experiments failed to find an effect that would be a strong update for me towards your view. Right now I buy the cancer analogy on conceptual grounds but I don’t think we have enough evidence to conclude obesity is like cancer in this regard, though it very well could be the case.
On top of that I also have a separate disagreement with you about emphasis in the context of my comment, since the point of my top comment is to draw attention to 400 kcal/day not being a small increase in calorie intake. You agree with me about this but you just don’t think it’s worth focusing on, probably because you think it’s a trivial observation. I still think it’s something that should be corrected given that SMTM explicitly said that it’s a small increase.
As far as I can see the only object-level point I disagree with you about is that I don’t think the evidence for obesity being like cancer or pregnancy is as strong as you seem to think it is.
Some people certainly are obese because of literal cancer and literal pregnancy. We seem to have strong evidence for that.
The interesting question is about how much of the obesity pandemic is explainable by such factors and not whether evidence for such factors exists.
We certainly don’t see enough pregnancy and cancer to explain the obesity epidemic but there might be other factors that are similar but harder to see. Thermodynamic arguments don’t help us rule out other effects that are similar to pregnancy/cancer.
I agree with everything you said, so again I’m confused why you thought you should make this comment.
I feel like I don’t really disagree with most of the commenters but they either think I do disagree with them or that I did a very bad job of communicating exactly what my point was. It’s hard for me to understand.
I still stand by this claim, again with the caveat that you take it as a correlational use of the word “explain” (which is not at all uncommon e.g. when talking about “fraction of explained variance” and so forth) and not one that suggests a causal explanation of the form “people wanted to eat more food, so they ate more food, so they got fatter as a result”.
Ok. My main point is just to clarify that other people are reading you as talking about explanation in general, not just strictly correlational explanation (if that’s what’s happening).
I do also think that’s not a great use of the word “explain” and “mystery”, because it’s not why the colloquial word is useful. The colloquial words “explain”/”mystery” are useful because they index “more information and ideas given/needed about this”. So just because X correlationally explains Y, and X is true, doesn’t mean there’s no mystery about Y.
I never said there’s no mystery about Y, just that there’s no mystery about Y being true conditional on X being true.
It’s a fair point that my usage of “explain” and “mystery” confused some people but I’m not too sure how else I would have made my point. Should I have said “people today are eating about as much more compared to the past as we would expect given how much fatter they’ve gotten”?
Someone may have made this point somewhere else already, but there’s really no mystery in Americans getting fatter if we condition on the trajectory of mean calorie intake.
Read literally, this says: There’s a mystery of why intake went up. Conditioned on intake up, there’s no mystery of why fat went up. I think this isn’t right. If we agreed that sustained high intake implies weight increase, it’s still not right. That’s because conditional probability isn’t the same as explanation, and if there’s a mystery, what we’re after is explanation. That’s one of the points of the tumor example: If fat causes intake, then saying “there’s no mystery about fat” is pointing away from the explanation of intake, which is that fat causes intake and something causes fat.
They later tried to address this because people picked up on it and eventually said “well, even if the 20% increase explains the obesity epidemic, that still leaves the question of why people are eating more open”. I think this is bizarre: to me it’s quite obvious that in the long run more calorie intake has to lead to higher body mass, though not necessarily in a proportional way as I’ve idealized above. They should have been focusing on the causal channel going through calorie intake from the start. Instead, it seems like to them this was a secondary channel to fall back on.
Taken literally, this seems to be consistent with beliefs of people who disagree with you? It’s just that they have different conclusions about the causality going through calorie intake. Interpreted that way, I don’t see how this statement is consistent with saying “there’s really no mystery in Americans getting fatter”.
I don’t understand your objection. I never made any claims about causality. My whole argument is about energy balance, which says nothing about causality.
On the other hand, I think it’s just true that conditioned on the increase in calorie intake there’s no mystery in the increase in body mass. I don’t understand why you’re disputing this point. You can say this is not an interesting observation (which I agree with, though as far as I can tell SMTM did not, which is why I wrote my comment) but I don’t see how you can say it’s not right.
My impression is that some people are engaging in a bizarre combination of steelmanning SMTM’s point while strawmanning my own. SMTM didn’t make the best version of the claim they could have made, they made the actual claim that I quote in my post. I think their claim is wrong. Do you disagree with this or not?
Meta: my interest here is to see if there are miscommunications here that I can clear up. I’m not carefully following the object-level debate. (In particular, I think that you Ege should feel extremely free to ignore what I’m saying as unhelpful to you; if I’m not helping you understand what’s happening in the thread then I’m not doing what I’m trying to do.)
I don’t understand your objection. I never made any claims about causality. My whole argument is about energy balance, which says nothing about causality.
On the other hand, I think it’s just true that conditioned on the increase in calorie intake there’s no mystery in the increase in body mass. I don’t understand why you’re disputing this point.
(Note: you did mention causality in the passage I quoted: “They should have been focusing on the causal channel going through calorie intake from the start.” That’s not a claim about what causes what, but it is a claim about what questions are the right questions to ask.)
I’m pointing at the word “mystery”. I’m saying that to me, “mystery” means “explanation wanted”. I’m saying that just because P(X|Y) is high, doesn’t mean Y is a good explanation of X. (For a silly example, setting Y=”X and 2+1=3″ makes P(X|Y) = 1 and is obviously doesn’t explain anything.) I agree (based on my preconceptions, ~0 independent data) that P(body mass high | high sustained intake in the wild) is high. My read of some of the comments on your comment, e.g. Yudkowsky’s, is that they are taking you to be saying “high intake explains fat, such that there is no further interesting question about fat, though there may be further questions about why high intake”, based on the passages from your comment I quoted. Reading your comment closely, you didn’t actually say that, if by “conditional on Y, there’s no mystery about X” you mean “P(X|Y) is high”. In fact, what you said is consistent with believing that “Alice is fat” explains (in the contextually relevant sense) that “Alice has high intake”, and you recommend that if you believe this then you should “focus[] on the causal channel going through calorie intake”, i.e. investigate why Alice is fat in order to explain her high intake.
SMTM didn’t make the best version of the claim they could have made, they made the actual claim that I quote in my post. I think their claim is wrong. Do you disagree with this or not?
I don’t know. I think your argument makes sense, but the actual situation is going to be more complicated.
I think I understood what you’re saying the first time around and again I agree that your account of things is certainly possible, but even in your case the tumor or fat mass has to be sustained by calorie consumption from the outside. The increase in calorie intake we’ve seen is about the amount we would have expected if someone had just told us that people are 15% fatter on average without any increase in exertion or heat dissipation to compensate for that.
Another way to say this is that I’m just making a claim about the conditional expectation E[average body mass increase | average calorie consumption increase = 20%] and pointing out this expectation is not far off from the actual observed body mass increase we see. In that sense the increase in calorie intake can account for the increase in body mass in correlational terms. The question of why both of these variables went up, though, is difficult and their level is obviously not determined by this argument.
Externally, why people feel the need to keep eating until they become obese is not a question with a clear (to me) answer. It could be because there is some process that’s operating in the body that’s taking priority over other activities and hoarding a lot of the energy intake to produce fat cells, which I think is your story. However, it could also be that some part of the brain is malfunctioning and leading the lipostat to be poorly calibrated. It could also be that the food you’re eating messes with the natural feedback loops in your body that are supposed to make you feel full when you’ve eaten enough.
I think this question is interesting and your account is logically possible. I just think that
I don’t have any special insight into which account is right, and
I don’t think your particular explanation is favored that much over other competing explanations.
I don’t think the overfeeding experiments provide strong evidence for your scenario, though I agree that they should be a Bayesian update in favor of accounts in that broad neighborhood. What would convince me is experiments which involve smaller increases in calorie intake but sustained over much longer periods of time, on the order of a year or so. If such experiments failed to find an effect that would be a strong update for me towards your view. Right now I buy the cancer analogy on conceptual grounds but I don’t think we have enough evidence to conclude obesity is like cancer in this regard, though it very well could be the case.
To give an analogy of my own, the overfeeding studies look similar to attempts to settle disputes about which programming languages are best for productivity by asking undergraduates to complete some simple tasks in them over the course of a few weeks. What really matters is how you do when you’re working with big and complex software in the real world that has to be developed and maintained by large teams with turnover for years, sometimes decades; but that obviously doesn’t lend itself to a simple experimental design so people still keep arguing about it.
Retrying again:
By the same reasoning:
“Pregnancy” probably isn’t a thing. “Pregnant” people eat around 500 more calories per day. This is sufficient to explain all the weight gain from “pregnancy” without supposing anything other than thermodynamics at work—anyone who eats an extra 500 calories per day will probably gain that much weight over the course of 40 weeks.
I think Ege’s alleging that SMTM presented two causal graphs:
calories → ? → weight gain
calories → weight gain
Ege’s saying that 2 is simpler and sufficient, so we don’t need to posit a ? in the middle.
You’re pointing out that we still need to address a third causal graph:
? → calories → weight gain
Edit: And maybe that there’s also a scenario where ?, calorie intake, and weight gain are all in some complex interrelationship. Maybe contaminants cause more fat deposition and less energy and more hunger, thereby increasing weight gain per calorie, increased calorie intake, and increased contaminant intake via food. Or something.
Ege’s agreeing with you, but wants to emphasize that this is compatible with criticism of SMTM’s alleged emphasis on graph 1.
Note: I say “alleged” only because I’m sidestepping evaluating the truth of Ege’s claim. Just trying to clarify what it is (AFAICT).
I don’t agree with this presentation of what I’m saying.
I’m not terribly sure what SMTM means when they say “the increase in calorie intake is small”, but all possible interpretations of their claim seem wrong. For instance, one plausible interpretation in causal graph lingo is “if you applied the do operator on calorie intake and raised it by 20%, we would have seen an increase in body mass that’s significantly smaller than what we’ve actually seen”. I think this claim is wrong, basically for long-run energy balance reasons.
I’m not saying anything else about the structure of the causal graphs, which could be arbitrarily complicated and involve arbitrarily many nodes and dependencies. I’m just saying that if you apply the do operator on calorie intake and raise it by 20% then you’d get an increase in mean body mass that’s about as big as what we’ve seen.
Thanks for clarifying that I misrepresented your view. Based on your response here, you’re pointing out that there’s a strong correlation between increased caloric intake at the population level, and increased obesity. You are also saying that the explanations you’ve read from SMTM for why this correlation exists seem wrong, and also that they underestimate the magnitude or importance of the caloric intake.
WRT Eliezer’s arguments, you seem to be agreeing that there may be some underlying force(s) causing that increased caloric intake. However, you are very uncertain about which, if any, of the hypothesized forces(s) are the true causes of increased caloric intake.
Eliezer and others seem to be perhaps mistakenly interpreting you as denying the existence of, or “need for,” a deeper explanation for increased caloric intake and consequent weight gain. You are confused about why they are making this mistake.
Is that a more accurate account of your position?
Yes, this summary is accurate.
I’m not sure who is “to blame” for the miscommunication but I suspect I simply was not clear enough in my top comment. Now it’s likely too late to clear up the issue for most readers as they won’t be following the developments in this thread.
Feel free to adapt, or copy/paste, the summary into your parent comment if you like.
This is “assuming there’s no link “increased calorie intake → increased energy expenditure”″, right? I think one of the things Eliezer is saying is that there seems to have been such a link in the past and now there isn’t / it’s much weaker.
That’s not quite true—there is at least the naive link that a higher equilibrium body mass leads you to expend more energy in daily activities even if you exercise the same amount as before. In my very naive model I assume these are directly proportional, but Natalia cites some better research that does a log-linear regression of calorie expenditure on equilibrium (I think? I didn’t check this part) body mass which seems to be more accurate empirically.
I think it’s unclear whether we had the link you mention in the past, too. We definitely had a correlational link: people who did hard labor and ended up exercising a lot every day took in much more calories, as we would expect, and they were generally not obese. However, I think my argument would work just as well in the past if you just applied the do operator on calorie intake per day and looked at the causal impact on equilibrium body mass, as I don’t think there’s evidence that there’s a big downstream link from calorie intake per day to exercise.
You left out weight gain->calories, as in the pregnancy example, and calories ← X → weight gain.
I don’t understand why you’re “retrying”. I already agree with your point and you not saying “yes, you already agree with me” is quite confusing to me. As I say in my comment:
Do you think this characterization of your position is unfair or wrong? If so, why?
As far as I can see the only object-level point I disagree with you about is that I don’t think the evidence for obesity being like cancer or pregnancy is as strong as you seem to think it is. It’s definitely possible for it to be like that but I would bet against it at even odds. I explain this here:
On top of that I also have a separate disagreement with you about emphasis in the context of my comment, since the point of my top comment is to draw attention to 400 kcal/day not being a small increase in calorie intake. You agree with me about this but you just don’t think it’s worth focusing on, probably because you think it’s a trivial observation. I still think it’s something that should be corrected given that SMTM explicitly said that it’s a small increase.
Some people certainly are obese because of literal cancer and literal pregnancy. We seem to have strong evidence for that.
The interesting question is about how much of the obesity pandemic is explainable by such factors and not whether evidence for such factors exists.
We certainly don’t see enough pregnancy and cancer to explain the obesity epidemic but there might be other factors that are similar but harder to see. Thermodynamic arguments don’t help us rule out other effects that are similar to pregnancy/cancer.
I agree with everything you said, so again I’m confused why you thought you should make this comment.
I feel like I don’t really disagree with most of the commenters but they either think I do disagree with them or that I did a very bad job of communicating exactly what my point was. It’s hard for me to understand.
(The thread continues to look to me like what I described here https://www.lesswrong.com/posts/7iAABhWpcGeP5e6SB/it-s-probably-not-lithium?commentId=NxzEfuyGfuao25mrx
i.e. Yudkowsky is responding to the part of your original comment where you said
)
I still stand by this claim, again with the caveat that you take it as a correlational use of the word “explain” (which is not at all uncommon e.g. when talking about “fraction of explained variance” and so forth) and not one that suggests a causal explanation of the form “people wanted to eat more food, so they ate more food, so they got fatter as a result”.
Ok. My main point is just to clarify that other people are reading you as talking about explanation in general, not just strictly correlational explanation (if that’s what’s happening).
I do also think that’s not a great use of the word “explain” and “mystery”, because it’s not why the colloquial word is useful. The colloquial words “explain”/”mystery” are useful because they index “more information and ideas given/needed about this”. So just because X correlationally explains Y, and X is true, doesn’t mean there’s no mystery about Y.
I never said there’s no mystery about Y, just that there’s no mystery about Y being true conditional on X being true.
It’s a fair point that my usage of “explain” and “mystery” confused some people but I’m not too sure how else I would have made my point. Should I have said “people today are eating about as much more compared to the past as we would expect given how much fatter they’ve gotten”?
That’s clearer to me, yeah. It’s unambiguous that it’s about conditional prediction (“we would expect given”) rather than explanation-in-general.
In your original comment, you wrote:
Read literally, this says: There’s a mystery of why intake went up. Conditioned on intake up, there’s no mystery of why fat went up. I think this isn’t right. If we agreed that sustained high intake implies weight increase, it’s still not right. That’s because conditional probability isn’t the same as explanation, and if there’s a mystery, what we’re after is explanation. That’s one of the points of the tumor example: If fat causes intake, then saying “there’s no mystery about fat” is pointing away from the explanation of intake, which is that fat causes intake and something causes fat.
Taken literally, this seems to be consistent with beliefs of people who disagree with you? It’s just that they have different conclusions about the causality going through calorie intake. Interpreted that way, I don’t see how this statement is consistent with saying “there’s really no mystery in Americans getting fatter”.
I don’t understand your objection. I never made any claims about causality. My whole argument is about energy balance, which says nothing about causality.
On the other hand, I think it’s just true that conditioned on the increase in calorie intake there’s no mystery in the increase in body mass. I don’t understand why you’re disputing this point. You can say this is not an interesting observation (which I agree with, though as far as I can tell SMTM did not, which is why I wrote my comment) but I don’t see how you can say it’s not right.
My impression is that some people are engaging in a bizarre combination of steelmanning SMTM’s point while strawmanning my own. SMTM didn’t make the best version of the claim they could have made, they made the actual claim that I quote in my post. I think their claim is wrong. Do you disagree with this or not?
Meta: my interest here is to see if there are miscommunications here that I can clear up. I’m not carefully following the object-level debate. (In particular, I think that you Ege should feel extremely free to ignore what I’m saying as unhelpful to you; if I’m not helping you understand what’s happening in the thread then I’m not doing what I’m trying to do.)
(Note: you did mention causality in the passage I quoted: “They should have been focusing on the causal channel going through calorie intake from the start.” That’s not a claim about what causes what, but it is a claim about what questions are the right questions to ask.)
I’m pointing at the word “mystery”. I’m saying that to me, “mystery” means “explanation wanted”. I’m saying that just because P(X|Y) is high, doesn’t mean Y is a good explanation of X. (For a silly example, setting Y=”X and 2+1=3″ makes P(X|Y) = 1 and is obviously doesn’t explain anything.) I agree (based on my preconceptions, ~0 independent data) that P(body mass high | high sustained intake in the wild) is high. My read of some of the comments on your comment, e.g. Yudkowsky’s, is that they are taking you to be saying “high intake explains fat, such that there is no further interesting question about fat, though there may be further questions about why high intake”, based on the passages from your comment I quoted. Reading your comment closely, you didn’t actually say that, if by “conditional on Y, there’s no mystery about X” you mean “P(X|Y) is high”. In fact, what you said is consistent with believing that “Alice is fat” explains (in the contextually relevant sense) that “Alice has high intake”, and you recommend that if you believe this then you should “focus[] on the causal channel going through calorie intake”, i.e. investigate why Alice is fat in order to explain her high intake.
I don’t know. I think your argument makes sense, but the actual situation is going to be more complicated.