This idea of calorie intake and expenditure being an epiphenomenon… Taubes certainly does say things that seem to suggest that, but what it would even mean for that to be true?
“Epiphenomenon” is somewhat hyperbolic, but it does make a sensible claim. To make clear what that claim is, it is necessary to think about causal graphs, because intervention in a system to produce a desired result can only be successful if it is based on a correct understanding of how the system works.
“dW/dt = Calories in—calories out”, while literally true, carries with it the suggestion that a sufficiently accurate causal graph for this problem is one with two arrows, from input to weight and from output to weight. All you have to do to lose weight is to eat less and/or exercise more.
If the causal model is correct, the predicted result of an intervention will happen. If the predicted result does not happen, the model is wrong.
It seems to be more often the experience than not, that the predicted result does not happen. This brings the model into question.
Causal models make two sorts of claim: the claims that are seen, and the claims that are not seen. The claims that are seen are the variables and the arrows of the model: they claim that these properties of the world exist, and these causal influences exist among them. The claims that are not seen are the absences of variables and arrows. Where there is no arrow, the model claims that there is no direct causal effect. Where there is no variable, the model claims there is no other phenomenon in the world causally relevant to the things being modelled.
To repeat in the face of the failed prediction, “but...input minus output!” is to attend only to the claim that is seen. One of the claims that is not seen in this model is the absence of an arrow from input to output. Suppose we hypothetically add one: suppose that restricting calorie intake makes the body reduce its expenditure also. (Or in concrete terms: skip eating for a day and collapse with exhaustion.) What is now the effect on weight of eating less? That depends on the details and relative magnitudes of how these things influence each other. That is just one example. There are many ways in which “dW/dt = Calories in—calories out” could be embedded in a larger graph for which the claimed remedy for overweight will fail. When they fail, it is not because “dW/dt = Calories in—calories out” is false, but because it is incomplete.
For the case of growing children, where it is was said that they eat because they are growing, rather than growing because they eat, the claimed causal graph appears to be something like this: the body’s internal processes of development cause a demand for food; the demand for food causes eating; eating makes materials available for growth; growth is sensed by the body’s internal processes of development, which adjusts demand for food accordingly. The causal arrows form a cycle, part of which I’ve nebulously called “the body’s internal processes”. There will be an arrow into that node from other internal processes, specifying how fast to grow. (Observe that overfed chidren do not develop normally, but faster; they develop at the same speed and also grow fat.) That is what is driving the cycle, hence the paradoxical sounding “they eat because they are growing, rather than growing because they eat”.
Notice that most of these hypothetical causal graphs describe processes internal to the organism and difficult to observe or intervene on, and not all that much is definitively known. This is what makes this a hard problem. It is doubly hard if one does not realise that one must think in these terms to make any progress.
“Epiphenomenon” is somewhat hyperbolic, but it does make a sensible claim. To make clear what that claim is, it is necessary to think about causal graphs, because intervention in a system to produce a desired result can only be successful if it is based on a correct understanding of how the system works.
“dW/dt = Calories in—calories out”, while literally true, carries with it the suggestion that a sufficiently accurate causal graph for this problem is one with two arrows, from input to weight and from output to weight. All you have to do to lose weight is to eat less and/or exercise more.
If the causal model is correct, the predicted result of an intervention will happen. If the predicted result does not happen, the model is wrong.
It seems to be more often the experience than not, that the predicted result does not happen. This brings the model into question.
Causal models make two sorts of claim: the claims that are seen, and the claims that are not seen. The claims that are seen are the variables and the arrows of the model: they claim that these properties of the world exist, and these causal influences exist among them. The claims that are not seen are the absences of variables and arrows. Where there is no arrow, the model claims that there is no direct causal effect. Where there is no variable, the model claims there is no other phenomenon in the world causally relevant to the things being modelled.
To repeat in the face of the failed prediction, “but...input minus output!” is to attend only to the claim that is seen. One of the claims that is not seen in this model is the absence of an arrow from input to output. Suppose we hypothetically add one: suppose that restricting calorie intake makes the body reduce its expenditure also. (Or in concrete terms: skip eating for a day and collapse with exhaustion.) What is now the effect on weight of eating less? That depends on the details and relative magnitudes of how these things influence each other. That is just one example. There are many ways in which “dW/dt = Calories in—calories out” could be embedded in a larger graph for which the claimed remedy for overweight will fail. When they fail, it is not because “dW/dt = Calories in—calories out” is false, but because it is incomplete.
For the case of growing children, where it is was said that they eat because they are growing, rather than growing because they eat, the claimed causal graph appears to be something like this: the body’s internal processes of development cause a demand for food; the demand for food causes eating; eating makes materials available for growth; growth is sensed by the body’s internal processes of development, which adjusts demand for food accordingly. The causal arrows form a cycle, part of which I’ve nebulously called “the body’s internal processes”. There will be an arrow into that node from other internal processes, specifying how fast to grow. (Observe that overfed chidren do not develop normally, but faster; they develop at the same speed and also grow fat.) That is what is driving the cycle, hence the paradoxical sounding “they eat because they are growing, rather than growing because they eat”.
Notice that most of these hypothetical causal graphs describe processes internal to the organism and difficult to observe or intervene on, and not all that much is definitively known. This is what makes this a hard problem. It is doubly hard if one does not realise that one must think in these terms to make any progress.