Cross-posted from Je fais, donc je suis
I’ve spent this weekend at the Rotman Institute’s big annual conference, this year on climate science. In this post, I want to follow up on an exchange Saturday morning, between Gavin Schmidt and Eric Winsberg.
Some quick context: In a 2012 paper, Winsberg applied Heather Douglas’ inductive risk argument to climate science. Put very generally, Douglas’ argument is that, when scientists make judgments that have non-epistemic consequences, they have a responsibility to take those consequences into account when making those judgments. Winsberg more-or-less simply pointed out that both conditions in the antecedent apply to building climate models: building these models involve numerous judgments, and (at least) insofar as these models will be taken up and used for public policy purposes, they have quite significant non-epistemic consequences.
Schmidt’s response to this argument was made very briefly, and I don’t think he quite understood what the argument was saying. He seemed to understand Douglas and Winsberg as raising concerns about bias: that, whenever scientists make judgments, there’s a live possibility that they will tweak models in ways that promote certain ethical and political interests. For example, they might increase the frequency or severity of cyclones, in order to enhance worries about climate change. But this is worrisome, in several ways: it seems like scientists shouldn’t be trying to manipulate public concerns in these ways, and in response people might reject manipulated climate models.
Schmidt’s response is that climate model phenomena like cyclones are emergent; that is, they’re produced by interactions among the low-level model dynamics, but we don’t have precise explanations that reduce these phenomena to the low-level dynamics. Consequently, these phenomena generally can’t be tweaked in the way suggested in the previous paragraph.
Now, Douglas and Winsberg aren’t suggesting that climate modelers are biased; nor are they suggesting that climate modelers should tweak their models in the way suggested two paragraphs ago. Rather, they (or, at least, Douglas) are arguing that scientists have a responsibility to take the non-epistemic consequences of their judgments into account. For example, if modifying a parameter makes a difference for cyclone frequency, which makes a difference for the level of public concern about climate change, modelers should take this into account; for example, in both scientific and public communications they might discuss this relationship between the parameter and cyclone frequency.
But something like Schmidt’s response is quite relevant here. Reconstructing a little bit, Schmidt’s point was that, because these model phenomena are emergent, modelers aren’t in a position to trace causal connections from the judgments they make through the model phenomena and on to the non-epistemic consequences. This means that they can’t take the downstream consequences into account when making the upstream judgments.
This is a more significant problem than it might seem at first, if we accept a meta-ethical principle called ought implies can. According to this principle, someone can have a moral responsibility — they ought to do something, X — only if it’s possible for them to do X. Equivalently, if it’s impossible for you to do X, then you aren’t morally obligated to do X.
In the case of climate modeling, this implies that, if Schmidt is right and climate modelers can’t take the downstream consequences into account when making model design judgments, then Douglas and Winsberg are wrong and the modelers don’t have a responsibility to take these consequences into account. This is similar to a problem I recently raised, with Charles Pence, in the context of what we called massive interconnectivity in technology development: when researchers aren’t in a good position to understand how their work will be taken up and used by others, there’s no way for them to take into account the non-epistemic consequences of this work.
Does this mean that ethical and political values have no role in building climate models? No, I don’t think so — but for rather different reasons than Douglas’ and Winsberg’s concerns about inductive risk.
On my view, climate models are (or should be) evaluated by both epistemic and pragmatic standards. These two kinds of standards are related, but the relationship is complicated, and for the purposes of this post we can treat them as roughly independent. Roughly, epistemic standards concern the truth or empirical or theoretical adequacy of the models, while pragmatic standards concern the practical usefulness of the models. A model that falls short pragmatically is deficient, whatever its epistemic merits. More concretely, a good climate model needs to be both skillful for prediction and useful for developing adaptation and mitigation policy.
Now, “useful for developing policy” is a standard that’s highly dependent on ethical and political values. A pragmatically useful model — for “us” — is one that’s useful for promoting “our” ethical and political values. (Not in the sense of bias, manipulation, or wish-fulfillment, but rather in the sense that you should use a hammer to hang a picture because it’s the appropriate tool for the job.) At this point, I think my view could help elaborate Wendy Parker’s notion of “adequacy for purpose,” which has been discussed repeatedly this weekend.
How does my view avoid Schmidt’s response to Douglas and Winsberg? On my view, ethical and political values probably won’t come into play in making low-level model design judgments. They’re probably not relevant to decisions about parameter values, for instance, in part for the kinds of reasons Schmidt discussed. But ethical and political values are enormously importantly in high-level model design, and in ways that can feed back into epistemological issues.
Consider the question of whether climate modelers should focus on improving precision and realism — simulating climate systems in as much detail as possible — or on improving understanding — using a complex hierarchy of models, from simple to complex, to better understand the processes underlying climate phenomena. This might seem to be a strictly epistemological issue, concerning the relative epistemic value of empirical adequacy vs. explanatory power. But it’s plausible to argue that for purposes of developing good, robust adaptive climate policy, significantly more precision will probably not be very useful, while significantly better understanding will be enormously useful. We probably don’t need to know the number of heating-degree days in a given city in a given year later this century; yet we probably do need to understand the key drivers of the jet stream precession, which caused the “polar vortex” this past winter in North America. In this light, pragmatic ethical and political considerations — we want models that can usefully inform adaptive policy development — provide a reason for pursuing the understanding strategy rather than the precision strategy.
Douglas’ stated account is more specific: for reasons I’ve never really understood, she insists that scientists have these responsibilities only when there are harmful non-epistemic consequences of error. ↩