by Judith Curry
For the paper that I am writing on uncertainty and the IPCC, I am including a section on “Consensus, Disagreement, and Argument Justification.” While googling around on this this topic, I encountered a fascinating body of work by Princeton philosopher Thomas Kelly, which I find mind-blowingly relevant to the climate conflict. Here are some excerpts from a few of his papers that I found to be particularly provocative.
Authors summary: Suppose that you and I disagree about some non-straightforward matter of fact. Psychologists have demonstrated the following striking phenomenon: if you and I are subsequently exposed to a mixed body of evidence that bears on the question, doing so tends to increase the extent of our initial disagreement. That is, in response to exactly the same evidence, each of us grows increasingly confident of his or her original view; we thus become increasingly polarized as our common evidence increases. I consider several alternative models of how people reason about newly-acquired evidence which seems to disconfirm their prior beliefs. I then explore the normative implications of these models for the phenomenon in question.
The first set consists of purely descriptive, psychological questions about how exactly You and I are responding to our evidence so as to generate the relevant phenomenon. The second set consists of normative questions. Given that You and I are responding to our evidence in such-and-such a way, is there any chance that our doing so is anything other than blatantly unreasonable? What is the epistemic status of the views at which we ultimately arrive by responding to our evidence in this way? How (if at all) should we attempt to counteract or correct for the relevant psychological tendency? As we will see, these normative questions are less straightforward than one might expect; pursuing them raises a number of rather subtle and delicate issues about what is to be objective or open-minded (on the one hand) as opposed to dogmatic or biased (on the other), as well as questions about the role that one’s background beliefs should and should not play in the assessment of new evidence.
All else being equal, individuals tend to be significantly better at detecting fallacies when the fallacy occurs in an argument for a conclusion which they disbelieve, than when the same fallacy occurs in an argument for a conclusion which they believe.
Light bulb: How many hundreds of examples have we seen of this?
Suppose that one is presented with an argument for a conclusion which contradicts something that one believes. One examines the argument, judges that it is not a good one, and so retains one’s original view. Suppose, moreover, that the fact that one judges that the argument is not a good one is contingent on the fact that one already disbelieved its conclusion prior to having been presented with the argument: if one had not already disbelieved the conclusion—if, say, one had been an agnostic or had not yet formed an opinion about the relevant issue—then one would have been persuaded by the argument. This might look extremely suspicious. After all, if an argument is sufficiently attractive that it would have convinced one if one had initially examined it from a standpoint of neutrality, how can it be legitimate for the crucial difference to be made by the fact that one already had an opinion about the issue in question, an opinion that, ex hypothesi, one arrived at in ignorance of the argument?
Of course, all of this might lead one to think that You and and I are guilty, not of giving too much scrutiny to evidence that seems to tell against our beliefs, but rather of giving too little scrutiny to evidence that seems to tell in their favor. Or better: perhaps our fault lies in the fact that we subject such evidence to different levels of scrutiny. That is, perhaps whatever absolute level of scrutiny we ought to devote to newly encountered evidence—indeed, even if no absolute level of scrutiny is rationally required of us—in any case, the one thing that we are rationally required not to do is to devote different levels of scrutiny to evidence depending on how well it coheres with our prior beliefs.
And, all else being equal, the more cognitive resources one devotes to the task of searching for alternative explanations, the more likely one is hit upon such an explanation, if in fact there is an alternative to be found.
Light bulb: personal insight – the climategate emails made me concerned that I had been duped by the IPCC, and I started questioning the stuff that I used to agree with, without having given it adequate scrutiny. Hence, Climate Etc.
The Key Epistemological Fact: For a given body of evidence and a given hypothesis that purports to explain that evidence, how confident one should be that the hypothesis is true on the basis of the evidence depends on the space of alternative hypotheses of which one is aware.
In general, how strongly a given body of evidence confirms a hypothesis is not solely a matter of the intrinsic character of the evidence and the hypothesis. Rather, it also depends on the presence or absence of plausible competitors in the field. It is because of this that the mere articulation of a plausible alternative hypothesis can dramatically reduce how likely the original hypothesis is on one’s present evidence.
Light bulb: Big implications for the “convergence of multiple lines of evidence” argument, when only one side of the argument is presented.
For the sake of explicitness, let’s bring the psychological and normative considerations together. As a psychological matter, when we encounter data that seem to go against what we believe, we are disposed to devote resources to the project of generating rival hypotheses to account for that data. To the extent that we are successful in generating plausible rivals, apparent counterevidence gets considered against a relatively rich space of alternative explanatory hypotheses. This fact tends to diminish the extent to which any particular hypothesis in the field gets confirmed or disconfirmed by the original evidence, inasmuch as the competitors tend to divide up the support conferred by the novel evidence among them. That is, the support which any one of the hypotheses receives is diluted by the presence of the others. (This last fact is a normative consequence of the operation of the relevant psychological process.) On the other hand, when we encounter evidence that is plausibly explained by things that we already believe, we typically do not devote additional resources attempting to generate alternatives. Data that seem to support hypotheses that are already believed thus tend to get considered against a comparatively impoverished or sparse background of alternative hypotheses. As a result of the less competitive milieu, the support conferred by the new evidence is not siphoned away, and thus tends to go in relatively undiluted form to the already accepted hypothesis. Over time, this invisible hand process tends to bestow a certain competitive advantage to our prior beliefs with respect to confirmation and disconfirmation.
There is an important asymmetry in the way that we respond to evidence that seems to tell against our prior beliefs and evidence that seems to tell in favor of those beliefs.
Light bulb: explains confirmation bias
On what I take to be the correct view of these matters, questions about (e.g.) how much time or effort one should devote to scrutinizing a given argument or piece of evidence are practical questions. Typically, the reasons that one has to devote further thought to a given argument or piece of evidence (if any) compete with other practical considerations. In such cases, rationality is always in part a matter of opportunity cost, in the economists’ sense.
Note to John O’Sullivan: I viewed spending too much time on Sky Dragon would result in an opportunity cost. But I spent enough time on it to refute it in my own mind based upon my background understanding of physics and my decades of research on this topic. To provide an adequate hearing for Sky Dragon, I opened it up to discussion at Climate Etc. for those who did not view spending time on this this to be an opportunity loss.
On the contrary, the tendency to devote more thought to that which seems to violate or run counter to one’s expectations would seem to be the natural or default state, which prevails unless one deliberately makes a conscious effort to devote equal thought to those considerations which seem to support what one already believes.
As we’ve seen, one manifestation of our lack of even-handedness in responding to new evidence is our tendency to devote fewer cognitive resources to searching for alternative explanations of a given fact when we already believe some hypothesis that is contrary. Scientists do not treat the anomalous phenomena and the non-anomalous phenomena on a par. On the one hand, scientists devote relatively little attention and effort to attempting to devise plausible alternative explanations of phenomena for which the accepted theory already offers a plausible explanation. On the other hand, scientists devote a great deal of attention and effort attempting to generate hypotheses that allow the existence of the anomalies to be reconciled with the presently accepted theory (to the extent that such is possible). Is this rational behaviour?
Indeed, Lord, Ross, and Lepper suggest that You and I are properly subject to criticism only insofar as our initial convictions are held more strongly than is warranted by our original evidence.
Light bulb: the “overconfidence” thing
The suspiciousness of this is perhaps even greater when we focus once again on the interpersonal case of two individuals who have been exposed to both pieces of evidence, differing only in the order in which they encountered that evidence. If they both reason in the way described, the model predicts that they might very well end up with different levels of confidence towards the proposition that capital punishment is a deterrent, despite apparently having the same total evidence. In that case, it looks as though two individuals who share the same total evidence end up believing different things because of historical facts about the relative order in which they encountered the elements that comprise that total evidence. This is because, when individuals reason in the envisaged way, they do not in fact end up with the same total evidence in the relevant sense. Here it’s important to distinguish between two different senses of ‘evidence’, a broad sense and a narrow sense. Evidence in the narrow sense consists of relevant information about the world. Statistical information about crime rates is, perhaps, a paradigm of evidence in the narrow sense. As a rough rule of thumb: evidence in the narrow sense consists of things that it would be natural to call ‘data’. In the narrow sense of evidence, the individuals in the scenario described above have the same total evidence. On the other hand, we can also speak about evidence in the broad sense. Evidence in the broad sense includes everything of which one is aware that makes a difference to what one is justified in believing. Clearly, evidence in the broad sense includes evidence in the narrow sense, inasmuch as relevant data or information of which one is aware typically does make a difference to what one is justified believing. But one’s evidence in the broad sense will include, not only evidence in the narrow sense of data, but also things such as the space of alternative hypotheses of which one is aware. For (by the Key Epistemic Fact) which hypotheses one is aware of can make a difference to what one is justified in believing. Now, even if two individuals have exactly the same evidence in the narrow sense, they might have different evidence in broad sense, in virtue of differing with respect to the set of hypotheses of which they are aware. But if they have different evidence in the broad sense, then they might differ in what they are justified in believing, despite having exactly the same evidence in the sense of data. (Again, this will be admitted by anyone who accepts the Key Epistemic Fact.)
Light bulb: this explains the different perspectives on the climate change issue between climate researchers versus weather forecasters, and versus other people that might have just started looking at this motivated by the climategate emails.
Historical facts about when one acquires a given piece of evidence might make a causal difference to which body of total evidence one ultimately ends up with. One acquires a given piece of evidence at an early stage of inquiry; this might very well influence the subsequent course of inquiry in various ways, by way of making a difference to how one subsequently thinks and acts (which possibilities one considers, which routes get explored as the most promising and fruitful, and so on.) And this in turn can make a difference to what evidence one ends up with. In such cases, there is an undeniable element of path-dependence.
The following is, I believe, a not uncommon pattern. Relatively early on, one picks up a view about some controversial matter, a view that is not shared—and indeed, is explicitly rejected–by some who have considered the question. Perhaps one even picks up the view at One’s Parent’s Knee. Once one first begins to hold the view, one retains it thereafter. Perhaps at various times one is somewhat more confident than at other times, but after one first comes to hold the view, one can from then on be correctly described as believing the relevant proposition. Over time, however, the reasons for which one holds the view evolve. That is, the reasons for which one believes that [EXAMPLES] are not identical to the reasons for which one held this belief, when one first began to hold it.
For as we have seen, the fact that a belief is held at earlier times can skew the total evidence that is available at later times, via characteristic biasing mechanisms, in a direction that is favorable to itself. The concern is not (simply) the banal point that an individual who has long held a given view might easily fall into overestimating how well-supported it is by the considerations available to him; rather, the very fact that he has this particular body of considerations available, rather than one that is significantly less favorable, might very well be due to the fact that he has long been a believer.
In deciding what level of confidence is appropriate, we should taken into account the tendency of beliefs to serve as agents in their own confirmation.
Light bulb: perfect one sentence summary of the problem with the IPCC.
It is a striking fact that almost everyone holds at least some beliefs that are explicitly rejected by others who have been exposed to all of the same evidence and arguments. When a belief that one holds is explicitly rejected by individuals over whom one possesses no discernible epistemic advantage, does this give one a reason for skepticism about that belief? In deciding what to believe about some controversial question, how (if at all) should one take into account the considered judgements of one’s epistemic peers? I explore these and related questions. I argue that an awareness of the relevant kind of disagreement need not undermine the rationality of maintaining one’s original views.
Let us say that two individuals are epistemic peers with respect to some question if and only if they satisfy the following two conditions:
(i) they are equals with respect to their familiarity with the evidence and arguments which bear on that question, and
(ii) they are equals with respect to general epistemic virtues such as intelligenc thoughtfulness, and freedom from bias.
Light bulb: the PNAS paper was an attempt to dismiss skeptics as not being at the same epistemic level as the consensus group.
The beliefs of a reasonable individual will thus constitute higher-order evidence, evidence about the character of her first-order evidence. Given the general reasonableness of one’s epistemic peers, what they believe on the basis of one’s shared evidence will thus constitute evidence about what it is reasonable to believe on the basis of that evidence. There are subtle questions, I think, about how one should integrate such higher-order considerations into one’s own deliberations and what difference such considerations make to what it is reasonable for one to believe.
The question at issue then, is whether known disagreement among those who are epistemic peers in this sense must inevitably undermine the rationality of their maintaining their respective views.
Disagreement among epistemic peers then, is disagreement among those who disagree despite having been exposed to the same evidence. According to this line of thought, the only thing that would justify one in maintaining views that are rejected by one’s epistemic peers would be if one had some positive reason to privilege one’s own views over the views of those with whom one disagrees.
On the present view, the rationality of the parties engaged in such a dispute will typically depend on who has in fact correctly evaluated the available evidence and who has not. If you and I have access to the same body of evidence but draw different conclusions, which one of us is being more reasonable (if either) will typically depend on which of the different conclusions (if either) is in fact better supported by that body of evidence. Therefore, the question of which one of us is doing a better job evaluating the evidence will often be a non-trivial, substantive intellectual question.
Given that reasonable individuals are disposed to respond correctly to their evidence, the fact that a reasonable individual responds to her evidence in one way rather than another is itself evidence: it is evidence about evidence.
Light bulb: the “quality of evidence” thing.
The beliefs of a reasonable individual will thus constitute higher-order evidence, evidence about the character of her first-order evidence. Of course, an awareness of disagreement can serve to call one’s attention to arguments that one might never have considered or to evidence of which one might have been unaware.
As more and more peers weigh in on a given issue, the proportion of the total evidence which consists of higher order psychological evidence [of what other people believe] increases, and the proportion of the total evidence which consists of first order evidence decreases. As the number of peers increases, peer opinion counts for progressively more in determining what it is reasonable for the peers to believe, and first order considerations count for less and less. At some point, when the number of peers grows large enough, the higher order psychological evidence will swamp the first order evidence into virtual insignificance.
Light bulb explodes from high voltage of the argument
Throughout, we have been concerned with the probative force of peer opinion in cases in which the peers arrive at their opinions independently of one another. This assumption of independence tends to maximize the probative force of peer opinion relative to the probative force of first-order evidence. Impressive evidence that a given answer to a question is the correct answer is afforded when a large number of generally reliable peers independently converge on that answer. On the other hand, the less their convergence is an independent matter, the less weight such convergence possesses as evidence.
The general moral: even in cases in which opinion is sharply divided among a large number of generally reliable individuals, it would be a mistake to be impressed by the sheer number of such individuals on both sides of the issue. For numbers mean little in the absence of independence. If one uncritically assumes that the members of the contending factions have arrived at their views independently, then one will tend to overestimate the importance of other people’s opinions as evidence and underestimate the importance of the first order evidence and arguments. One will be too quick to conclude that agnosticism is the reasonable stance in cases in which opinion is sharply divided, and too quick to conclude that deference to the majority is the reasonable course in cases in which opinion is not sharply divided.
Memo to the NAS, NIPCC, Gang of 18, etc.: you can stop signing all those petitions.
A few other papers of potential interest:
Author summary: If you are more likely to continue a course of action in virtue of having previously invested in that course of action, then you tend to honor sunk costs. It is widely thought both that (i) individuals often do give some weight to sunk costs in their decision-making and that (ii) it is irrational for them to do so. In this paper I attempt to cast doubt on the conventional wisdom about sunk costs, understood as the conjunction of these two claims.