In January 2013 Jime Sayaka interviewed me on the topic of postmortem survival for his now defunct blog Subversive Thinking. In what turned out to be a lengthy interview (and preview of arguments in my forthcoming book), I outlined in considerable detail my critique of empirical arguments for survival, as well as explained why common survivalist defenses of these arguments lack cogency. Below I repost my lengthy response to Sayaka’s fourth question, something of an invite for me to share my main criticisms of empirical survival arguments.
4 – Sayaka: “Professor Sudduth, you have been a philosophical critic of the survivalist hypothesis to explain the empirical data from mediumship and other putative evidence for survival of consciousness. What are your objections for the survival hypothesis?”
Sudduth: Well, let me begin with some important caveats and clarifications. Unlike many other philosophers, I don’t object to the survival hypothesis itself, nor do I deny that people can be epistemically justified in believing in survival. I’ve already stated that I subscribe to the eastern philosophical and spiritual tradition of Vedanta. So I don’t believe that what I essentially am shares in the limits or destiny of my body or individual mind. I am a survivalist. I also don’t deny that empirical evidence can add to the justification of belief in survival, for instance, by adding to the evidential probability of the survival hypothesis. And I think there’s much to be said for how the survival hypothesis may draw support from multiple grounds, for example, empirical, philosophical, and religious or spiritual. But this requires a very different approach than has been traditionally taken by the majority of empirical survivalists. My present project is, therefore, concerned with the critique and dismantling of the existing and deeply entrenched tradition of classical empirical arguments for survival. Hopefully it paves the way for new and fruitful approaches to empirical arguments for survival.
So let’s unpack some of the details of my argument.
As I see it, there’s really no way to make an empirical case for survival unless we can show that the features of the world marked out by the relevant data are what we would expect if the survival hypothesis is true, and furthermore that these features are more to be expected if survival is true than if survival is false (or, more modestly, if some alternative non-survival hypothesis is true). So what is often called predictive power, at least understood in a broad sense, is essential to an empirical case for survival. As it happens, most survivalists have either claimed or assumed the same, usually in connection with how the “explanatory power” of the survival hypothesis is parsed. But it’s more generally relevant because predictive power, or the probability of the evidence given a hypothesis, plays an important role in the two dominant approaches to evidence assessment in confirmation theory, Bayesian and Likelihoodist approaches, both of which I will subsequently discuss.
However, predictive salience subjects the survival hypothesis to anauxiliary hypothesis requirement. Theoretically, this arises from the general Duhem-Quine thesis in philosophy of science that single statements rarely have predictive consequences, unless they’re supplemented with auxiliary hypotheses. So hypotheses can only be tested via their predictive consequences in bundles or sets. This is repeatedly demonstrated in the history of science, but I remember first seeing it dramatically illustrated in the old television series Columbo. When detective Columbo tests his hypothesis that Dr. Brimmer murdered Mrs. Kennicut, he relies on a number of additional assumptions, many of which are statements about Dr. Brimmer (e.g., having a particular connection to the victim, being left handed, having a temper, wearing a diamond ring with a unique shape). These auxiliary assumptions, together with the hypothesis that Dr. Brimmer committed the crime, leads Columbo to expect to find the crucial pieces of evidence, which only function as “clues” because they are linked to the murderer by way of a particular set of added assumptions.
It’s a central part of my argument that this is true with respect to the survival hypothesis. The data collected from mediumship or cases of the reincarnation type only serve as evidence for personal survival once various auxiliary hypotheses are introduced to facilitate the link between the data and the continued existence of the deceased person. This is often glossed over, or simply not acknowledged at all, because empirical survivalists routinely treat the survival hypothesis as a generic survival hypothesis, for example, the survival of individual consciousness, the mind, or the self. But this kind of simple survival hypothesis does not lead us to expect the relevant data, unless it is supplemented with a wide range of auxiliary statements about the knowledge, intentions, and causal powers of postmortem persons, as well as the mechanism or process of postmortem communication (in the case of mediumship) and rebirth (in reincarnation cases).
The necessary reliance on auxiliary hypotheses is clear if we carefully read classic works on the empirical arguments for survival such as E.R. Dodds’s “Why I Do Not Believe in Survival” (1934), Hornell Hart’s Engima of Survival (1959) and Alan Gauld’s Mediumship and Survival (1982). Hence, inasmuch as the empirical case for survival depends on predictive derivations that logically link the survival hypothesis to specific features of the empirical world (captured by the relevant data), the empirical case for survival requires what I call a robust survival hypothesis. While empirical survivalists usually assume some robust version of the survival hypothesis, they rarely acknowledge this with adequate transparency; much less do they critically explore it. Consequently, they fail to consider its significance to the overall case for survival. And this is a crucial issue as I see it because the satisfaction of the auxiliary hypothesis requirement has significant consequences for assessments of both the prior probability of the survival hypothesis and its explanatory power, the two determinants of the posterior probability of the survival hypothesis.
Here I make two points.
First, I argue that the survival hypothesis can only adequately satisfy the auxiliary hypothesis requirement at the cost of a significant reduction of prior probability. The predictive power of the survival hypothesis (i.e., its ability to lead us to expect the relevant data) is inversely proportional to its prior probability: as the predictive power of the survival hypothesis is increased, its prior probability is decreased, specifically as a result of increased complexity and less fit with background knowledge. So a survival hypothesis with great explanatory power will I’m afraid not have very high prior probability, and certainly not greater prior probability than the nearest competitors. Within a Bayesian framework, this will significantly lower the posterior probability of the survival hypothesis.
Second, the survival hypothesis can only adequately satisfy the auxiliary hypothesis by adopting assumptions that lack independent support and testability. In this way, they are quite different from Columbo’s auxiliary hypotheses, or the kinds of auxiliary statements employed by scientists. For example, there is no independent evidence for supposing that persons, should any of them survive death, will have the intention and requisite powers to communicate with living persons, much less in ways that as much as approximate the modality of mediumship or apparitions. We also have no independent reason to suppose that discarnate persons will have awareness of events taking place in our world or the mental lives of living persons, which is required if mediumistic communications genuinely originate from discarnate persons. Furthermore, we have no good independent reason to suppose that some or all living persons would reincarnate on earth, much less as humans or with past life memories, congenital birth marks corresponding to the manner of their death in a former life, etc. In short, we don’t know what would happen to consciousness if it should survive death, nor do we know anything about the causal laws to which postmortem existence and agency would be subject. And, at present at any rate, there is no way to independently test hypotheses at this juncture. In fact, if the afterlife is anything like dream experiences or some other similar altered states of consciousness—the closest conjectured analogues of the afterlife—I would say the relevant data are actually not what we would expect.
Now the lack of independent testability has important implications for the explanatory power of the survival hypothesis. Since the epistemic credentials of the auxiliary hypotheses are quite weak, they can only be methodologically sanctioned by a very permissive principle governing the inclusion of auxiliary hypotheses to test the survival hypothesis. The problem here is that it is prima facie implausible to suppose that any such liberal principle will simultaneously entitle empirical survivalists to their stock of auxiliary hypotheses and not entitle others from including whatever auxiliary hypotheses are needed to generate predictive consequences for proposed alternative non-survival explanations. In other words, the empirical survivalist faces the problem of purchasing predictive power for the survival hypothesis at the cost of indirectly purchasing it for alternative hypotheses as well. So it won’t be the case that a robust survival hypothesis will lead us to expect data that are otherwise improbable, nor even that the data would be more likely given a robust survival hypothesis than robust alternative hypotheses.
Now consider the bearing of these points on run-of-the-mill defenses of empirical arguments for survival.
First, consider defenses of the prior probability of the survival hypothesis. When empirical survivalists defend the prior probability of the survival hypothesis, they consider the hypothesis only in its simple form, for example, the mere supposition of one’s individual consciousness persisting after death. There’s a lot of expended effort to defend substance dualism, critique materialist philosophies of mind, or dismantle arguments from cognitive neuroscience that purport to show the dependence of consciousness on neural substrates and hence a functioning brain. Important as these moves are, their success is limited. While they may remove prominent reasons for supposing that the prior probability of the survival hypothesis is low, they do not show that its prior probability is high. More importantly, they do not defeat arguments that purport to show that the prior probability of the survival hypothesis is low, not because of the supposition of survival itself, but because of the nature and consequences of the auxiliary hypotheses that are needed to generate predictive power for the survival hypothesis.
Next, consider critiques of the nearest explanatory competitors. There’s a pretty widespread consensus in the survival literature that the nearest explanatory competitor, which ostensibly accounts for the relevant data, is the appeal to living-agent psi in the form of extra sensory perception and/or psychokinesis among living agents. Now among empirical survivalists it’s virtual orthodoxy that this counter-explanation fails, for at least two reasons:
1. Appeals to living agent psi are rejected since they are allegedly inferior in explanatory power. For example, living-agent psi does not lead us to expect living persons exhibiting personality traits and skills characteristic of the deceased, as if the case in the better cases of the reincarnation type and trance mediumship. Also, living-agent psi would allegedly not lead us to expect the complex sets of veridical information found in these cases. This would require that the data be psychically derived from multiple sources, but outside survival-type cases there’s no evidence that living-agent psi has this kind of efficacy.
2. The second line of attack is to concede a possible version of the living-agent psi hypothesis that might explain these data. If living-agent psi were stretched into a “super-psi” hypothesis—positing living-agent psi functioning of a quite extraordinary degree or kind—and further supplemented with various supplemental assumptions about how human abilities and (conscious and unconscious) motivations are likely to play a role in accounting for the data. But empirical survivalists typically reject this strengthened living-agent psi hypothesis because it’s highly complex and lacks independent support. In Bayesian terms, this explanatory competitor can only purchase predictive success at the cost of significantly lowered prior probability.
In the light of my earlier observations, it should be clear why these objections fail. The strategy suggested by the above objections is essentially to argue that a robust survival hypothesis has greater explanatory power than simple explanatory competitors (e.g., a vanilla living-agent psi hypothesis), and a simple survival hypothesis has greater prior probability than the nearest robust competitor (living-agent psi + auxiliaries). This may be true, but it’s ultimately irrelevant. We must compare the values assigned to explanatory power and prior probability of robust versions of each of the candidate explanations. When we try to do this, I argue that (a) the prior probability of the robust survival hypothesis is either equal to or less than the prior probability of the nearest robust explanatory competitor(s) and (b) the predictive power of the robust survival hypothesis is equal to or less than the predictive power of the nearest robust competitor(s). From a Bayesian approach to calculating posterior probabilities, I think (a) and (b) significantly deflate the posterior probability of the survival hypothesis. Consequently, the survival arguments fail to show that the posterior probability of the robust survival hypothesis, given the evidence and usual assignments to background knowledge, exceeds ½.
It is, of course, crucial to this argument that the content of the background knowledge and scope of the evidence be carefully spelled out, and I do so in my book. And there’s a thorny problem here concerning just where to draw the parameters that isolate the total available and relevant evidence. The problem also appears with respect to identifying the parameters of background knowledge. What we include as evidence and background knowledge has consequences for judgments of the posterior probability of h because it affects the values assigned with respect to prior probabilities (of h and e) and the posterior probability of e given h (i.e., predictive power).
For example, I would say that the robust survival hypothesis has greater explanatory power than the robust living-agent psi hypotheses when the parameters of the evidence are more narrowly drawn, e.g., vis-à-vis mediumship—excluding evidence that that communicators provide inconsistent and unreliable information and that mediumistic controls are sometimes fictitious and yet convey accurate information. In fact, evidence within narrow parameters frees the survival hypothesis from the need to adopt a number of auxiliary assumptions, and thereby circumvents conditions that would further lower the prior probability of the survival hypothesis. So if we pick and choose the evidence, constrain its parameters in particular ways, the case for survival actually looks pretty good. I suspect this is why some empirical survivalists think that the evidence for survival is good. In much the same way, it looks like we have a good case for supposing that conditions are optimal for swimming at the beach given that the weather is warm, the ocean water isn’t turbulent, and there are only a modest number of people at the beach. However, all this changes once we add that several sharks have been spotted in the waters earlier in the morning. It’s a canon of inductive logic that you consider the total evidence available in assessing the net plausibility of a hypothesis. I think this is yet another point where survival arguments are vulnerable because they typically operate with implausibly narrow parameters on the relevant evidence. So one of my interests is to identify and carefully describe the total available evidence, as well as consider the implications of different parameters for background knowledge.
Follow