Author Archives: Farhan Islam

Week 2: Psillos

Psillos critically examines the epistemology of Bas van Fraassen, not with regard to any specific questions about realism and antirealism, but with its implications for rationality. In particular, he takes issue that van Fraassen’s conception of rationality has no compelling story about prior beliefs, ignores the content of belief, unfairly rejects IBE,  and calls rational what we really ought to be able to call irrational.

Background: Sailing between Bayesianism and Tradition

To understand van Fraassen’s view of rationality, let’s first get a handle on the positions that he is trying to mediate between. 

Traditional Epistemology: 

Traditional Epistemology is not a particularly well defined position per se, but the relevant features of it for this debate is that it suggests that we are rationally obligated to follow ampliative rules in our belief formation. A rule is ampliative if it (supposedly) allows for one to derive conclusions that extend beyond the premises. A way to think of it is that they are arguments that can be good but not sound in the strict sense. In particular, the ampliative rules in question are Induction and Inference to Best Explanation (IBE). Induction is the inference to a general law from a sufficient set of particular instances (eg. every day of my life, the sun has risen, thus the sun will rise everyday) and IBE is the belief in the explanation that best suits the evidence (eg. all of my electronics stopped working at the same time, so the power is out). This is especially important because in traditional epistemology, since for a belief to be rational, it must be justified, which leaves nothing to choice. Even though these inferences are not deductive, we say that it is rational to make them, and more saliently, irrational not to

Bayesian Epistemology:

Bayesian Epistemology designed to put induction in purely logical terms. Bayesians do that by focusing on beliefs not as binary belief and disbelief, but as a range of credences one can hold towards a proposition. In doing so, inductive inferences can be treated with the same level of deductive rationality as ordinary propositional logic. There are two views on rationality within Bayesianism: synchronic and diachronic. Synchronic Bayesianism constrains rational belief by whether or not beliefs can be held at the same time coherently. The corresponding rule in ordinary logic is the principle of noncontradiction. Since credences are a range, irrationality (to a synchronic bayesian) is demonstrated with a Dutch Book argument. These show that if you were to place a series of bets that correspond perfectly to your beliefs at the same time, you would suffer a loss (imagine a situation in which a coin would be flipped 100 times and you bet $150 dollars that it would land heads 75 times and tails 75 times; you would necessarily lose $50). A Bayesian can go another step and constrain the rules by which beliefs must be updated. This is Diachronic Bayesianism. Diachronic Bayesians constrain rational belief by whether or not one changes their belief in accordance with evidence (on top of already having them cohere in the synchronic sense). This corresponds to a belief in ordinary logic that if you believe that P and that  P-> Q, you are rationally obligated to believe Q. This is also demonstrated with a Dutch Book argument (imagine a situation in which you believe that there is a 1% chance that a coin is rigged to always land heads, if you were asked to bet (out of $100) on whether the coin will land heads, you should at first bet ~$51, but if I then flipped the coin a thousand times and they all landed heads, you would be irrational to not start betting higher). This process of updating beliefs is called conditionalization and follows the form Pr(H) = Pr(H|e) upon being shown evidence e for hypothesis H. Note that these rules are markedly non-ampliative, but still allow for our inductive intuitions to be satisfied. This has the benefit of avoiding the problem of induction, but the drawback of not having much of a story about how we get our initial beliefs.

Some argue that since no one is consciously doing said probability math in their heads whenever examining any evidence, it is unfair to hold this as a meaningful picture of rationality. Do you think that is a fair argument? If not, how do we provide a readily applicable concept of Bayesian rationality to ordinary people?

Voluntarism (BvF’s view):

Van Fraassen’s voluntarism attempts to find a comfortable middle ground between these two views in epistemology by introducing our will in the formation of our beliefs. He argues that it is rational to hold beliefs that extend beyond evidence (use ampliative inferences) but he does not hold that beliefs that extend beyond evidence are rationally obligatory. Rather he believes that they are permissible, so, by our will, we may believe them. 

BvF defines four epistemic principles to lay out exactly how his view differs from the others.

  1. There can be no independent justification to continue to believe what we already find ourselves believing
  2. It is irrational to maintain unjustified opinions
  3. There can be no independent justification for any ampliative extrapolation of the evidence plus previous opinion to the future
  4. It is irrational to extrapolate ampliatively without justification

Belief in all four of these is tantamount to skepticism, since it says that we need to justify our beliefs but cannot justify our initial beliefs, nor can we justify extending our beliefs from our prior, irrational beliefs. Bayesians avoid this by denying premise 2. Since Bayesians do not concern themselves with justification, they can sustain having credences and changing them (according to evidence) without needing a story for what justifies them initially. Van Fraassen goes another step and also denies the need for justification for extrapolation (he rejects premise 2 and 4). Thus, van Fraassen believes that we are neither rationally obligated to justify our initial beliefs, nor are we rationally obligated to contitionalize our beliefs on the evidence. 

Now, not just any belief is permissible, but note the important distinction between rationality deciding what one can believe as opposed to what one must believe. The bounds for what one can believe are provided by three items. (1) Prior opinion (2) logical and synchronic (not diachronic!) coherence (3) does not a priori preclude the possibility of vindication.

Logical but nonetheless Irrational: Psillos’ arguments against Bayesianism

Before getting into Psillos’ arguments against voluntarism. Let’s note his argument for why Bayesianism leaves much to be desired for a theory of rationality.

  1. Bayesians believe that our initial beliefs need not be justified by evidence, only that they must be responsive to evidence so that credences can be recalculated
  2. It is irrational to believe a claim without evidence
  3. Bayesians are being irrational when accepting a new belief

What Psillos gets at with this argument is that the strict non-ampliative nature of Bayesianism leaves a gap in explanation for how and why one would go about acquiring new beliefs or deciding that a new piece of evidence is worth considering. He makes the example that there is a lot of evidence that the earth is round, and a little evidence that the earth is flat. Yet, one could, without transgressing any Bayesian rules of rationality, ignore all of the former and believe that the earth is flat. 

Is this a problem for the Bayesian? Can rationality only refer to how our beliefs hang together, or must it make reference to the content of our beliefs? In other words, can it be the case that any belief is rational so long as it is in the right context?

Saving the bacon or burning it? Psillos’ arguments against van Fraassen

Focusing on the third item with regard to how rationality bounds the range of possible beliefs, van Fraassen argues that our beliefs are constrained by our prior beliefs. These prior beliefs are sensitive to an external factor, namely truth. That is to say that even though we choose our beliefs, that does not change the fact that some beliefs can be right upon investigating the relevant facts about the world. Psillos argues that van Fraasen’s voluntarism makes the error of ignoring the content of belief, when it is still a necessary consideration for rationality.

  1. Some beliefs are decidably true or false and likely to be false upon investigation
  2. It is irrational to hold beliefs that, upon checking, are likely to be false
  3. Some beliefs are irrational to hold, regardless of other beliefs
This argument is a bit strange to me (and may be an incorrect reading) as it seems like it is no different from the call for synchronic coherence. Premise two is doing a lot of work by declaring the belief in question to be likely to be false. If we already believe that the belief is likely to be false, would we not be susceptible to a Dutch Book argument? This argument seems to assume that we can know beforehand how likely it is that a belief can be vindicated.

Psillos next argues that BvF’s view does not have an adequate story about how evidence can be incorporated into one’s belief-corpus.

  1. Having true beliefs aids us in our goals
  2. Not pursuing our goals is irrational
  3. Evidence aids in our ability to have true beliefs
  4. It is irrational to not conditionalize on evidence
We may say that it is unwise to not pursue our goals, but is it correct to say that it is irrational?

Psillos then defends IBE against an argument that van Fraassen provides. Van Fraassen argues that if one uses IBE they are susceptible to a Dutch Book situation, since they are providing an unfair probability bonus to a hypothesis H because it satisfies the “best” criterion, which may not necessarily connect to the probability of H being true. 

Psillos argues that one does not run into this problem if they apply IBE consistently before and after learning that H is the best explanation. 

Lastly, Psillos argues that BvF’s voluntarism permits obvious cases of irrationality. Namely…

  1. Cases where one disregards evidence that pertains to the truth or falsity of their beliefs.
  2. Cases where one changes between two permissible beliefs for arbitrarily reasons
  3. Cases where one could refute their own belief with another one of their beliefs
Are these cases irrational? Who gets to decide which evidence pertains to their beliefs? Who gets to decide which reasons are arbitrary?