Project concept map for TFCA project (Theoretical Foundations of Cognitive Agents) | Synthese (2009) 167: 125-143

Refer to TFCA-0-Index.md for an overview of the course.

Project assumes basic familiarity with Bayes' rule and conditionalization.

Motivation

Several authors point out that IBE and Bayesianism "are our two most prominent theories of scientific inference" (Weisberg 2009, 125). Idem Henderson 2017, openings pagina.

More motivation for compatibilism:

Part of the purpose of giving a Bayesian model of IBE is to elucidate the epistemic significance of explanatory considerations by connecting them to an established account of evidential support. (256-257)

Henderson 2014, refers back to Weisberg, arguing where IBE and Bayesianism could be mutually beneficial:

  • Benefit to IBE: Bayesianism provides a formal and widely useful framework for studying inference (cf. Weisberg 2009)
  • Benefit to Bayesianism: IBE may fill in some gaps that the formal theory does not strictly cover.
    • At this point Henderson does not elaborate, but later on (696) she for example mentions that Bayesianism is silent on what hypotheses are selected for the hypothesis space in the first place.

Poston 2014, also puts emphasis on where Bayesianism is silent (being a main proponent of IBE):

Yet the Bayesian structure is silent about explanation and its virtues. How do the explanatory virtues fit into a Bayesian story? An old methodological rule is to seek the most simple, powerful, and conservative explanation (i.e., the ‘best’ explanation) of some phenomena and to believe the best explanation when it is good enough. [ejw: ref to Harman 1965 here]. It is not apparent from the Bayesian apparatus why the norm to seek good explanations should be reflected in a Bayesian view of inference. (Poston 2014, 149)

Weisberg 2009:

I suspect that the explanationist has much to offer the Bayesian, but the potential benefits of a marriage of the two views are lost when we make IBE subservient to Subjectivist Conditionalization. If we grant IBE primacy instead, and use it to shape a more objective Bayesianism that rejects Subjectivist Conditionalization, we are in a position to develop a Bayesianism that is free of the limitations that come with Subjectivist Conditionalization. Such a view would be a truer friendship between the explanationist and the Bayesian than the heuristic approach to compatibilism, since each party would stand to benefit from the other. The explanationist can help the Bayesian fill in the gaps in her view, by providing a way to construct priors for new theories, for example. And the Bayesian provides the explanationist with a powerful and well-studied formal framework in which to formulate the details of her view. (Weisberg 2009, 136)

Henderson-Bayesianism_and_IBE.md also has a nice motivation: Bayesianism fits confirmatory aspects of scientific inference, IBE fits explanatory aspects of scientific inference. They both are applicable to many cases of scientific inference which suggests they should be broadly compatible.

Non-trivial compatibilism

The challenge for a good compatibilistic account is to do so without making IBE trivial. (Bayesianism is a well-established formal framework that can be presented weakly (subjective) or strongly (objective), but whose core of conditionalization you cannot strip out).

E.g. Lipton argues we should not reduce IBE to interpreting "best" only as the most probable.

Weisberg (2009) argues that IBE is only made compatible with Bayesianism by making it "subservient to conditionalization in a way that robs IBE of much of its substance and interest." (Weisberg 2009, 125) Arguing, instead, that we require an objective Bayesian account to take in IBE's (normative?) claims.

Epistemic attitudes

Henderson (2014) shows that the compatibilist debate assumes that you at least allow explanationist to have degrees of belief on hypotheses, rather than being fully committed to full beliefs. I.e. the debate assumes that it is at least be possible that Bayesianism and IBE address similar epistemic concerns are are not essentially different.

Given that they do, Van Fraassen's position is "that IBE is bad epistemology because violating conditionalization is irrational" (125).

Related point: I've read several arguments (can't recall where exactly, but I think Glass) that point out that Bayesianist are in no way committed to actually accepting any inference (I think this is what is meant by "full" belief) based on their degrees of belief. (See next point)

Bayesianism does not actually entail inference

Bird (109) and ... I think Glass?

It is good to note that strictly speaking the Bayesian framework only provides probabilities over hypotheses in the given hypothesis space and does not actually tell you which hypothesis to infer.

In the typical interpretation of IBE, it is in fact an account concerning (whether it be descriptive or normative ) of inferring a hypothesis (e.g. the most 'lovely' hypothesis).

Extending Bayesianism with a rule for inference has its own problems. (Identify those). E.g. the most straightforward problem is indeed that the most probable explanation is not necessarily the most lovely. It could be a hypothesis that is trivially true but adds no understanding whatsoever.

Yet the Bayesian structure is silent about explanation and its virtues. How do the explanatory virtues fit into a Bayesian story? An old methodological rule is to seek the most simple, powerful, and conservative explanation (i.e., the ‘best’ explanation) of some phenomena and to believe the best explanation when it is good enough. [ejw: ref to Harman 1965 here]. It is not apparent from the Bayesian apparatus why the norm to seek good explanations should be reflected in a Bayesian view of inference. (Poston 2014, 149)

Qualitative vs. quantitative beliefs

In Weisberg's (normative) reading of IBE: "IBE is a claim about what we ought to believe, but bare bones Bayesianism has nothing to say about qualitative belief; it deals solely in degrees of belief." (Weisberg 2009, 127)

A related point may be (Lipton's? or Glass's?) remark that identifying the hypotheses with the highests posterior in no way is a prescription for actually inferring that hypotheses (whereas this is IBE's goal).

Weisberg 2009, 127 continues... There are two options:

  • Introduce a qualitative notion into Bayesianism
  • frame IBE in terms of degrees of belief

However, several authors reject the idea of Bayesian qualitative beliefs, e.g. Jeffrey 1970, Christensen 2004. (see Weisberg 127)

Internalist and externalist reading of IBE

Two ways of understanding IBE (Weisberg 2009, 131):

  • "internalist": you should prefer the explanation that you judge better
  • "externalist": you should prefer the explanation that is in fact better

Explanatory virtues

Common denominator: "It is often suggested that a theory provides the best explanation if it possesses the optimal combination of explanatory virtues in relation to the phenomena, where explanatory virtues include considerations such as simplicity, unification, scope, and fruitfulness." (Henderson 2014, 689-690)

The explanatory appeal of IBE is "its agreement with common sense and paradigm instances of good scientific inference" (Weisberg 2009, 129). One may therefore argue there's some objectivity to the theory of IBE that subjective Bayesianism in the heuristic view does not sufficiently acknowledge (allowing for rational inter-subjective disagreement).

Why do explanatory virtues matter? We can see the relevance of explanatory virtues by sketching cases where IBE and subjective conditionalization may diverge: despite the "rational" procedure of subjective conditionalization, we have a dominant gut feeling about which one is the better explanation; and we can also usually indicate why that is the case in terms of explanatory virtues.

Weisberg (2009) gives a nice example on page 130 about the most likely explanation of finding your house with the front door open, broken lock, valuables missing, mess in the living room.

Incompatibilism

For a sketch of Van Fraassen's position, see "Van Fraassen: IBE contradicts Bayesianism" in Henderson-Bayesianism_and_Inference_to_the_Best_Explanation.md .

Henderson argues that Van Fraassen's sketch of IBE in terms of "boosting" probabilities is not the only possible explanation of IBE in a probabilistic context. Van Fraassen thus seems to provide a dilemma between choosing IBE or Bayesianism, under the assumption that IBE necessarily violates the rational axioms of probability (conditionalization, specifically). This dilemma may be false and the task is to find another interpretation of IBE in a probabilistic context (e.g. explanatory virtues play a role on prior determination, but do not involve boosting).

Truth-tracking

We need "a correlation between higher posterior probabilities and the truth" (Bird, 112)

Cf. Glass.

Are explanatory considerations descriptive or normative?

Henderson: the issue is "whether the role of explanatory considerations is seen as purely descriptive or also normative" (Henderson 2014, 693)

This is actually a quite important issue! Multiple other relevant conceptual distinctions depend on it.

  1. Descriptive -> IBE is a heuristic for what Bayesianism normatively prescribes

If you argue explanatory considerations are descriptive (e.g. Lipton, 2004) then you are likely to commit to the heuristic view of IBE, where explanatory considerations actually guide people's hypothesis selection process in a way that approximates what Bayesianism prescribes normatively. The strength of this position is that is accommodates the fact that people are shitty Bayesians, i.e. their actual behavior diverges from what Bayesianism prescribes as the rational thing to belief/do.

  • Lipton 2004
  • Okasha 2000

Okasha, for example, says that our explanationist inferences use “explanatory considerations as an aid to calculating the priors and likelihoods needed to apply Bayes’s theorem,” (Okasha, 2000, p. 703) and Lipton and McGrew make similar pronouncements. (cited in Weisberg 2009, 132)

An alternative way of phrasing the descriptive account of IBE is to say that, insofar as IBE has any normative power, it is derived from being an approximation of Bayesianism (i.e. in the subjective account, the correct application of subjective conditionalization).

  1. Normative -> Objective Bayesianism

If on the other hand you argue that explanatory considerations are normative, then you almost certainly have to commit to objective Bayesianism because then IBE will have to impose normative constraints on priors and likelihoods (meaning: for a correct application of Bayesianism these are the right priors and likelihoods). Example: Weisberg, 2009.

I suspect that the explanationist has much to offer the Bayesian, but the potential benefits of a marriage of the two views are lost when we make IBE subservient to Subjectivist Conditionalization. If we grant IBE primacy instead, and use it to shape a more objective Bayesianism that rejects Subjectivist Conditionalization, we are in a position to develop a Bayesianism that is free of the limitations that come with Subjectivist Conditionalization. Such a view would be a truer friendship between the explanationist and the Bayesian than the heuristic approach to compatibilism, since each party would stand to benefit from the other. The explanationist can help the Bayesian fill in the gaps in her view, by providing a way to construct priors for new theories, for example. And the Bayesian provides the explanationist with a powerful and well-studied formal framework in which to formulate the details of her view. (Weisberg 2009, 136)

Whether IBE is descriptive or normative, leads to different views on what it means to constrain Bayesianism. Do explanatory considerations guide or determine the right choice for prior and likelihood? It is good to note that in the case of an objective Bayesian, explanatory considerations are constraints in addition to the restrictions that were already in place on the choice of priors and likelihoods.

There is a similar point w.r.t. Bayesianism: does it describe how people ought to behave, or how they actually behave?

The obvious point here is that people are actually really shitty Bayesians (e.g. consider Kahneman), so usually the Bayesian framework is seen as normative.

This point is relatively interesting for those who apply Bayesianism in cognitive science; it may be perceived to have descriptive aspirations there. Consider for example the critique of Bowers and Davis on Bayesian "just so" stories .

Contraint-based vs emergent compatibilism

Henderson makes the distinction between constraint-based compatibilism and her own position of emergent compatibilism.

Cited as proponents of constraint-based compatibilism are:

Subjective and heuristic view: Lipton, Poston. The paper of Alexander Bird contains a great overview (but he himself ultimately argues for objective Bayesianism)!

  • NOT IN READING LIST YET Okasha, S. [2000]: ‘Van Fraassen’s Critique of Inference to the Best Explanation’, Studies in History and Philosophy of Science, 31, pp. 691–710.
  • Lipton 2004 (the book)
  • Williams (see Henderson 2017, 250)
    • "Local" view: i.e. only adjust priors in special circumstances for specific hypotheses

Q: What constraint-based form is not also immediately the heuristic view? Any examples?

Objective:

  • Weisberg 2009: Locating IBE in the Bayesian framework

Subjective vs. Objective Bayesianism.

See "## Bayesianism (minimal, subjective, objective)" in Henderson.

When it comes to the matter of compatibilism with IBE, note that those who see IBE as normative (e.g. Weisberg) are committed to objective Bayesianism.

Henderson is sort of an outside category with her emergent compatibilism.

But Bird is also interesting, because he argues that not only subjective, but also objective Bayesianism is still too weak for compatibilism. Afaik he is however not anti-compatibilist, but rather one with an extremer form/interpretation of objective Bayesianism? -> See p.112: "'super-objective' Bayesianism"

My general impression is that Bird requires a "super-objective" Bayesianism because he has a very strong account of IBE. So he is on the complete opposite spectrum from those Bayesians who make compatibility easy by making IBE too trivial.

Plausibility vs credence

See Bird's "super-objective" position that drops personalism. Apparently, Salmon also does something in that direction.

Working with plausibility has some advantages. Probabilities of 0 and 1 do no longer involve problems of people accepting crazy bets on the truth of a proposition.

What more?

Relevant explanatory virtues

Priors:

Lipton: "explanatory considerations of unification, simplicity, and their ilk" (Lipton 2004, p. 115)

"simplicity, elegance, unification" (Weisberg 2009, 132).

Emergent compatibilism

I'm increasingly impressed by this position. It does some things right, in my opinion.

  • It's the only theory so far that clearly accounts for the difference between core theories and auxiliary hypotheses, which is a relevant conceptual distinction in the philosophy of science
  • An additional advantage may be that this potentially better counters criticasters of Bayesianism that you can fit anything basically by choosing your priors right.
    • Henderson instead heavily stresses on "natural" assumptions, also on the priors.
    • No "extra" constraining/boosting -> less arbitrary
    • Auxiliary assumptions are punished in the likelihood! So also a 'natural' answer to adding complexity to the theory.
    • Whereas, when constraining the prior on specific fitted theories, how does one successfully 'justify why the prior is higher for more explanatory specific hypotheses' (256)?
  • Allows discussion of less trivial examples (Henderson started this, but more work needs to be done)

The first component, as we saw above, naturally reflects the explanationist concern with fine-tuning since it involves an averaged likelihood. By contrast, in Bayesian reconstructions which work with fitted models, explanatory considerations can be grafted onto the Bayesian machine by having them constrain the priors, but it remains rather opaque why they, rather than some other considerations, should be what determines the priors. (Henderson 2017, 256)

You can still make assumptions on the priors, for example apply the Principle of Indifference. So emergent compatibilism is compatible with objective Bayes!

Hendersons stresses multiple times that for equal priors, hierarchical Bayes can perfectly fine account for an explanatory difference between theories due to the amount of fine-tuning required for fitting a general theory.

But the main point is: compatibilism of IBE and Bayesianism does no longer depend on the (possibly arbitrary) choice of priors! (Henderson 2017, 256)

Bayes-IBE Position Matrix

Would be interesting to make a 2x2 matrix, with constraint-based vs emergent, and subjective vs. objective compatibilism. Maybe this is of limited interest if it turns out Henderson is the only one advocating emergence-based compatibilism.