Explanatory Bayesianism and Inference to the Best Explanation

Background

Deductive reasoning is quite well understood, but the verdict is still out on a well-rounded theory of abductive reasoning. This is especially interesting because science often proceeds through abductive reasoning from observed evidence to a hypothesis that would best explain that evidence, if the hypothesis is indeed true. This type of reasoning is studied as "inference to the best explanation" or IBE [@Harman1965]. The central claim of IBE is that "explanatory considerations are a guide to inference" [@Lipton2000, p.184, my emphasis], meaning that a hypothesis is not just inferred and then explains some evidence, but that in fact it is inferred precisely because it explains the evidence.

Any complete theory of IBE should offer some methodology for choosing one hypothesis over another, but should also give an account of why the chosen explanation is in fact better [@Glass2012, p.413]. The Bayesian approach to scientific inference seems to be a natural candidate for providing an IBE account, because Bayesian conditionalization can express degrees of belief in various hypotheses, given (new) evidence. But although Bayesianism offers a clear method for comparing hypotheses, it has no explicit account of what explanatory virtues good explanations have.

Various authors argue that even though IBE and Bayesianism could be made compatible by interpreting "the best" as "the most probable", this would be at the cost of making the account of IBE trivial, as it loses its ability to describe what explanatory virtues guide the inferences that scientist make [@Lipton2000; @Glass2012]. Peter Lipton has argued that scientists do not straightforwardly choose the most probable explanation, but the most lovely one, i.e. explanations that add the most understanding [@Lipton2000]. This raises the question if there is a Bayesian account of IBE such that it does not make IBE trivial, but maintains its core "explanationist" idea.

To that end, I want to explore the various attempts that have been made to provide an "explanationist Bayesianism" under the working assumption that "the Bayesian and the explanationist should be friends" [@Lipton2001, 109-119; @Lipton2004a, 120]. Within the compatibilist perspective, and often in response to Lipton, there is a variety of views on exactly how they should be friends [e.g. @Hitchcock2007; @Ladyman2005]. Henderson suggests a general division into "constraint-based compatibilism" and "emergent compatibilism" [@Henderson2017, p.249-250].

The constraint-based strategy investigates where explanatory considerations and virtues can play a role in the Bayesian framework. Proposals include using explanatory considerations for likelihood judgements, "assessing the priors", "determining relevant evidence" and "guiding hypothesis construction" [@Lipton2004a, p.119]. A related discussion is whether IBE is merely a heuristic that approximates the outcomes of Bayesian probabilistic reasoning, because explanationists may argue that IBE can reasonably lead to true knowledge [@Bird2017, p.104-111]. This matters because compatibilism with IBE may then require objective rather than subjective Bayesianism [@Weisberg2009; @Iranzo2008]. For example, Poston argues that simple theories should receive higher priors than complex ones based on objective considerations, in order to avoid inductive scepticism [@Poston2014, p.174-179].

Emergent compatibilism instead suggests an alternative strategy, namely to show how the Bayesian framework itself already favors better explanations based on inherent properties. For example, rather than fine-tuning priors based on explanatory considerations, one may point out that Bayesian model selection already penalizes more complex hypotheses (simplicity being an explanatory virtue) [@Henderson2017]. Another measure could be the amount of fine-tuning required to make the general Bayesian framework fit the given data, where less fine-tuning indicates better explanations [@Henderson2014].

Yet another strategy may be to nevertheless redefine IBE, but without making it trivial. Cabrera for example argues that whereas Bayesianism assigns credences to hypotheses, IBE tells us when to accept and infer a hypothesis. In that sense, IBE and Bayesianism may be seen not only as compatible but as complementing each other [@Cabrera2017].

Given that inference to the best explanation is a core theme of the course, this specialization project is a good fit. More specifically, this project combines philosophical considerations on IBE with the main framework that is studied in this course, namely Bayesianism.

Research question

The proposed specialisation project is an in-depth literature study that addresses the following research question:

What are the current main proposals for making Bayesianism compatible with IBE, without making IBE trivial?

The project is thus specifically intended to study the efforts of those who argue IBE and Bayesianism can be made compatible. Of course, compatibilists often respond to arguments against compatibility, e.g. arguments from Van Fraassen, but these arguments are considered only insofar as they are relevant for understanding the compatibilist positions. This is done to limit the scope of the project and maintain an appropriate focus. Additionally, the research question reflects that a main challenge for compatibilists is to show the connection between IBE and Bayesianism without making IBE a trivial account, as explained in the background section. Finally, the research question is appropriate for a literature study, because the focus lies on mapping the current state of a debate in a specific academic subfield.

Learning objectives

The learning objectives of this project are to:

  • Explain the core ideas of IBE and the relevant aspects of Bayesianism

In order to understand why and how compatibilists connect IBE and Bayesianism, the respective accounts must first be sufficiently clear independent of each other. Moreover, otherwise it cannot be clear how IBE becomes trivial when "best" is interpreted as "most probable".

  • Differentiate the main academic positions arguing for the compatibilism of IBE and Bayesianism

This is necessary in order to gain insight in the relevant academic debate and answer the research question. What personally motivates me here is the interdisciplinary nature of this research field, because I have an interest in philosophy next to my interest in AI. Combining both interests is a plus.

  • Summarize the main arguments for these respective positions

In order to compare and evaluate the differentiated academic positions it is important to consider how they are supported.

  • Evaluate the merits of and challenges for these respective positions and arguments.

This objective facilitates being able to give an informed opinion on explanatory Bayesianism. Additionally, this helps to get a sense where future work is needed. Are there inconsistencies to be resolved or gaps to fill?

The output of the project will be a literature review, which is a suitable format given these learning objectives. A provisional literature list is provided at the end of the proposal. Note that the above objectives also roughly describe the temporal steps of the project. The following section contains a more detailed study plan for meeting these learning objectives.

Study plan

The bulk of the project work is reading papers and book chapters, and making notes of them. I generally make quite extensive notes while reading, so as a rule of thumb I estimate to read about 10 pages per hour. The preliminary literature list contains 329 pages (including bibliographies), which boils down to around 300 pages and 30 hours of reading and note taking. The literature list contains a few lengthy book chapters of which sometimes only a subpart focuses on IBE in relation to Bayesianism. It also contains some papers I have already read. In practice, there should therefore still be some space to include highly relevant literature if I encounter it during the project. I do however think my reading list is already fairly complete.

Reading and writing often go hand in hand for me and I do them in parallel, so I attribute relatively little time for dedicated writing at the end of the project. Part of the writing effort is included in the note taking effort. The time schedule therefore contains several written deliverables throughout the project, not just at the end.

At the beginning of the project I'll prioritize book chapters and papers providing a comprehensive account of the main issues related to compatibilism, in order to get a general overview of the academic debate. This should already help me identify the main positions in the debate, so that I can focus my reading efforts in the rest of the project. After having a clear idea of the main positions in the debate, I'll focus on papers working out more fine-grained issues. The focus then shifts from mapping standpoints to reviewing arguments for and against those standpoints.

Allocated time

Reading + taking notes (30h):

  • Around 300 pages of literature
  • Based on the conservative speed of 10 pages per hour
  • Includes the possibility to incorporate new literature

Writing literature review (20h):

  • Introduction & research question (2h)
    • The proposal already forms a template for this section
  • Background on IBE, Bayesianism, compatibilism (5h)
  • Main section: discuss most important papers (10h)
  • Conclusion and reflection on learning objectives (3h)

Time schedule

The project is planned to take about 50 hours, distributed over 7 weeks. To compensate for the exam weeks, I'll instead distribute the 50 hours over 6 weeks, meaning that I will spend approximately a full workday per active week on the project. I do not specify a detailed order for reading papers in the time schedule, but will follow the strategy as specified above.

Week Dates Description
W21 18-24 May Read overview papers to
get an outline of the debate
W22 25-31 May Introduction + Research Question
W23 1-7 June Background section:
explain IBE and Bayesnianism
first sketch of main positions
W24 8-14 June Read and write about central papers
W25 15-21 June Read and write about central papers
W26 22-29 June Exams
Revise work done so far
W27 29-3 July Conclusion + Evaluation Learning Goals
Finalize