Project Proposal | Brainstorm & Roadmap
Formal remarks
Evaluation scheme
- Background: 4 points
- General description of research area (2)
- References (1)
- Embedding w.r.t. course goals (1)
Remarks: Describe the general topic / area (including references) and explain how it fits within the course (e.g., with respect to abduction, relevance, stochastic computation, Bayes in Cog(Neuro)Sci etc.) This typically takes 1/2 to one page.
- Research question: 2 points
- Motivation (1)
- Clarity, preciseness (1)
Remarks: Make sure the research question you want to investigate is explicit and precise. Both for literature research as well as projects. See e.g. https://writingcenter.gmu.edu/guides/how-to-write-a-research-question
- Learning objective: 2 points
- Motivation of objectives (1)
- Are objectives SMART? (1)
- Specific, measurable, attainable, realistic, time-dependent
Remarks:
Part of the course goals is to train you to set your own learning objectives and reflect on them. See e.g. https://teachonline.asu.edu/2012/07/writing-measurable-learning-objectives/
Goal of project proposal
Goal is to formulate your own learning goals
- What do you want to study? (background and RQ)
- How you will achieve that (SMART objectives)
- How you will allocate the available time (SMART)
- How you will interpret results etc.
N.B. the project proposal does thus not necessarily have to be a research proposal (more experimental approach). It could be a requirement analysis (software project), or a plan for a literature study (literature study).
In any case you need:
- background description of the topic with proper references
- plainly stating what you want to do.
- and having a clear plan of work (with timeschedule).
Possible formats
- Extensive literature study
- Philosophical or computational issue, in-depth
- Output: Review paper and presentation.
- Study a computational model:
- a) algorithm design
- b) computer implementation
- c) simulation
- Output: working program and report
- Formalization of conceptual idea discussed in course and some exploration of it
- Output: exploration/experiment e.g. with Pepper robot.
- Intersection with other courses, suggestions:
- parameterized complexity
- learning algorithms
- societal impact of autonomous agents
Possible topics
What topics did I find relatively interesting so far?
Intuitively, something with the philosophy block ór the Bayesian block.
- Abuction problem: making sense of observations / finding new explanations
- Inference to the best explanation
- Subgoal could be to make an effort to map the relevant concepts in the literature; abduction and IBE tend to be conflated.
Questions related to abduction:
- What is an explanatory inference?
- What is "best" in "inference to the best explanation"
- Zie Glass 2012 uit syllabus; doet ook computer simulations.
- How do we generate candidate hypotheses?
- How do we evaluate and compare hypotheses?
-
How do we decide what information is relevant
- Literature overview: measures of relevance.
-
Philosophical issues related to computational modeling?
Something related to the interpretation of subjective probability.
Uncertainty - Foundations of subjective probabilities:
- "Where do subjective degrees of belief come from?"
- "How should we update beliefs in light of new evidence?"
- Something with Bayesianism in cognitive science and neuroscience
Uncertainty - stochastic computations
- Stochastic computations: how can we use noise?
Negative:
- No complexity analysis: too dry for me, and I miss a bit of background.
- We did not have the computational model lectures yet, so I'll not consider those.
Example research questions
- How did researchers (like Jaynes) approach ( Bertrand ) Russell's paradox?
- What is the current status of this paradox?
- Can I defend my own informed opinion on whether this has been solved to satisfaction?
- To what extent is 'explainable AI' generic versus sensitive to cultural differences? Is XAI subject to gender/ethnic/other biases?
- What is the relation between 'insight' (as in insight in problem solving) and the frame problem? What could cause the apparent individual variance in 'flexibility' in one's frames?
- What is known about the relations between BPP and PP and the complexity classes P and NP? Why do people believe P = BPP and what evidence is there for this belief?
- How can we de-randomize stochastic computations and when can this be done tractably?
- What other notions of 'best' have been approached to 'best = most probable' in the context of Bayesian models? What are the pros and cons of each alternative?
Brainstorm
TODO: make sure to count the number of pages of the literature referred to in the proposal.
- Relevancy and abduction
- Explanatory virtues; Lipton paper mentions they are not understood very well; no consenus (Lipton p.6)
- Ik zou dus kunnen kijken naar de status van dit debat
- Zijn er pogingen tot formalisatie?
Pearls "Theoretical Impediments to Machine Learning"
- Check responses to Pearls "Theoretical Impediments to Machine Learning"
- Many papers just seem to refer to the causal hierarchy (see second bullet)
- Association; intervention; counterfactuals;
- What is the current debate?
On Reddit: https://www.reddit.com/r/MachineLearning/comments/7qgwue/r_theoretical_impediments_to_machine_learning/
Discussion focuses on whether Pearl's claim that statistical "model-free" learning cannot learn causal relations. Some people are not convinced by that and mention various deep learning/NN-models that perhaps do learn causal relations.
Raven paradox
Bayesian answers to the Raven paradox.
Raven paradox and relevancy: do you gain information about ravens by looking at a white shoe?
What evidence is relevant in an inductive reasoning process (measured according to degree of confirmation?)
Wikipedia: "To confirm the statement 'All ravens are black', it would be necessary to observe all ravens. This is difficult but possible." But if logic has a normative role, does that statement not also have implications for all future ravens?
Doelen:
- What is the relevance of the raven paradox?
- Historical reception of the paradox; main positions
- Is the Raven Paradox solved?
Literature:
- Original paper of Hempel
- In response to earlier view of Jean Nicod (1930) that observing a black raven is evidence for the statement "all ravens are black"
- Wikipedia entry
-
- Okay there is a SHIT ton of responses to the Raven Paradox.
- Ik zou kunnen focusen op de "Standard Bayesian Response"
-
Good, I.J., "The white shoe is a red herring," British Journal of Philosophy of Science, 17 (4), 1967, p. 322.
- https://academic.oup.com/bjps/article-abstract/17/4/322/1516288?redirectedFrom=fulltext
- Vroege refutatie; Good verwijst ook naar een eerdere contributie met een oplossing
- "De crux van het voorbeeld is dat de beoordeling van een hypothese altijd moet geschieden in concurrentie met andere hypothesen die (ook) zodanig zijn ingericht dat de waarschijnlijkheid van de waarnemingen bepaald kan worden." (Wikipedia)
- Oftewel: een Bayesian framework dat wél hypotheses tegen elkaar afweegt is geschikter en zou de paradox moeten oplossen.
- Antwoord van Hempel -> The White Shoe: No Red Herring
- Weer antwoord van Good: "The White Shoe Qua Herring is Pink"
David A. Freedman - Black Ravens, White Shoes, and Case Selection
To summarize, the illusion of paradox is created by blurring the distinction between the sample and the population. The illusion is dispelledby deciding whether we are discussing the data at hand, or extrapolat-ing from the data to a larger population—although, in the second case, asampling model is needed.
De Bayesian solution accepteert de tegenintuitieve conclusie, maar stelt dat het gewicht van een niet-zwarte-raven observatie extreem klein is; dat hangt wel van de ratio zwarte raven en alle niet-zwarte-raven objecten af. Het artikel van Freedman biedt ook een antwoord dat afhangt van de ratio tot de algehele populatie, geloof ik. Dus het is relevant, maar nauwelijks relevant (in alledaags taalgebruik zou ik het alsnog irrelevant noemen).
Wikipedia, Carnap's variant:
Hence, from the Bayesian-Carnapian point of view, the observation of a non-raven does not tell us anything about the color of ravens, but it tells us about the prevalence of ravens, and supports "All ravens are black" by reducing our estimate of the number of ravens that might not be black.
De hedendaagse discussie gaat dan meer over de voordelen en nadelen van de Bayesian aanpak:
- A Solution to the Raven Paradox: A Redefinition of the Notion of Instance
- A Conciliatory Answer to the Paradox of the Ravens
- Peter Vranas (2004) - Hempel's Raven Paradox: A Lacuna in the standard Bayesian Solution
-
Fitelson and Hawthorne (2010) - How Bayesian Confirmation Theory Handles the Paradox of the Ravens
-
Susanna Rinard - A new Bayesian Solution to the Paradox of the Ravens
What is "best" in inference to best explanation
Zie ook: laatste example vraag. Ik zat in zo'n richting te denken:
-
What other notions of 'best' have been approached to 'best = most probable' in the context of Bayesian models? What are the pros and cons of each alternative?
- Subquestion, see next subsection: is there an alternative that can also be a "lovely" explanation?
-
Inzoomen op Likely vs lovely?
-
Under which understanding of "best" is the best inference/explanation truth tracking?
How compatible is IBE with Bayesianism?
Background reasoning could be something like:
- Overarching question: how do scientists pick a particular explanation/hypothesis to explain the evidence.
- See Glass and Lipton for "requirements" of IBE. More or less: 1) How do they choose one explanation over another competing explanation? 2) What makes one explanation better than the other? What explanatory features makes up a good explanation?
- "Best = most probable" as a straightforward answer
- Advantage of Bayesian confirmation theory is that it always considers hypothesis in a contrastive settings
- This is the core idea behind many Bayesian solutions to Hempel's paradox.
- Lipton and Glass: IBE with best=most probable -> push to triviality!
- Accommodates the second requirement (why one better than the other) but not the first (it does not say anything, descriptively, about explanatory features; what are the characteristics of a good explanation?)
- Introduce compatibilist vs incompatibilist position; relate to the criticism on best=most probable
- Look up Van Fraassen 1989
- We focus on attempts to make Bayesianism and IBE compatible.
- Challenge is to show: Likelihood an important features of lovely explanations; but loveliness is guiding in IBE. How exactly? What's the exact relationship between Bayesianism and IBE?
- Two possible strategies:
-
- Find ways to include explanatory considerations into Bayesian models
- E.g. explanatory constraints on priors
- Criticism: are explanatory constraints on priors are form of fine-tuning model for desired outcomes? Hacky?!
-
- Alternative position is "emergent Bayesianism"; show that Bayesian models themselves already favor the more lovely explanation (again, see below under "Henderson")
-
First attempt at RQ: what are, in the current debate, suggestions for making the Bayesian account of IBE address requirement 1.
- Learning goal: map current debate
- In particular the relevant approaches in the "compatibilist" camp
- Understand the merits and downsides of these solutions; e.g. are explanatory constraints on priors are form of fine-tuning model for desired outcomes? Hacky?!
- Learning goal: understand (quantify this!) the merits and downsides of a Bayesian IBE account
Literature
TODO Henderson2014 ziet er top uit.
Incompatible position:
- Van Fraassen (1989, p: 160-70) -> B. C. Van Fraassen. Laws and Symmetry. Clarendon Press, Oxford, 1989.
- Is specifically a critic of IBE; of compatibility with Bayesianism by extension, I guess.
- Salmon, W. C. 2001. Explanation and confirmation: A Bayesian critique of inference to the best explanation. In G. Hon and S. S. Rakover (Eds), Explanation: Theoretical Approaches and Applications, pp. 61–91. Dordrecht: Kluwer.
- N.B. heeft een response van Lipton
- Plus weer een response van Salmon op Lipton.
- TODO: wat is de controverse in deze woordenwisseling?
Compatibalism:
- "Best Explanations: New Essays on Inference to the Best Explanation"
- Contains at least two articles explicitly about Bayesianism:
- Alexander Bird: Inference to the Best Explanation, Bayesianism, and Knowledge, 97-120. In "Best Explanations: New Essays on Inference to the Best Explanation"
- Leah Henderson: Bayesianism and IBE: The Case of Individual vs. Group Selection. In "Best Explanations: New Essays on Inference to the Best Explanation"
Met referentie:
- Hitchcock, C. (2007). “The lovely and the probable,”Philosophy and Phenomenological Research,74(2): 433–440.
- L. Henderson (2014). Bayesianism and inference to the best explanation. British Journal for the Philosophy of Science, 65: 687–715, 2014.
- T. Poston. Reason and Explanation: A Defense of Explanatory Coherentism. Palgrave Macmillan, Basingstoke, 2014.
- P. Lipton. Inference to the Best Explanation. Routledge, London, 2nd edition, 2004.
- J. Weisberg. Locating IBE in the Bayesian framework. Synthese, 167: 125–44, 2009.
- James Ladyman: Wouldn't it be lovely: explanation and scientific realism. Metascience (2005) Review of Peter Lipton (2004)
- Valeriano Iranzo: "Bayesianism and inference to the best explanation"
- Cabrera, F. Can there be a Bayesian explanationism? On the prospects of a productive partnership. Synthese 194, 1245–1272 (2017). https://doi-org.ru.idm.oclc.org/10.1007/s11229-015-0990-z
-
Harman, G. (1965). “The Inference to the Best Explanation,”Philosophical Review, 74: 88–95.
-
Mackonis, A. (2013). “Inference to the best explanation, coherence and other explanatory virtues,” Synthese, 190(6): 975-995.
- Myrvold, W. (2003). “A Bayesian Account of the Virtue of Unification,” Philosophy of Science70: 399-423
Kan geen toegang krijgen:
- Psillos, S.(2004). “Inference to the Best Explanation and Bayesianism,” in Stadler (ed.), Induction and Deduction in the Sciences.Dordrecht: Kluwer, 83–91.
Toch niet relevant:
- Douven, I.and Schupbach,J.(2015). “Probabilistic alternatives to Bayesianism: the case of explanationism,”Front. Psychol.,(6):459, doi: 10.3389/fpsyg.2015.00459.
- Douven, I. and Wenmackers, S. (2015). “Inference to the Best Explanation versus Bayes’s Rule in a Social Setting,” Brit. J. Phil. Sci., in press.
Henderson
Explanationist Bayesianism: how can explanatory considerations be brought into a Bayesianism account of IBE? (in a feasible way)
From Henderson, 248, on how to bring into play "explanatory considerations"
Various philosophers have then recommended what might be called ‘explanationist Bayesianism’, according to which (depending on the version) it is either de facto the case, or reasonable, or even essential that explanatory considerations play critical roles in the Bayesian machinery (Lipton 2004; Weisberg 2009; Poston 2014).
- Lipton 2004:
- Weisberg 2009:
- Poston 2014:
Henderson distinguishes between (see p. 249)
-
"constraint-based compatibilism" (to which 'explanationist Bayesianism' belongs because it e.g. the choice of priors is guided/constrained by explanatory considerations)"; and
- "a dominant approach to reconciling the two paradigms has been to recommend that Bayesianism and IBE should be amalgamated into an explanatorily driven form of Bayesianism. Minimally, this position requires finding roles that explanatory considerations could play in the Bayesian framework." (Henderson, 250)
-
'emergent compatibilism'; the idea that the Bayesian framework already penalizes bad explanations (?)
- "For natural choices of priors, the Bayesian method of calculating likelihoods for general frameworks of theories inherently penalizes such reliance on fine-tuning." (249)
- Then IBE emerges from the Bayesian account; rather than Bayesianism only being an IBE account when constrained.
- "A stronger version requires defending the idea that explanatory considerations are somehow critical or essential for the operation of Bayesianism." (Henderson, 250)
What's the exact relation between Bayesianism and IBE?
The fact that both Bayesianism and IBE appear to be broadly applicable to cases in the history of science suggests that they are compatible with one another. But then the puzzle is how to characterize their relationship. How does IBE fit in with the evaluation of theories in terms of their empirical support? Is IBE just another way of talking about empirical support? Or is it an additional component which needs to be included in a full account of theory evaluation? (Henderson, 250)
Explanatory constraints on priors, debate:
Ted Poston, for example, argues that if more explanatory hypotheses are not given higher priors, inductive scepticism follows (Poston 2014). Others have advocated that simpler hypotheses should be given higher priors (e.g. Jeffreys 1998 [1939]), though there has also been serious opposition to this view (e.g. Kelly 2007). The basic problem, critics urge, is that we have no reason to think the world is more likely to be simple than complex, so no grounds for making that assumption in the prior. (Henderson 250)
Henderson's own contribution: favor an emergent approach; using that Bayesian model selection inherently favors simpler explanations; motivation for Bayesianism as IBE thus can be found within Bayesianism itself.
The principles at work here are those that underpin the Bayesian approach to ‘model selection’, which is known to favour simpler hypotheses (MacKay 2003).
Bird
Bird begins with a clear distinction between normative and descriptive accounts.
Inference to the Best Explanation, Bayesianism, and Knowledge
Describes Standard Bayesianism, which includes personalism i.e. that evaluations of hypotheses is a matter of credences
Some distinctions:
Standard Bayesianism is normative, a view that asserts requirements about how reasoning ought to proceed, in addition to a descriptive kernel (personalism). A standard Bayesian is also a subjectivist if she holds that probabilism and conditionalization are the only rational constraints on credences. An objective Bayesian extends standard Bayesianism with further constraints, such as a principle of indifference.2 Some Bayesians are non-standard in that they do not accept all the features of standard Bayesianism. For example, one might accept that there are counterexamples to conditionalization. (98)
Difference between IBE and Bayesianism may be between description and normative accounts:
Descriptive explanationism I take to be the view that people, scientists included, often hold the explanatory character of a hypothesis to be relevant to its epistemic evaluation. Inference to the Best Explanation (IBE), understood as a description of an inferential practice central to science, is explanationism par excellence: it holds that such subjects come to accept a hypothesis because it provides a better explanation of the evidence than its rivals. Per se, normative and descriptive claims are easy to reconcile. Even if IBE and Bayesianism are entirely different, it might be that IBE describes how we do reason while Bayesianism describes how we ought to reason (but do not). (98)
The first major question is: is IBE descriptive or normative?
Incompatibilism arguments seem to rely quite heavily on the Dutch book arguments that show that IBE cannot be a normative account of reasoning.
Van Fraassen argues (see Bird, 99) that
"if explanationist considerations were relevant, then they would encourage a thinker to boost the credence p(h|e) above what is mandated by Bayes’s theorem, when h is a better explanation than competitor hypotheses while reducing the corresponding credences for those competitors. Such a thinker would be open to a Dutch book (a set of bets he takes to be fair but which will lose money whatever the outcome). An account of reasoning cannot be normatively correct if it enjoins accepting a Dutch book. So IBE is not a normatively correct account of reasoning." (Bird, 99)
Remainder
- What other notions of 'best' have been approached to 'best = most probable' in the context of Bayesian models? What are the pros and cons of each alternative?
- Subquestion, see next subsection: is there an alternative that can also be a "lovely" explanation?
See Glass, 412: Can the Bayesian interpretation of IBE consider lovely rather than the most "likely" explanations?
Glass mentions critique of Van Fraassen (on IBE or IBE as Bayesianism?). Look up ref.
See Glass, 418: variant of the previous question; can explanatory virtues also be taken into account? (Is that a matter of psychology?)
Explanatory virtues
TODO make a separate note for this.
Lipton, p. 6:
Candidates for explanatory virtues:
- Scope (explain more types of phenomena)
- Precision
- Mechanism (provide information about underlying mechanisms)
- Unification (unify apparently disparate phenomena)
- Simplicity (cf. Ockham's Razor)
Example: there is no "uncontroversial analysis of unification or simplicity"; and are these artefacts of the specific formulation of the explanation or the explained hypothesis?
Leah Henderson. "Bayesianism and IBE" in Best Explanation: New Essays on Inference to the Best Explanation:
On the other hand, scientific arguments for one theory over another often appeal more or less explicitly to explanatory advantages of that theory. For example, Darwin argues very explicitly in The Origin of the Species that the theory of natural selection provides a better explanation of ‘large classes of facts’ than special creationism (Darwin 1962 [1859]). Copernicus argues that his heliocentric theory possesses greater ‘harmony’ than its Ptolemaic rival, which may be seen as an appeal to a kind of explanatory virtue (Copernicus 1543). (250)
Draft mail aan Johan
Beste Johan,
Ik wilde even aan de pols voelen of ik met mijn TFCA voorstel voor aankomende vrijdag in de goede richting denk.
Volgens Lipton en Glass moet een theorie over "inference to the best explanation" (IBE) een methodologie bieden om één verklaring/hypothese over een andere te verkiezen, maar daarnaast ook een beschrijving kunnen geven van wat precies een (goede) verklaring is. Een van de voordelen van de Bayesiaanse aanpak is dat hypothesen altijd in contrast met andere hypothesen worden beoordeeld en dus heel geschikt is als methodologie voor het kiezen tussen hypothesen. Maar om de Bayesiaanse aanpak als een goede variant van IBE te zien - ervan uitgaande dat je IBE überhaupt niet afwijst - moet Bayesianism ook een antwoord bieden op de vraag wat een goede verklaring/hypothese precies goed maakt. Lipton en ook Glass hameren er op dat wanneer Bayesianism en IBE compatibel worden gemaakt door "best" als "most probable" te interpreteren, IBE triviaal wordt en diens meerwaarde verliest als poging om te beschrijven hoe (hoofdzakelijk) wetenschappers sommige hypothesen over andere verkiezen. De onderliggende hoofdvraag in dit debat over Bayesianism en IBE dus: hoe kun je beide compatibel maken zonder dat IBE triviaal wordt?
Ik zou graag in kaart willen brengen wat de recente suggesties zijn voor een Bayesiaanse versie van IBE die compatibel is met "lovely explanations", wat ik in de literatuur ook wel "Explanationist Bayesianism" genoemd heb zien worden. Ik kon vrij goed literatuur vinden die reageren op de problematiek die Lipton aankaart. Ik denk bijvoorbeeld aan werk van Leah Henderson (2014, 2017) die onderscheid maakt tussen een "constraint-based compatibilism" dat "explanatory considerations" laat meespelen, bijvoorbeeld in de keuze van priors, en "emergent compatibilism" dat betoogt dat het Bayesiaanse framework uit zichzelf al betere verklaringen (in een bredere zin) prefereert.
Ik beperk me dus door 1. specifiek de Bayesiaanse benadering van IBE te onderzoeken en 2. door niet het compatibel vs. non-compatibel debat helemaal uit te pluizen, maar me te concentreren op voorstellen binnen het compatibel-kamp.
Is deze focus scherp genoeg?
Met vriendelijke groet,
Edwin Wenink
Zie hieronder een voorlopige lijst met literatuur (niet geordend en consistent geformatteerd):
- L. Henderson (2014). Bayesianism and inference to the best explanation. British Journal for the Philosophy of Science, 65: 687–715, 2014.
- J. Weisberg. Locating IBE in the Bayesian framework. Synthese, 167: 125–44, 2009.
- Salmon, W. C. 2001. Explanation and confirmation: A Bayesian critique of inference to the best explanation. In G. Hon and S. S. Rakover (Eds), Explanation: Theoretical Approaches and Applications, pp. 61–91. Dordrecht: Kluwer.
- Myrvold, W. (2003). “A Bayesian Account of the Virtue of Unification,” Philosophy of Science70: 399-423
- Alexander Bird: Inference to the Best Explanation, Bayesianism, and Knowledge, 97-120. In "Best Explanations: New Essays on Inference to the Best Explanation"
- Leah Henderson: Bayesianism and IBE: The Case of Individual vs. Group Selection. In "Best Explanations: New Essays on Inference to the Best Explanation"
- Hitchcock, C. (2007). “The lovely and the probable,”Philosophy and Phenomenological Research,74(2): 433–440.
- Valeriano Iranzo: "Bayesianism and inference to the best explanation"
- T. Poston. Reason and Explanation: A Defense of Explanatory Coherentism. Palgrave Macmillan, Basingstoke, 2014.
- P. Lipton. Inference to the Best Explanation. Routledge, London, 2nd edition, 2004.
- James Ladyman: Wouldn't it be lovely: explanation and scientific realism. Metascience (2005) Review of Peter Lipton (2004)
- Cabrera, F. Can there be a Bayesian explanationism? On the prospects of a productive partnership. Synthese 194, 1245–1272 (2017). https://doi-org.ru.idm.oclc.org/10.1007/s11229-015-0990-z
- Douven, I. (2005). “Wouldn't It Be Lovely: Explanation and Scientific Realism,”Metascience, 14: 338-343
- Douven, I.and Schupbach,J.(2015). “Probabilistic alternatives to Bayesianism: the case of explanationism,”Front. Psychol.,(6):459, doi: 10.3389/fpsyg.2015.00459.
- Douven, I. and Wenmackers, S. (2015). “Inference to the Best Explanation versus Bayes’s Rule in a Social Setting,” Brit. J. Phil. Sci., in press.
- Psillos, S.(2004). “Inference to the Best Explanation and Bayesianism,” in Stadler (ed.), Induction and Deduction in the Sciences.Dordrecht: Kluwer, 83–91.
- Harman, G. (1965). “The Inference to the Best Explanation,”Philosophical Review, 74: 88–95.
- Mackonis, A. (2013). “Inference to the best explanation, coherence and other explanatory virtues,” Synthese, 190(6): 975-995.
Draft
Volgens Lipton en Glass moet een theorie over "inference to the best explanation" (IBE) een methodologie bieden om één verklaring/hypothese over een andere te verkiezen, maar daarnaast ook een beschrijving kunnen geven van wat precies een (goede) verklaring is. Een van de voordelen van de Bayesiaanse aanpak is dat hypothesen altijd in contrast met andere hypothesen worden beoordeeld en dus heel geschikt is als methodologie voor het kiezen tussen hypothesen. Maar om de Bayesiaanse aanpak als een goede variant van IBE te zien - ervan uitgaande dat je IBE überhaupt niet afwijst - moet Bayesianism ook een antwoord bieden op de vraag wat een goede verklaring/hypothese precies goed maakt. Lipton en ook Glass hameren er op dat wanneer Bayesianism en IBE compatibel worden gemaakt door "best" als "most probable" te interpreteren, IBE triviaal wordt en diens meerwaarde verliest als poging om te beschrijven hoe (hoofdzakelijk) wetenschappers sommige hypothesen over andere verkiezen. De onderliggende hoofdvraag in dit debat over Bayesianism en IBE dus: hoe kun je beide compatibel maken zonder dat IBE triviaal wordt?
It is a common scenario for any scientist to reason about which hypothesis is the best explanation for some observed phenomena, and infer that hypothesis if it is the best out of the available explanations. This type of reasoning is studied as "inference to the best explanation" (IBE). According to
Requirement.
One framework that has been successful in modeling this type of inference is Bayesianism.
that, when confronted with a set of observations, one has to reason about which of the possible hypotheses is the best explanation for a given
In science one is presented with a particular observation,
Both in science and everyday life, we often have to make inferences from