#### This event was financed by

#### PhilPharm project

#### that has received funding from the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme – G.A. n. 639276

#### 2018 Program

#### Thursday 13th

######

9:00 - Opening: Barbara Osimani

Morning Session (Chair: Stefano Bonzio)

9:10 - 9:50 - Tommaso Flaminio

9:50 - 10:30 - Hykel Hosni

Coffee Break

11:00 - 11:40 - Serena Doria

11:40 - 12:20 - Jan Sprenger

Lunch

Afternoon Session (Chair: Tommaso Flaminio)

14:00 - 14:40 - Katya Tentori

14:40 - 15:20 - Vincenzo Crupi

Coffee Break

15:40 - 16:20 - Rosaria Gesuita

16:20 - 17:00 - Barbara Osimani

Excursion to Sirolo

and Social Dinner

#### Friday 14th

######

Morning Session (Chair: Armando Sacco)

9:10 - 9:50 - Roberto Festa

9:50 - 10:30 - Gustavo Cevolani

Coffee Break

11:00 - 11:40 - Giacomo Sillari

11:40 - 12:20 - Paolo Galeazzi

Lunch

Afternoon Session (Chair: Francesco De Pretis)

14:00 - 14:40 - Carlo Martini

14:40 - 15:20 - Mattia Adreoletti

15:20 - 16:00 - Discussion

#### Abstracts

#### Mattia Andreoletti

Assessing causality of adverse events during early phase clinical trials in oncology: bayesian networks meet case-based reasoning

This paper addresses the issue of causality assessment of adverse drug reactions in early phase clinical trials, with a focus on oncology. A growing body of evidence suggests that, in such contexts, causal judgments are highly subjective and might have a negative impact on the quality of drugs safety data. To overcome this, medical researchers have developed many different methods and tools to assist physicians in causality attribution. Nonetheless, all these methods have significant limitations. I am going to restate this debate in philosophical terms, exploring whether and how recent philosophical research on singular causation can suggest a plausible solution to this problem and contribute to clinical practice.

Gustavo Cevolani

Probability, truthlikeness, and rational belief

Keywords: rational belief, probability, truthlikeness, Lottery paradox, Preface paradox, Lockean thesis, Carneades thesis

Providing a theory of rational belief in the face of uncertainty remains a crucial issue in both (formal) epistemology and philosophy of science. Well-known difficulties, like the Lottery and Preface paradoxes, trouble the attempt to combine a quantitative, probabilistic account of uncertain belief with our pre-systematic intuitions about plain or categorical belief. In this paper, I first present and motivate a non-conventional view of these issues based on the idea, called here “Carneades thesis”, that rational belief aims at truth approximation. Then, I show how this account deals with the Lottery and Preface paradoxes. Finally, I discuss the pros and cons of this proposal, as well as its connections with the Lockean thesis, the distinction between belief and acceptance, and Leitgeb’s stability theory of belief.

#### Serena Doria

Representations of preference orderings by coherent upper and lower conditional previsions defined with respect to Hausdorff outer and inner measures

Sufficient conditions to assure that a maximal random variable is a Bayes random variable are given when the preference ordering is represented by coherent upper and lower conditional previsions.

#### Roberto Festa

Laws of likelihood and Matthew effects for Bayesian confirmation theory

A Bayesian confirmation measure expresses the degree to which evidence E confirms hypothesis H in terms of the probabilistic relations between E and H. In particular, an incremental measure c(H,E) is a function of p(H│E) and p(H) such that c(H,E) increases both when p(H│E) increases and when p(H) decreases. Bayesian theorists have proposed and defended several incremental measures, characterized by different, and partially incompatible, formal properties. In the last few years, such properties have been systematically investigated, in order to identify the structural conditions characterizing specific (classes of) incremental measures. In this paper, we contribute to this ongoing work on the grammar of Bayesian confirmation by focusing on two families of structural conditions. The first include some well known “laws” of likelihood, among which the Weak Law of Likelihood (WLL) plays a prominent role; the second deals with what we call “Matthew effects” (or properties) for confirmation. First, we discuss and motivate these two sets of conditions as applied in Bayesian confirmation theory. Second, we prove that, although likelihood and Matthew principles are inspired by very different intuitions, they show some surprising, logical and conceptual relationships. Third, we present a simple class of confirmation measures, which satisfy some counterintuitive principles but still find place in our classification. Finally, we suggest that measures of this kind capture some important intuitions concerning confirmation, that makes them highly plausible in a number of interesting cognitive contexts.

#### Tommaso Flaminio

#### Strict coherence on many-valued events

This talk offers a logical perspective on the subjectivist foundation of probability through the analysis of coherence. We do so by revamping the interest in its "strict" version which was put forward in the mid 1950's by Shimony and Kemeny (and favoured by Carnap). In particular we investigate the property of strict coherence in the setting of many-valued logics. Our main results read as follows: (i) a map from an MV-algebra to [0,1] is strictly coherent if and only if it satisfies Carnap’s regularity condition, and (ii) a [0,1]-valued book on a finite set of many-valued events is strictly coherent if and only if it extends to a faithful state of an MV-algebra that contains them. Remarkably this latter result allows us to relax the rather demanding conditions for the Shimony-Kemeny characterisation.

#### Paolo Galeazzi

#### Multi-criteria games

In most interesting decision problems, action choices heavily depend on the decision criterion adopted by the decision maker in evaluating the alternative options. Different choices will in turn produce different outcomes and different rewards to the decision maker. In the talk, I will try to introduce and justify game-theoretic situations where different players may choose actions according to different decision criteria. Time permitting, the presentation will then focus on two things: (i) how players could infer the others' decision criteria from observed play; (ii) how beneficial different decision criteria are in terms of rewards to the players.

#### Hykel Hosni

Boolean algebra of conditionals: probabilities and logic

Conditionals play a fundamental role both in qualitative and in quantitative uncertain reasoning. In the former, conditionals constitute the core focus of non-monotonic reasoning [1, 4, 5]. In quantitative uncertain reasoning, conditionals are central both for conditional probability, and more generally, for conditional uncertainty measures [6].

In this contribution we present a construction that builds, from any Boolean algebra A , an algebra of conditionals C (A) as a suitable quotient of the Boolean algebra freely generated by A x ( A \ { ⟂ } ), [2, 3]. The algebra C (A) is called the Boolean algebra of conditionals of A . If A is finite, C (A) is finite as well and hence atomic. The atomic structure of C (A) is fully characterized from the atoms of A . Besides presenting Boolean algebras of conditionals, a main purpose of this contribution is to show these structures provide a suitable algebraic setting in which conditional probability theory and certain nonmonotonic logics can be investigated.

In particular, in the area of conditional probability theory, Boolean algebra of conditionals are applied to provide an answer to the longstanding question if a conditional probability may be regarded as a simple (i.e., unconditional) probability on conditional objects [7, 5]. In these regards, although not every simple probability on C (A) satisfies all axioms of a conditional probability on A , [6], we will show the following result.

Theorem 1 . For every positive probability measure P on a Boolean algebra A, there exists a positive probability measure µp on

C (A) which agrees with P on the basic conditionals of C (A). In other words, for every basic conditional (a | b),

Then, we introduce the logic of Boolean conditionals (LBC) and prove its completeness with respect to the natural semantics induced by the structural properties of the atoms in a conditional algebra. Further. LBC is proved to be sound and complete with respect to a suitably defined Boolean algebra of conditionals. Finally we conclude with a result to the effect that LBC is indeed a preferential consequence relation, in the sense of the well-known System P, [1].

Aknowledgements.

The author acknowledges partial support by the Spanish Ramon y Cajal research program RYC-2016-19799; the Spanish FEDER/MINECO project TIN2015-71799-C2-1-P and the SYSMICS project (EU H2020-MSCA-RISE-2015 Project 689176).

References

[1] D. Dubois, H. Prade. Conditional Objects as Nonmonotonic Consequence Relationships, IEEE Transaction on Systems, Man and Cybernetics 24(12): 1724-1740, 1994

[2] T. Flaminio, L. Godo, H. Hosni, On the algebraic structure of conditional events, in S. Destercke and T. Denoeux (eds.) Symbolic and Quantitative Approaches to Reasoning with Uncertainty ECSQARU 2015, Lecture Notes in Computer Science Volume 9161, 106-116, Springer, 2015.

[3] T. Flaminio, L. Godo, H. Hosni, On Boolean algebras of conditionals and their logical counterpart, In: A. Antonucci, L. Cholvy, O. Papini (eds) Symbolic and Quantitative Approaches to Reasoning with Uncertainty ECSQARU 2017, Lecture Notes in Computer Science Volume 10369, 246-256, Springer, 2017.

[4] A. Gilio, G. Sanlippo, Conjunction, Disjunction and Iterated Conditioning of Conditional Events. SMPS 2012: 399-407, 2012.

[5] I. R. Goodman, R. P. S. Mahler, H. T. Nguyen. What is conditional event algebra and why should you care? SPIE Proceedings, Vol 3720, 1999.

[6] J. Y. Halpern. Reasoning about Uncertainty . MIT Press, 2003

[7] D. Lewis. Probabilities of Conditionals and Conditional Probabilities. The Philosophical Review 85(3): 297{315, 1976.

####

#### Carlo Martini

#### Biases in scientific judgment

Since the 1950s, Paul Meehl and a number of his collaborators researched how biases affect scientifici judgment, and argued that there were differences in performance between actuarial and expert judgment. Since then, actuarial judgment and expert judgment had to be traded off against each other in science-based decision-making. In this paper I review the history of the face-to-face between actuarial and expert judgment with a focus on biases. The history of the debate will suggests directions for how to handle the problem of biases in both expert and actuarial judgment.

####

####

#### Barbara Osimani

Varieties of Error and Varieties of Evidence in Scientific Inference

As Paul Meehl remarks, “Any working scientist” is more impressed with 2 replications in each of 6 highly dissimilar experimental contexts than he is with 12 replications of the same experiment”; this intuition has been formalised in formal epistemology as the Variety of Evidence Thesis that is, the fact that ceteris paribus, more heterogeneous evidence coming from independent sources is more confirmatory than less varied evidence. The thesis is known to fail for random evidence and for deterministically biased evidence ([Bovens and Hartmann(2003),Claveau(2013)]). However, Bovens and Hartmann’s results concerning the failure of the variety of evidence thesis (VET), mainly rely on the unreliable instrument being a randomiser, and of a very specific kind. When the rate of positive reports delivered by the instrument (no mater what the truth is) is .5, the instrument is a proper randomiser. However, as soon as such probability is higher than .5 the instrument tends to be a “yes-man”, whereas it is a “nay-sayer” if this probability drops below .5. In the former case consistency of positive reports from the same instrument speaks in favour of it being a randomizer (and therefore weakens their confirmatory strength), whereas the opposite holds for the latter case, and that’s the reason for VET failure in the latter situation.

Furthermore, Bovens and Hartmann's results run against the “too-good-to-be-true” intuitions underpinning suspicion of bias for considerable long series of reports from the same testing instrument. This happens because having divided the hypothesis space for the truth-tracking properties of the instrument in either perfectly reliable, or randomizer, then a series of consistent reports become less and less likely under the hypothesis of a randomizing instrument, and, consequently, more and more likely under the complementary hypothesis.

In order to account for the “too-good-to-be-true” intuition, and for the related suspicion of systematic bias, we developed a model where the instrument may either be reliable but affected by random error, or unreliable and systematically biased towards delivering positive reports (but non-deterministically so).

In our model the VET fails as well, but the area of failure is considerably smaller and affects borderline cases where the ratio of false to true positive reports for the two instruments become favourable for the biased one. In our case, VET failure simply follows from the fact that receiving two positive reports from the same instrument increases the probability that this is the positively biased one; and once you are there, then, if the the assumed ratio of false to true positive reports is more favourable for the biased instrument, then receiving the two reports is more confirmatory if they come from the same instrument. Otherwise, in case the ratio of false to true positive reports is unfavourable for the biased instrument with respect to the other one, then the two report are more confirmatory if they come from independent instruments, and therefore the VET holds in this case.

Also, we identified the settings where the role of the strength of the tested consequences of the hypothesis matters for VET holding/failing and discovered why. Finally, we explain why the apparently counterintuitive result in Bovens and Hartmann model, according to which the area of the VET failure grows for increasing reports.

#### Jan Sprenger

Trivalent Logics of Indicative Conditionals and Bayesian Inference* (joint work with Paul Égré and Lorenzo Rossi)

The semantics of indicative conditionals are a notoriously difficult philosophical problem. Many philosophers have adopted the view that they don't have classical truth conditions: the truth value of 'If A, then B', cannot be determined as a function of the truth values of A and B. Focus has therefore shifted to the assertability and/or probability of indicative conditionals (e.g., Adams, Edgington), where a broadly Bayesian analysis takes over.

However, this view severs the ties between the semantics and the epistemology of indicative conditionals. Instead, we propose a principled solution that goes back to an idea by de Finetti: to adopt a trivalent semantics, based on reading indicative conditionals as conditional predictions of the consequent, or as conditional bets on this proposition. We characterize an appropriate trivalent logic of such conditionals (i.e., we identify the most appropriate truth table and validity relation) and we provide a proof theory (sequent calculus, algebraization). Finally we show how a Bayesian account of the assertability of conditionals, including the famous equation p(if A, then B) = p(B|A), naturally flows from this approach.

#### Katia Tentori

Evaluating the descriptive accuracy of forecasting models

Alternative measures have been proposed to evaluate the accuracy of forecasts. I will report the results of two experiments that compare the adequacy of three scoring rules as descriptions of accuracy judgments in a probabilistic prediction context