Section: New Results

Dealing with Imperfect Information

Participants : Pierre Bisquert, Patrice Buche, Abdelraouf Hecham, Madalina Croitoru, Jérôme Fortin, Rallou Thomopoulos, Bruno Yun.

Reasoning in presence of inconsistencies is a challenging task both from a theoretical and an application point of view. From a theoretical point of view, it means finding methods that can tolerate the inconsistency. From an application point of view, it means providing meaningful results to the user. In the works carried out this year inconsistency arose while developing a decision support system assisting human experts with a given task. The main challenges we faced where first to provide the user with a comprehensive vision over the different possibilities and then to assist her while she makes a decision.

Argumentation in the Existential Rule Setting

Our first line of work focussed on the use of argumentation-based methods for reasoning in presence of inconsistencies within knowledge bases expressed in the formalism of existential rules. Such inconsistency can occur either in the facts or in the rules of the knowledge base.

Logical based argumentation instantiates abstract argumentation frameworks by $i)$ constructing arguments from inconsistent knowledge bases, $ii)$ computing attacks between them, and $iii)$ using so-called argumentation semantics in order to select acceptable arguments and their conclusions. The advantage of using argumentation for reasoning in an inconsistent setting lies in the explanatory power of argumentation frameworks. We considered the first case of inconsistency arising from the factual level and investigated the formal properties of the argumentation frameworks. We showed then that the argumentation is of practical use as it allows for a principled explanatory dialogue. Finally, we carried out an experiment that compared the explanatory power of argumentation in this setting and found out that positive results are only achieved if particular attention is given to the phrasing of such interaction.

• These results were published at ESA 2017 [12], IJAR 2017 [13], and DL 2017 [28]

In logical based argumentation, arguments are sometimes based upon equivalent data. Cores are notions introduced in that delete such arguments. We investigated two different notions of core in such a logically instantiated argumentation framework (more details about the instantiation can be found in [12]) that will remove redundant arguments and attacks in a different manner. We do not follow the argumentation semantics “a la Dung” but study ranking semantics that return a total order over the set of arguments in the logical argumentation framework. We show that the manner of defining the core of a logically instantiated argumentation framework affects the ranking output of ranking semantics.

• These results were published at AAMAS 2017 [38] and IDA [37]

Another setting we explored was when the inconsistency arises from the rules (also sometimes referred to as incoherence). In this setting we investigated defeasible logics and proposed a refined formalism for defeasible existential rules. We showed that in the case of defeasible reasoning one may be interested in generating all provenance paths of an atom, an issue which raises an interesting technical challenge. In order not to lose paths due to the skolemisation process we introduced a new combinatorial structure called the graph of atom dependency and showed how using this structure prevents provenance paths loss. We implemented our approach and showed that it has a very good performance with respect to the other argumentation based tools that could be used for defeasible reasoning in existential rules.

• These results were published at RuleML+RR 2017 [29] and AAMAS 2017 [30]

Human Interaction and Decision Making

Our second line of contributions focussed on how to bridge the gap between the human and the machine in a decision support setting. We thus investigated how human reason and how cognitive biases can influence decision making. Then, we approached the decision making process by developing methods based on voting theory and classical decision theories allowing us to achieve desirable properties.

We proposed a dual system for artificial agents combining deductive logical reasoning with intuitive reasoning for the sake of argumentation. Our contribution is the definition of a new formal model of flexible argument evaluation. We consider that, when it is not possible for an agent to make a logical inference (since it requires too much cognitive effort or she has insufficient knowledge), she might replace certain parts of the logical reasoning with mere associations. We applied our work on the Durum Wheat variety selection in the context of the French National Agency (ANR) Dur-Dur project.

• These results were published at MM 2017 [15]

Collective decision making is classically done via social choice theory with each member of the group expressing preferences as a (total) order over a given set of alternatives, and the group’s aggregated preference is computed using a voting rule. However, such methods do not take into account the rationale behind agents’ preferences. Our research hypothesis is that a decision made by a group of participants understanding the qualitative rationale (i.e., arguments) behind each other’s preferences has better chances to be accepted and used in practice. To this end we proposed a novel qualitative decision process which combines argumentation with computational social choice for modelling the decision-making problem. We show that a qualitative approach based on argumentation can overcome some of the social choice deficiencies. A first version of this approach was implemented and practically demonstrated in [25].

• These results were published at ADT 2017 [41]

A recent work in cooperation with Laval University (Canada) and AGIR joint research unit (Toulouse) deals with the combination of argumentation and system dynamics simulation for decision support in the agri-food sector. We propose a systematic method to assess possible options, based on the complementarity of argumentation modeling and system dynamics (SD) simulation, in conjunction with field experimentation. As a practical application, we assess various options available to agri-food chain stakeholders when considering the adoption of cereal-legume intercrops as an alternative to sole crops. Moreover, we carried out complementary studies to explore the possible added-value of argumentation for decision support in practical cases related to agri-food chains. We proposed the introduction of numerical indicators in argumentation systems in order to evaluate to what extent the system studied (a short food supply chain) is polemical, i.e. subject to divergent viewpoints, and which criteria are mainly involved in these divergences. As a study case, we considered a food policy about bread-making which illustrated that a given argument may be interpreted through different scenarios, among which unexpected worst-cases can occur.

• These results were published at IEA/AIE 2017 [33], [35] and WCCA 2017 [34], [32]