M/F Traduction des énoncés de préférence formulés en langage naturel vers des langages de préférence formels ; applications à la prise de décision collective

Job Description

Thesis Subject

Standard collective decision-making mechanisms exhibit a certain rigidity, both in the format used to express voters' preferences and in the way constraints on the outcome can be stated. This is not an issue for simple problems involving the election of a single candidate, but it becomes one as soon as multiple candidates or projects are elected, or objects are allocated to individuals.

Consider, for example, the election of a representative council (such as a laboratory council): a voter may wish to express complex preferences such as " I would like a council that is more or less balanced in terms of gender and seniority, but above all, I want my research group to be represented at least in proportion to its importance within the laboratory. All other things being equal, I would like Ann, Bob, and possibly Carol to be members of the committee, and I would like Daniel and Elena not to be members."

Another example: if I am conducting a poll to organize one or more meeting(s), rather than simply counting the number of attendees, I may want to specify soft constraints on the participants (e.g., "I prefer at least one member of each group to attend, and roughly three times as many senior members as PhD students}), and still others if several meeting dates have to be found".

Yet other examples can be found in fair division of indivisible goods.

Such preferences or constraints (by voters in the first example, by the organiser in the second one) are complex, because they involve preferential dependencies (e.g., my preference for the presence of a participant cannot be specified independently because it depends in which other participants are present).

On the one hand, preferences with such dependencies (technically, {\em nonseparable preferences}) have been considered in quite a lot of papers in individual and collective decision making (preference languages and logics, social choice on combinatorial domains). On the other hand, studies have mostly remained at the level of theoretical developments, for a pragmatic reason: they rely on complex formal languages for expressing preferences or constraints over combinatorial domains, which individuals or non-specialist decision-makers are likely to find difficult to use.

However, if they are allowed to express preferences or constraints in natural language, with an LLM-type tool downstream to perform the translation (along with verification tools that would need to be designed) in an interpretable formalism, new possibilities emerge. Hence the key question: how can existing social choice mechanisms be generalized to accommodate complex preferences and constraints, with the help of current AI tools?

The first part of the PhD thesis will consists in using LLMs to translate preference statements into various formalisms such as preference logics, soft constraints, or other formal preference languages such as CP-nets. The second part will apply the findings of the first part to social choice mechanismsn especially, multiwinner voting, participatory budgeting, and fair division.

Your Work Environment

The PhD will be performed in LMSADE, Université Paris-Dauphine, within the PRAIRIE-PSAI institute, under the supervision of Jérôme Lang and Benjamin Negrevergne

Constraints and risks

nothing special

Compensation and benefits

Compensation

2300 € gross monthly

Annual leave and RTT

44 jours

Remote Working practice and compensation

Pratique et indemnisation du TT

Transport

Prise en charge à 75% du coût et forfait mobilité durable jusqu’à 300€

About the offer

About the CNRS

The CNRS is a major player in fundamental research on a global scale. The CNRS is the only French organization active in all scientific fields. Its unique position as a multi-specialist allows it to bring together different disciplines to address the most important challenges of the contemporary world, in connection with the actors of change.

CNRS

The research professions