PhD position 08 - MSCA COFUND, AI4theSciences (PSL University, France) - "Can we imagine a decision-making system as a support for access to law? Illustration around the European regulation on AI ".

“Artificial intelligence for the Sciences” (AI4theSciences) is an innovative, interdisciplinary and intersectoral PhD programme, led by Université Paris Sciences et Lettres (PSL) and co-funded by the European Commission. Supported by the European innovation and research programme Horizon 2020-Marie Sklodowska-Curie Actions, AI4theSciences is uniquely shaped to train a new generation of researchers at the highest academic level in their main discipline (Physics, Engineering, Biology, Human and Social Sciences) and master the latest technologies in Artificial Intelligence and Machine Learning which apply in their own field. 26 doctoral students will join the PSL university's doctoral schools in 3 academic cohorts to carry out work on subjects suggested and defined by PSL's scientific community. The 2023 call will offer up to 4 PhD positions PhD on 10 PhD research projects. The candidates will be recruited through HR processes of high standard, based on transparency, equal opportunities and excellence.

Subject description: 

Context

The purpose of this doctoral research project would be to imagine a decision-making system (decision support) based on the analysis of the way in which a legal text was adopted and then interpreted. The starting point would therefore be to access, process and analyse/interpret a large mass of legal data comprising all the preparatory work for the development of a legal text.

A decision-making system, like the law, is based on different steps that can help stakeholders find, among other things, relevant information to improve their decisionmaking. It is therefore an automated decision-making aid which remains the responsibility of the stakeholder.

Such a project starts from the paradigm that a decision-making system, like legal analysis, involves extracting information from a mass of data and creating knowledge. It will therefore be a question of seeing to what extent a decision-making system could support legal analyses.

In other words, the objective here will be to provide a tool to help lawyers handle a large amount of evolving texts/data, compare, and analyse them.

Scientific objectives, methodology and expected results

The starting point of this work is therefore a large volume of data (rather textual) in evolution (over time) which must be compared and analysed in order to help lawyers to make (good) decisions.

For this, it will be necessary to address (for the most important): the management of large volumes of data, textual analysis, and the management of document versions.

The management of large volumes of data will make use of Big data, since today, legal texts/data are accessible in different places and in different formats. It will therefore be necessary to raise issues related to recovery, cleaning, and storage for effective analysis. To our knowledge, no standard approach exists for the specific treatment of legal texts [1]. However, such “automated” help would greatly facilitate the work of lawyers who today carry out this work manually.

Version control or revision control consists in managing all the versions of one or more files (generally in text). Essentially used in the field of software creation, it mainly concerns the management of source codes. It is a tedious and relatively complex activity. Software support is almost essential. There are different version control software which, although having common concepts, each bring their own vocabulary and their own uses. For example, there is a rudimentary versioning mechanism in Wikipedia: for each article, the history is available by clicking on the « View history » link; each line is a version of the article. Such a system is linear, as opposed to a more elaborate content management, according to a tree structure. Unfortunately, such possibilities do not exist for legal texts [2, 3].

Textual analysis, on the other hand, raises different problems related in particular to language and vocabulary (here, for example, we will be interested in the European regulation on AI, which can be written in different languages and each language is structured differently), document format (some documents are in proprietary formats, e.g., PDF, and it is difficult to extract parsable text from them), and textual similarity. Evaluating the similarity between textual documents is one of the important issues in several disciplines such as the analysis of textual data, information retrieval or the extraction of knowledge from textual data (Text Mining). In each of these areas, similarities are used for different processing: (i) in textual data analysis, similarities are used for description and data mining; (ii) in information retrieval, the evaluation of the similarities between documents and queries is used to identify the relevant documents in relation to the information needs expressed by the users; (iii) in Text Mining, similarities are used to produce synthetic representations of large document collections[4].

The techniques used to calculate the similarities obviously vary according to the disciplines, but they are however most often integrated into the same general two-step approach:

• Textual documents are first associated with specific representations that will serve as the basis for calculating similarities. Although the precise nature of the representations used is highly dependent on the domain of application, it should be noted that in almost all cases documents are represented as elements of a high-dimensional vector space.
• A mathematical model is chosen to measure the similarities. 

Again, to our knowledge, very few works in the literature have been carried out in this direction [5, 6, 7].

The objective of the thesis is therefore to propose a model of decision-making system, which will initially be applied to legal texts related to the European regulation on AI, and then generalized for other cases of application of legal texts. It will be:

• Model a text: according to its format, type, etc. It will therefore be up to us to define in our system what exactly it is, and to model it,
• Develop a decision-making system model to integrate large, specific, versioned data/information, manage it and enable its efficient analysis.
• Validate the usability of this system with real data, here from the European regulation on AI. This will go through the development of the system. We must also plan for the system to be able to evolve easily to consider new elements that would become available. A difficult point will be to guarantee a computation time short enough to ensure a quasi-real-time response.
• Generalize the system for other application cases. The main objective of this step is to have a generic decision-making system model (regardless of the scope of the texts, the types of texts, etc.).

In summary, the originality of this research consists in applying Big Data techniques, decision-making systems, information retrieval, text mining, machine learning, among others, to the still little explored field of Comparative (Public) Law. The approach should be designed to consider large, versioned data and help decision-making.

References

1. Moses, L. B., & Chan, J. (2014). Using big data for legal and law enforcement decisions: Testing the new tools. The University of New South Wales Law Journal, 37(2), 643–678.
2. Biasotti, A.A., Francesconi, E., Palmirani, M., Sartor, G., & Vitali, F. (2008). Legal Informatics and Management of Legislative Documents.
3. John Garofalakis, Konstantinos Plessas, and Athanasios Plessas. 2015. Automated analysis of greek legislative texts for version control: limitations, caveats and challenges. In Proc. of the 19th Panhellenic Conference on Informatics (PCI '15). Association for Computing Machinery, NY, USA, 115–116. 
4. Negre E. (2013), Comparaison de textes: quelques approches..., Paris, Cahiers du LAMSADE.
5. J. Brett Crawley and G. Wagner, "Desktop text mining for law enforcement," 2010 IEEE Int. Conf. on Intelligence and Security Informatics, Vancouver, BC, Canada, 2010, pp. 138-140.
6. Feinerer, I. (2008). A text mining framework in R and its applications. [Doctoral thesis, WU Vienna].
7. Chou, S., Hsing, TP. (2010). Text Mining Technique for Chinese Written Judgment of Criminal Case. In: Chen, H., Chau, M., Li, Sh., Urs, S., Srinivasa, S., Wang, G.A. (eds) Intelligence and Security Informatics. PAISI 2010. Lecture Notes in Computer Science, vol 6122. Springer, Berlin, Heidelberg. 
8. Devins, Caryn; Felin, Teppo; Kauffman, Stuart; and Koppl, Roger (2017) "The Law and Big Data,"  Cornell       Journal      of      Law      and      Public      Policy:      Vol.      27:      Iss.2,      Article      3. 

Thesis supervisors

ElsaNegre, MCF HdR, LAMSADE, Université Paris Dauphine - PSL

Olivia Tambou, MCF HdR, Cr2D, Université Paris Dauphine - PSL

PSL University PSL

Created in 2012, Université PSL is aiming at developing interdisciplinary training programmes and science projects of excellence within its members. Its 140 laboratories and 2,900 researchers carry out highlevel disciplinary research, both fundamental and applied, fostering a strong interdisciplinary approach. The scope of Université PSL covers all areas of knowledge and creation (Sciences, Humanities and Social Science, Engineering, the Arts). Its eleven component schools gather 17,000 students and have won more than 200 ERC. PSL has been ranked 36 in the 2020 Shanghai ranking (ARWU).