[TMP-010] A Transparent and Explainable Dialogue System for Immigration Services

Contact person: Andrea Galassi (a.galassi@unibo.it) 

Internal Partners:

1. University of Bologna, Andrea Galassi, a.galasi@unibo.it

External Partners:

1. University of Calabria, Bettina Fazzinga, bettina.fazzinga@unical.it
2. University of Naples, Margherita Vestoso, margherita.vestoso@unina.it

Migration is one of the top current concerns of the European Union that requires harmonizing ethical, legal and societal considerations. Unfortunately, the application rules and laws concerning immigration and asylum require a nontrivial evaluation of legal and factual facts, which makes it difficult for migrants to achieve a preliminary overview of their chances to obtain protection. In this micro-project, we explored how trustworthy AI can play a role in making asylum application processes more efficient and fair. To this end, we gathered a multidisciplinary team of computer scientists and immigration law experts. The team worked towards the creation of a chatbot aimed at supporting migrants who seek asylum in Europe. In particular, we developed a prototypical tool to support, inform, and guide asylum applicants in the process (and not to assist or replace a judiciary expert in a “predictive justice” fashion). We started from a previous micro project (“Ethical Chatbots”, Fazzinga, B., Galassi, A., and Torroni, P. (2022). A privacy-preserving dialogue system based on argumentation. Intelligent Systems with Applications, 16, 200113) and improved our argumentative chatbot architecture to address the complexity of the new domain and to exploit the power of LLMs. We maintained the focus on properties such as data governance and privacy preservation, transparency, explainability, and auditability.

Our experience highlighted the necessity of developing this kind of tool in close relationships with domain experts. The tool development process demonstrated that, besides the knowledge of the relevant legal framework, a crucial role is played by a set of unwritten best practices and conventions, often not explicitly represented anywhere, and mostly known by practitioners through experience. Moreover, since it is extremely difficult to have access to the asylum request and the corresponding court decisions are not shared publicly, for the sake of applicants’ safety, a data-oriented approach is not feasible, and neither do current LLMs, trained on available data, have the necessary information for providing meaningful answers.

Results Summary

We developed “ACME”, a prototype chatbot with the aim of supporting migrants in their requests for asylum. ACME is a hybrid architecture that combines a subsymbolic language understanding module based on NLP techniques and LLM, with a symbolic reasoning module based on computational argumentation.

The aim of the tool is to help migrants identify the highest level of protection they can apply for. This would contribute to a more sustainable migration by reducing the load on territorial commissions, Courts, and humanitarian organizations supporting asylum applicants.

Relevant properties ACME exhibits include: data governance and privacy thanks to its modular architecture; transparency and explainability thanks to argumentative reasoning; and the ability to integrate and reasoning with explicit, expert-made, formalized knowledge, ensuring auditability.

Tangible Outcomes

1. Bettina Fazzinga, Elena Palmieri, Margherita Vestoso, Luca Bolognini, Andrea Galassi, Filippo Furfaro, Paolo Torroni (2024). “A Chatbot for Asylum-Seeking Migrants in Europe”. IEEE International Conference on Tools with Artificial Intelligence (ICTAI) http://arxiv.org/abs/2407.09197 
2. chatbot code https://github.com/lt-nlp-lab-unibo/ACME-A-Chatbot-for-Migrants-in-Europe 
3. Video demonstration of the tool: https://www.youtube.com/watch?v=P8iW7FOZTYM&feature=youtu.be