Assessing the use of Artificial Intelligence in Asylum Procedures Date: July 20, 2023 Presented by: Ludivine Sarah Stewart What Do We Mean By Al? What is AI? 'Artificial intelligence system' (Al system) means a machine-based system that is designed to operate with varying levels of autonomy and that can, for explicit or implicit objectives, generate output such as predictions, recommendations, or decisions influencing physical or virtual environment" (Euractiv, 2023) Why Do Asylum Authorities Turn To Al? 1. Improve data quality 2. Reduce time- consuming tasks 3. A faster asylum processes 4. Potential to reduce errors 5. Uncover fraud & identify vulnerable individuals 6. Full automation not envisaged But... use of Al raises important normative questions: should we use Al in asylum procedures? When should we use it? Should the law regulate its use in asylum and how? The Article  Advances a normative framework of due process for assessing the use of Al in asylum procedures  Goal: guide the reflections on the use of Al in AP and understand where th challenges lie in the context of an emerging legal framework on Al in Europe  Research question: What does due process require from procedures and how can Al in asylum procedures potentially challenge those values? Structure Of The Paper 1. Introduction 2. A Due Process Framework for Al-driven Decision-Making 3. Al and Due Process: Benefits and Pitfalls 4. Framing the Debate 5. Conclusions Why Does It Matter?  Due Process in legal theory: a 'guarantee' against the state that protects individuals against abusive exercise of power and arbitrary decisions (Gordon Anthony 2011; Nunes Chaib, 2019)  Due Process as a legitimizing function of the state's power (Mashaw, 2003; Hovell, 2016)  In Asylum: 1. The asylum process as the 'gateway to protection' (Kagan, 2006) 2. Link between adequate procedures and non-refoulement What Does It Require For Procedures? 1. Accuracy Instrumentalist approach: procedures should ensure the correct application of the law to the facts. 2. Participation Non-instrumentalist: procedures should ensure opportunity to influence decisionmaking process Benefits And Pitfalls For Due Process: Accuracy 1. Potential For Errors In Decision-Making a. Poor performance of the AI system: 1) Errors at the design or development stage including errors in datasets 2) Importance of good quality in datasets! b. Reliance on correlation and statistics 2. Biased Decision-Making a. Main concern: bias in the data. 3. Failure To Rectify Errors And Bias a. From automation to confirmation bias b. A persistent challenge in Al: opacity Benefits And Pitfalls For Due Process: Participation Participation supposes at least reasons giving and explanation for an action/decision (Galligan, 1996) and opportunity to be heard. Challenges: 1. Asymmetry of information a. Lack of information given to individuals about the existence and functioning of the system b. Access to information and representation with technical knowledge 2. Explain ability of decisions Framing The Debate On Al In Asylum:  Proposal for an Al Act: potential mostly on accuracy and almost none on participation (e.g. no general right to information)  Al liability directive? Potential for participation - but ex-post  Council of Europe Zero Draft Convention on AI, Human Rights, Democracy and the Rule of Law?