A Regulatory and Human Factors Approach to Overreliance in Legal Information Chatbots

Open Access
Article
Conference Proceedings
Authors: Niki Georgiadou, Maria Rigou
Abstract

Chatbots are rapidly transforming the way individuals from diverse social and economic backgrounds access legal information. This free access to legal knowledge can act as a corrective mechanism within society, reducing inequalities among social groups. Notwithstanding their benefits, the growing use of chatbots for legal matters raises significant concerns. One major issue is the accuracy of the information provided, since correctly applying a rule is a complex exercise, not a simple act of reading. Additional risks involve the processing of personal and sensitive data, which may be leaked or over-shared without users' deliberate consent or awareness. Concerns also arise regarding algorithmic bias, since AI systems may reproduce discriminatory patterns embedded in training or design data. Furthermore, chatbots may raise issues concerning the unauthorized practice of law. This article argues that legal question-answering chatbots, regardless of their pre-designed purpose, may endanger users' fundamental rights, and that classifying them as limited-risk systems undermines the goal of ethical, responsible and trustworthy AI. It further argues that the transparency framework under Article 50(1) of the AI Act is insufficient and proposes embedding ethics-by-design and ethics-by-default through a set of concrete interaction safeguards to actively support verification of chatbot-provided information before reliance: cognitive forcing functions, interaction friction, scaffolded verification, and calibrated uncertainty and limitation cues.

Keywords: Legal Information Chatbots, Automation Bias, Trust, AI Act, Transparency, Ethics-by-Design

DOI: 10.54941/ahfe1008419

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← Designing Integrated Human-AI Systems: A Socio-Technical Approach to Knowledge Management in HealthcarePsychological Foundations of Epistemic Profiles for Misinformation Resilience: Implications for Defence-Oriented Agent-Based Modelling →
View all articles in Human Factors in Design, Engineering, and Computing →