L'algoritmo è uguale per tutti? | Ernesto Belisario | TEDxLink Campus University

Quick Overview

The algorithm is not yet equal for everyone in the justice system, as demonstrated by studies showing AI bias against minorities, but the speaker argues that carefully designed AI can help make justice systems less arbitrary, less unjust, and more truly equal by augmenting human decision-making.

Key Points: The legal maxim "The law is equal for everyone" is currently challenged by the use of AI in judicial settings, as illustrated by a Florida case where a Black woman received a high-risk score (8) while a white man received a low-risk score (3) for similar crimes. A study by researchers at Harvard found that AI risk assessment tools, like the one used in Florida, systematically show bias, disproportionately flagging minorities as high-risk for recidivism. The French legal system passed a law in 2019 that explicitly forbids the use of generative AI (like ChatGPT) in judicial activities, including interpretation of law, fact evaluation, and adoption of rulings. The speaker points to the existence of a database tracking "AI Hallucination Cases," which documented 597 instances of fabricated legal citations or arguments generated by AI in court filings. The speaker suggests that the solution is not to ban AI entirely, but to use it as a tool to support human judgment, making the system less arbitrary and more equitable. The bias in AI systems stems from the data they are trained on, often reflecting existing societal biases, such as those prevalent in North American and Western cultural profiles.

Context: This TEDx talk, titled "L'algoritmo è uguale per tutti?" (Is the algorithm equal for everyone?), is delivered by Ernesto Belisario, an attorney from the Ilexia Legal Studio. The presentation explores the ethical and practical implications of integrating Artificial Intelligence, particularly large language models like ChatGPT, into the justice system, contrasting the ideal of legal equality with the reality of algorithmic bias demonstrated in risk assessment tools.