AI ETHICS Moreover, AI systems can perpetuate and amplify biases present in their training data or design, especially when used in the context of the criminal legal system. 63 When AI is trained on historical data reflecting discriminatory patterns, the AI may reproduce those patterns in its outputs. This concern applies across AI applications, from facial recognition systems that perform worse on darker-skinned faces to risk assessment algorithms that may disadvantage defendants from communities of color. Professor Brandon Garrett and Professor Cynthia Rudin have documented bias concerns in criminal justice AI. They note that “criminal justice data is often noisy, highly selected and incomplete, and full of errors.” 64 When AI systems train on this flawed data, they may encode existing biases rather than correcting them. A risk assessment tool trained on historical sentencing data, for instance, might perpetuate racial disparities present in that data. Defenders should be alert to bias in AI tools they use and in AI tools prosecutors employ. When AI outputs seem to disadvantage clients based on protected characteristics, defenders should investigate whether bias might explain the results. Challenging biased AI evidence requires understanding how bias enters AI systems and how to demonstrate its effects. When a defender lacks AI competence, at least three options exist. First, the defender can decline matters requiring AI assistance. Second, the defender can invest the time necessary to develop adequate AI knowledge and skills. Third, the defender can associate with another lawyer who possesses the necessary technological competence. A defender simply cannot proceed with AI-assisted representation while remaining ignorant of how the technology works and where it might fail.
2.
Rule 1.6 – Confidentiality of Information
Model Rule 1.6 prohibits lawyers from revealing “information relating to the representation of a client” without informed consent or implied authorization to carry out the representation. The rule further requires lawyers to “make reasonable efforts to prevent the inadvertent or unauthorized disclosure of, or unauthorized access to, information relating to the representation of a client.” Defenders should note that this Model Rule specifically addresses the ethical duty to keep all information about the client confidential. Attorney-client privilege is a separate evidence rule, although privileged information is considered confidential. AI systems pose significant confidentiality risks. Depending on the system’s architecture, the information might be stored on remote servers, used to train future AI models, or become accessible to the vendor’s employees or contractors. Even if the AI provider promises confidentiality, data breaches remain possible, and the lawyer has limited ability to verify or enforce the provider’s security practices.
27
Parity in Practice: The Defender’s Duty to Ethically Use AI
Made with FlippingBook Online document maker