Winston's AI Monthly Recap - July 2024

Judges Weighs In on Use of GENAI

The Fifth Circuit recently declined to adopt a procedural rule which would have required disclosure of the use of generative AI in court filings, citing existing federal court rules as sufficient to police misconduct. The court warned that “‘I used AI’ will not be an excuse for an otherwise sanctionable offense.” At the lower court level, a Northern District of Texas judge recently adopted a similar rule requiring certification of AI use and human review. Meanwhile, an Eleventh Circuit judge recently wrote, in a concurring opinion, his experimentation using generative AI in legal research, specifically to interpret the term "landscaping" in an insurance policy document. The judge concluded that large language models might be useful in interpreting legal texts, but cautioned about potential pitfalls such as hallucinations and bias.

New Hampshire Also Considers AI Bills

A New Hampshire’s state legislature is considering four bills relating to AI: 1) HB 1432-FN, a bill criminalizing the fraudulent use of deepfakes, establishing penalties including class B felony charges for creating or distributing deepfakes intended to cause harm, and creating a private right of action for victims, 2) HB 1500-FN, a bill criminalizing distribution of “synthetic media” with intent to mislead others about the acts of a person or influence an election, with stricter penalties for violations close to elections, 3) B 1596-FN, a bill requiring the disclosure of AI usage in political advertising, mandating clear labels on synthetic media in political campaigns and prohibiting distribution of such media within 90 days of an election without proper disclosure, and 4) HB 1688-FN, a bill regulating AI use by state agencies, prohibiting its use for discrimination, surveillance, or deceptive purposes, and requiring human oversight for AI-generated decisions affecting citizens’ rights.

Research Into ‘Hallucinating’ Generative Models Advances Reliability of Artificial Intelligence

Researchers at the University of Oxford have made strides in preventing AI hallucinations by coding AI models to tell the difference between situations in which the model is uncertain about the truth of the underlying information, or just uncertain about how to word the answer. In new models, when there is uncertainty about the truthfulness of the answer, models can inform users that the answer is potentially untrue due to a lack of knowledge. Hallucinations present a challenge to the widespread adoption of Gen AI technology, and this research indicates that hallucinations are something that developers may be able to potentially overcome in the near future. 

Pennsylvania Ethics Opinion

The Pennsylvania Bar Association’s recent Joint Formal Opinion No. 2024-200 [text] provides guidance on the ethical use of artificial intelligence in legal practice, addressing issues such as competence, confidentiality, and accuracy. Key insights include the need for lawyers to be technologically competent, safeguard client information when using AI, verify all AI-generated citations and content, and communicate transparently with clients about AI use. The opinion also highlights the importance of monitoring AI as one would an employee or vendor, ensuring unbiased and accurate information, and adhering to proper billing practices given AI’s time-saving capabilities. While the guidance is specific to Pennsylvania-licensed attorneys, it offers insights for legal professionals nationwide as they navigate the integration of AI into their practices.

The NYSBA’s Task Force Report on Artificial Intelligence

The New York State Bar Association (NYSBA) Task Force on Artificial Intelligence released a detailed report examining the legal, social, and ethical impact of artificial intelligence (AI) and generative AI on the legal profession. The report, approved by the NYSBA House of Delegates on April 6, 2024, provides an in-depth look at the evolution of AI technology, its potential benefits and risks, and its implications for legal practice. Key takeaways include: Task force recommendation that the NYSBA adopt a set of Generative AI guidelines and establish a standing committee to oversee periodic updates of the guidelines; emphasis on education of judges, lawyers, law students, and regulators; advising legislators to identify risks associated with AI that are currently not adequately addressed by AI, and exploration of AI governance, such as the expression of social values and reducing the risk of AI-generated misinformation, also known as hallucinations.

Colorado General Assembly Passes Senate Bill 24-205

The Colorado General Assembly recently passed Senate Bill 24-205, which aims to protect consumers from algorithmic discrimination and their interactions with AI systems, and is scheduled to become effective February 1, 2026. The bill defines key terms such as "algorithmic discrimination" and "high-risk artificial intelligence systems," and imposes specific duties on AI developers and deployers to implement risk management policies and provide transparency to consumers. More specifically, key provisions of the bill include requirements for developers to disclose the purpose, benefits, and risks of high-risk AI systems, and for deployers to notify consumers when such systems are used to make consequential decisions, provide explanations for adverse decisions, and offer opportunities to appeal. The Attorney General will have exclusive enforcement authority, with violations constituting an unfair trade practice under the Colorado Consumer Protection Act.

California Civil Rights Council Releases Proposed Regulations To Protect Against Employment Discrimination Resulting From the Use of Artificial Intelligence, Algorithms, and Other Automated Decision-Making Systems

On May 17, the California Civil Rights Council released proposed regulations to protect against employment discrimination resulting from the use of artificial intelligence, algorithms, and other automated decision-making systems, with aims to clarify how existing anti-discrimination laws apply to these emerging technologies, such as if they result in discrimination based on protected characteristics. The proposed regulations would also mandate that employers maintain employment records, including automated decision-making data, for at least four years, and emphasize that AI tools do not replace individualized assessments when evaluating applicants' criminal histories. The Civil Rights Council encourages interested parties to submit public comments by the July 18, 2024 deadline, and a public hearing on the proposed regulations will be held on the same date.