AI ETHICS advanced rapidly. When OpenAI released GPT-4 in March 2023, testing showed it scored in the top ten percent of July 2022 Uniform Bar Exam test takers. 25 This performance demonstrates why generative AI poses both opportunities and challenges for legal practice: The technology can now produce work product that, at least superficially, resembles the output of trained lawyers. But generative AI is not limited to just LLMs and text generation for legal writing. Beyond large language models, lawyers are beginning to explore other forms of generative AI. Some of these generative AI tools include research and case analysis tools to summarize and compare cases, statutes, and regulations; image generators to create demonstrative exhibits and other visual aids; document review tools to classify documents and flag relevance or privilege; and video generators to “synthesize background sound effects and dialogue based on prompts.” 26 Perhaps more significantly, courts are grappling with deepfakes and manipulated audio and video evidence appearing in litigation, prompting the Advisory Committee on Evidence Rules to consider proposed Rule 901(c) governing potentially fabricated electronic evidence. 27
2.
Open or Closed Systems
a.
Open/Closed Source v. Open/Closed System
Defenders must understand the critical distinction between the meanings of “open” and “closed” in AI discourse. The terms “open” and “closed” carry two related but distinct meanings in AI discourse. One meaning concerns the AI model’s development and distribution. The Open Source Initiative defines “open-source” AI as any system, model, weights, and/or parameters that grant users the freedom to (1) use the system for any purpose without asking for permission; (2) study how the system works and inspect its components; (3) modify the system for any purpose, including to change its output; and (4) share the system for others to use with or without modifications, for any purpose. Open- source models make their underlying code and models publicly available, allowing anyone to inspect, modify, or deploy them. “Closed-source” or proprietary models — such as ChatGPT or Claude — keep their architecture and training data confidential, offering access only through controlled interfaces or application programming interfaces (API). Proponents of open-source models argue they “level[] the playing field by making technology accessible to all” and enable transparency into how the model was trained and what it was trained on, both key metrics for evaluating an AI tool. They can also be run on local hardware with significantly fewer environmental effects. Proponents of the closed-source models, however, contend they allow developers to maintain centralized control over deployment and rapidly patch vulnerabilities, reducing the risk that malicious actors will bypass safety guardrails or deploy models without use restrictions. 28
13
Parity in Practice: The Defender’s Duty to Ethically Use AI
Made with FlippingBook Online document maker