Arkansas Public Defenders 2024
Michael Naughton presented to the Arkansas Public Defender Summit in November, 2024 on "Learning how to Utilize Justice Optimized to Power AI in Practice". The presentation focused on using AI tools, specifically large language models (LLMs), in a legal context. The presentation covered various aspects of AI, prompting techniques, and considerations for ethical and effective usage.
Key Points:
Introduction to AI: The presentation introduces different types of AI, focusing on chat-based models such as OpenAI's ChatGPT (including 4o and Preview o-1), Google's Gemini, Facebook's Llama3.2, Perplexity, and Claude 3.5. It also briefly touches upon the use of 3D models in court cases, specifically for visualizing locations.
Prompt Engineering: A significant portion of the presentation is dedicated to AI prompting. It emphasizes that AI is only as good as the prompts it receives. Effective prompts should be communicative, strategic, and on average, around 21 words. It outlines four main areas to consider when writing prompts:
Persona: Defining the role the AI should assume (e.g., criminal defense attorney, probation agent).
Task: Clearly stating what the AI should do (e.g., conduct research, draft a memo).
Context: Providing background information related to the task (e.g., details about a client's history).
Format: Specifying how the output should be presented (e.g., bullet points, long-form paragraphs).
Examples and Tips: The presentation includes numerous examples of prompts and follow-up prompts, covering scenarios such as summarizing letters, brainstorming arguments, and researching specific topics. It also offers tips for prompt writing:
Use natural language.
Be specific and provide context.
Specify the desired format.
Continue the conversation and refine prompts as needed.
Don't push AI into hallucinations (incorrect responses).
Consider the voice and persona, and maintain personal advocacy.
AI Limitations and Ethical Considerations: The presentation highlights limitations and ethical concerns related to using AI in legal settings. It specifically quotes the terms of use from OpenAI and Meta Llama 3, emphasizing that AI output should not be used for legal or material impact on individuals. It also discusses data collection policies of Google Gemini.
AI Model Comparison: The presentation briefly compares different AI models:
ChatGPT: Described as the gold standard, with fine-tuning capabilities.
Google Gemini: Noted for its integration with Google Workspaces.
Llama 3.2: Focuses on local hosting for increased security.
Claude 3.5: Mentioned for its robust LLM and ability to control computers.
Each model has its own data usage and policy considerations.
Application in Legal Practice: The core message is that AI should be used as a tool to enhance legal work, not replace it. Attorneys should use AI to:
Construct compelling narratives for their clients.
Research areas relevant to their cases.
Brainstorm ideas and perspectives.
Craft richer sentencing memorandums and verbal allocutions.
Avoid simply copying and pasting AI output.
Conclusion: The presentation concludes by emphasizing that the act of crafting prompts is crucial. This process helps attorneys actively analyze cases, investigate variables, and brainstorm persuasive arguments. The real value of AI lies in the way it can assist attorneys in thinking more deeply and comprehensively about their clients and their cases.
Link to the presentation here.
