Albert Einstein once purportedly said, “If you can’t explain it simply, you don’t understand it well enough.” This statement holds particular relevance for lawyers navigating the realm of generative AI models. A foundational skill taught in legal training is the art of crafting questions. Lawyers are trained to focus on phrasing clear and direct queries, a skill that translates well into creating prompts for AI systems.
The effectiveness of an AI’s response is contingent on the simplicity and clarity of the prompt it receives. Generative AI models are often likened to a literal-minded 6-year-old, requiring uncomplicated and straightforward questions to function effectively. Poorly worded prompts can result in erroneous outputs, a point emphasized by Wired’s Brian Barrett, who highlighted instances where Google’s AI provided misleading information with undue confidence.
For legal professionals, the task is reminiscent of examining witnesses where questions must be precise. An imprecise prompt can lead an AI model astray, much like a vague question can fluster a witness. For example, the innocuous question “Do you have the time?” can yield varied answers if not properly contextualized. Similarly, complex prompts, such as “Weave this year’s executive orders into a theme,” can confuse AI models trained to interpret questions literally, resulting in irrelevant responses related to actual weaving.
Legal practitioners leveraging AI must hence recall their foundational communication training and avoid complex analogies, excessive jargon, and ambiguous phrasing. They should aim to construct straight-to-the-point prompts to prevent AI “hallucinations”—instances where the model fabricates information. This approach not only minimizes confusion but also enhances the effectiveness of AI as a valuable tool in the legal industry.
As Kenneth N. Rashbaum, a partner at Barton, suggests, grounding queries in factual context and simplicity is critical. Though it may take several attempts to refine a prompt for the desired response, the well-honed questioning techniques from legal training are invaluable in shaping AI interactions. For further insights, explore Rashbaum’s full commentary here.