The future of legal AI is bright, but only if attorneys and engineers learn how to talk to one another. That requires a better understanding of how each’s educational and professional experiences shape their expectations and worldviews.
Having these conversations early and frankly can improve trust between these two professions with distinct foundations. When lawyers and AI engineers work together, they can make fantastic products that get used—not just elegant toy models that get close but not close enough.
Generative AI engineers and probably computer scientists, in general, have more precise mental models of how a large language model might behave—not just from repeated use but from their work and even their prior math experience. They are sensitive to how different prompts might trigger different outcomes or retrievals. When a large language model does not perform as expected, they understand how they might iterate on a prompt to improve it.
On the other hand, lawyers see a blank text box with no guidance on how to structure an inquiry, without a clear sense of the underlying mathematical concepts that ground large language models. Lawyers don’t need to study linear algebra to write good prompts (if they did, AI would be useless). Still, they need engineers’ guidance about how generative AI works and how to structure effective prompts for the particular large language model they are working with.
There are interface decisions software engineers can also make—like including citations for cases—that help guide lawyers to productivity with generative AI. But at the end of the day, they need support and prompt practice in the same way that senior associates need practice managing with junior associates.
Source: Forbes
Leave a Reply