Optimism and Caution (But Mostly Optimism) For the Future of GenAI In Your Law Firm

But what about the practicing lawyers, not just the HR, professional development, and non-legal staff?  A few practical use cases are useful.

• Imagine asking your genAI tool to read correspondence from opposing counsel requesting to meet and confer about a particular topic at a particular time.  Instead of personally reviewing all of your past correspondence, consulting calendar availability for you and your co-counsel, and re-examining pleadings and other court documents – and assuming your genAI tool had access to the appropriate matter-related information – attorneys could simply ask the genAI tool to draft a response, in a firm but professional tone, suggesting a meeting at least 8 days in the future, when all counsel are available, and suggesting the meeting focus on the most recent issue in contention between the parties. 
• Imagine asking your genAI tool to review a 300-page contract proposal in a complex intellectual property licensing and partnership arrangement and instructing it to cite to any paragraph that creates client obligations or responsibilities outside of certain pre-set parameters.
• Imagine receiving from the court an order to respond to opposing counsel’s eve-of-trial filing seeking disqualification of counsel for alleged conflicts of interest arising from the firm’s past representation of a witness in an unrelated, but similar, matter.

In each of these examples, the speed of communication goes up, efficiency goes up, client costs for attorney time and attention go down.  The floor of minimum quality drafting comes up.  The attorneys’ attendant stress decreases knowing they can rapidly produce the required materials, and they are more intellectually engaged because they can focus on the higher-order work of crafting litigation strategy, client engagement, and interface with the court, rather than the mechanics of first-drafting.

There are other possibilities, too.  With the right foundational data and well-crafted prompts, genAI can be a ‘thought partner,’ suggesting potential arguments, helping outline client proposals, and creating after-action summaries.  Beyond the written word, certain genAI tools also work with sound, images and video.  These can be used to create specific recreations or demonstratives for use in trial or presentations.  We think, though, that one of genAI’s superpowers will be its ability, in the near future, to create dynamic, live, infinitely configurable training scenarios that law firms and individual attorneys can use to develop practical legal skills, as well as nuanced interpersonal, managerial, or leadership capacities.

We imagine a digital training sandbox in which genAI can play the rule of opposing counsel and an expert deponent witness in order to improve deposition skills.  Or a judge, counsel, and reluctant witness to train examination and cross-examination.  Or belligerent opposing counsel intent on delivering an opening statement full of argument, in order to develop a sensitivity for when to object in such a situation.  Or six colleagues preparing for a significant client engagement event, where each has a slightly different understanding of the goals of the engagement and different personal interests that might shape their preferred scope of work.  Or a poor-performing associate to whom a new partner needs to give constructive (negative) feedback.  GenAI’s ability to respond dynamically, and uniquely to every interaction, means this sort of skills training no longer relies on scripts or multiple-choice scenarios.  Even running the same scenario consecutively will produce slightly different results. 

This can change how a law firm practices its business of delivering client service and also how it develops its attorneys and staff.

But.

But the future is not just a utopia of efficiency, and we have already seen examples of lawyers’ blind reliance on genAI without understanding its risks and of courts reacting in ways that may stifle innovation, increase client costs, and create additional barriers to access to justice.[4]  There are critical concerns—ethical imperatives—that lawyers need to be mindful of as they consider adopting genAI into their practice.

The first of these concerns centers on the tools themselves.  As described above, generative artificial intelligence is, basically, a very sophisticated language predictor.  When asked a question, it surveys its body of data and predicts, word by word, an answer that would likely be true.  Note, however, that it is not predicting truth, only a prediction of what truth might be.  Given the right data set and the right prompt, the answers provided are likely to be suitably and reliably accurate.  But not always.

Where the data, the model, or the prompting are lacking, genAI tools will generally ‘try their best’ to answer a question.  This predisposition to answering a question, satisfying the user, makes these tools susceptible to “hallucination.”  This is the possibility of creating a pleasing but factually inaccurate string of words.

The tool may ‘make up’ an answer.  Hallucination may occur as quotes or citations that do not exist, and this is, obviously, dangerous for attorneys.  They must understand, broadly, how genAI works and what the risks are to using it[5], not just its benefits.  It is (or should be) a standard of practice that attorneys fact-check and cite-check material produced by genAI.  Attorneys in a supervisory capacity—which includes any attorney working with another lawyer, paralegal, legal assistant, secretary, investigator, intern, or expert—should understand their obligation to oversee the work of those they supervise.[6]  Attorneys in a managerial capacity should ensure the practice groups, offices, and firms they manage have the training and are creating the culture necessary to use genAI effectively and ethically.[7]

Even more concerning is an issue that was not addressed at the conference.  Although terms of use may vary, most publicly available genAI tools capture user prompts and input to supplement their large language models.[8]  For lawyers using public tools like ChatGPT, Bard, Bing, and others, to experiment with genAI, this means that emails, pleadings, and other material they might use to provide context for a genAI prompt or request, will be subject to recording and use by the service provider.  Such a disclosure of confidential client information, even inadvertently, might constitute a violation of ethical obligations.[9]

[1] The generic term “genAI” is used to include the broad category of apps and services that leverage machine learning and large language models to provide new and unique responses to queries rather than returning particular terms or data via search.

[2] For simplicity, we refer to “law firms” with the intent of capturing all forms of legal services providers, including corporate general counsel offices, government legal offices, and non-profit legal providers.  The issues benefits and issues discussed here are relevant to every form of legal service delivery.

[3] The AI-generated text is provided verbatim.  Record of generation is available here - https://chat.openai.com/share/00fa1c55-82a5-4883-af5d-9e3bace1b969.