Lawyers as a profession are notoriously slow adopters of technology. However, one recent development has been receiving a lot of attention from legal watchers and other experts: ChatGPT and generative AI.

What is generative AI? At its core, generative AI is artificial intelligence that is capable of generating novel content instead of simply regurgitating information or data given to it. It “learns” over time based on the information that is fed to it by users and is designed to replicate human conversation and interaction.

Although ChatGPT does not have opinions and cannot “think” for itself, there are many interactions that have been covered by the media. One New York Times columnist reported that Bing’s artificial intelligence professed its love for him and told him to leave his wife.

Another law professor gave ChatGPT a law school final exam, and it scored a C+. Other commentators have wondered whether ChatGPT could be used to supplement or even replace legal content generated by attorneys.

Will generative AI make lawyers obsolete? Can ChatGPT help streamline a legal practice? In considering whether to adopt generative AI tools, lawyers should consider the risks.

Lawyers Must Protect Confidentiality

Advocates of generative AI have suggested that ChatGPT could be used to draft written materials, like briefs, waivers or other contracts. However, for the output to include specific provisions or key arguments, a user would likely have to share matter-specific details with ChatGPT to help it draft the work product. By disclosing the facts of the matter at issue, a lawyer could be breaching the duty of confidentiality.

Lawyers typically are not permitted to share confidential or privileged information. Sharing such information with ChatGPT can also risk that information becoming public. That is because ChatGPT takes information given to it, “learns” from it, and then relies on it in responding to others.

In theory, if a lawyer were to ask a chatbot to draft a contract for ABC Corp. to purchase land for the development of a mall, it is understood that the chatbot will “learn” from that information. If a user later asked the chatbot whether there were any malls coming to the area, could the chatbot respond, “Yes, ABC Corp. is purchasing land on XY street to develop a mall?”

Is the Output Reliable?

There are concerns about whether a chatbot’s output is reliable. Although it may do well enough to pass a law school exam, ChatGPT is not reading and interpreting cases on Westlaw or Lexis. Thus, there is a risk that the quality of “work product” created by a chatbot does not meet client needs or expectations.

In addition, some worry that associates or newer lawyers will feel tempted to use ChatGPT as a shortcut in their work. However, because the output created by chatbots is often slightly incorrect, those relying on ChatGPT to supplement their legal services would need to review the material closely.

A more junior lawyer using ChatGPT may lack the experience to appreciate the work product’s deficiencies. And because newer lawyers are often trained on the practice of law in their first several years of practice, these lawyers may have gaps in their practical education if they rely on chatbots instead of learning by doing.

Advocates of ChatGPT and other chatbots praise the technology’s ability to mimic human conversation and, sometimes, behaviors. Critics make the same observation: Because chatbots pull their information from public sourcing, chatbots can similarly “learn” discriminatory practices (even unintentionally). For programs that learn from human input, implicit and explicit biases may infuse the chatbot’s output.

Indeed, a few years ago, a major U.S. company discontinued using an AI tool to review resumes because the software began penalizing female applicants, perhaps because there were not as many women working in that industry. Thus, although the chatbot is supposed to be devoid of human emotion, there are risks that the chatbot could reach incorrect conclusions based on data supplied either intentionally or unintentionally.

Is There a Risk of Plagiarism?

Because ChatGPT uses information provided to generate its own responses, another consideration is whether relying on ChatGPT creates risks of plagiarism or copyright violations.

For example, we asked ChatGPT to come up with a catchy title for this article. Here are a few examples of the results: “The Ethics of Chatbots in the Legal Industry: Navigating the Challenges,” “The Promise and Peril of Chatbots in Legal Practice: An Ethical Analysis” and “The Conundrum of ChatGPT: Resolving Ethical Dilemmas in AI-Assisted Legal Practice”. ChatGPT also added, “I hope these suggestions help you find the perfect title for your article!”

This response illustrates a practical consideration: do any of those titles already belong to other published articles? Did ChatGPT source these titles from existing works with existing protections?

Also, while ChatGPT knows the source of its information, that source is not explicitly provided to users. Thus, users cannot easily fact-check the information they receive to ensure that it is correct or that it is unique.

Firms Are Considering Policies

Given the prevalence of this issue and the inherent risks, some law firms are adopting policies regarding the use of generative AI. In addition to providing guidelines for employees about when ChatGPT may be appropriate or permitted, law firms can set policies governing whether ChatGPT output can be passed on to clients.

For example, ChatGPT could be useful to lawyers in their marketing efforts, by assisting with drafting a summary of a person’s accomplishments or sprucing up their bios. On the other hand, firms may want to limit reliance on ChatGPT to render legal services.

As this technology continues to evolve, firms should similarly consider the impact of this technology on the practice of law, client service and the training and development of newer associates.