Generative Artificial Intelligence: How “Smart” Is Your AI?
There is no substitute for common sense. Today, clients want to know that their attorneys are up to date on all forms of cutting-edge technology. Generative Artificial Intelligence (“GAI”), a technology that allows a computer program to create new content and ideas, including legal briefs and contracts. Thus, GAI is a catchy, new trend that allows clients and attorneys to interact in new and different ways with information. However, there is a catch, and many people, including lawyers, misunderstand that a GAI computer program, like ChatGPT, is only as good as the information it contains and the programmer who created the GAI.
GAI uses machine-learning algorithms to process information, analyze large amounts of data, and create new information. The amount of data that GAI can analyze is enormous, beyond the capacity of a human. In order to do this work, GAI must “learn.” How does GAI learn? The same way children do, just on a larger scale. GAI can learn by trial-and-error or use other “learning” techniques to produce a better product. GAI analyzes the vast amounts of original source data to identify patterns and relationships between the words and ideas to create a new product that mimics the original. What is “machine learning”? The computer program’s algorithms analyze patterns in the original data set that build upon one another and draw inferences or conclusions from the patterns about what the new product should look like.
For instance, GAI can evaluate the political speeches that any politician has ever made and compose a new political speech. The content of the new political speech depends on the natural language algorithms that help the GAI “understand” the patterns of speech. The GAI’s new speech also depends upon the comparison speeches put into the GAI. Thus, the new political speech would be very conservative if only conservative speeches comprised source information. The new speech would also sound different if only speeches from the U.S. Senate or the British House of Commons were input. The new GAI’s political speech may not be able to address international farming issues well if the source information does not include farming data from South America, India or China. The bias of the source information directly affects the new product that GAI creates.
The same is true for legal GAI. The source information may include all U.S. Supreme Court cases, or every circuit court case decision from every county in the U.S. or elsewhere. Likewise, the algorithms used to compare and analyze a Supreme Court case and a judge’s decision from a small town may vary. Some GAI programs may weight specific terms more, while another algorithm may give more credence to a Supreme Court decision, giving a preference or drawing an inference based on the patterns of writing of Supreme Court justices.
Thus, the common-sense axiom of “garbage in, garbage out” applies to legal GAI. Indeed, early successes of legal GAI appear to depend heavily on the scope of the original data set. For instance, a county courthouse may input all of its own local rules and decisions into its legal GAI program, and a pro se litigant in that court may be able to understand better that particular court’s filing requirements and decisions if the GAI does not analyze or use original source data from other courts.
There is a famous movie where Robin Williams asks Matt Damon about quoting Shakespearean sonnets for love and smelling the Sistine Chapel. GAI’s new product might reference “Good Will Hunting,” but its algorithms could also misunderstand and misinterpret the prompts and all of the human reference points that allowed you, as the reader, to identify the movie. GAI might instead identify and insert “The Agony and the Ecstasy,” a 1965 movie about Michelangelo. The same is true for legal GAI, and a new lawyer or a judge reading a GAI legal brief or contract may not make the same references that a legal GAI does when it drafts the new product using an original data set of cases, contracts, legal textbooks and treatises or other source data.
Thus, depending on its algorithms, legal GAI can “hallucinate” or draw the wrong conclusions about a case, legal terms, the parties or the facts. The ethical rules and most courts now require a party using legal GAI to alert the judge that a brief was generated by legal GAI and that the lawyer signing the brief has reviewed the GAI brief and certifies it as true and accurate. Legal GAI is only as good as the lawyer who reviews the new product.
The contents of this Alert are for informational purposes only and do not constitute legal advice. If you have any questions about this Alert, please contact the Shulman Rogers attorney with whom you regularly work.
To receive Legal Alerts and other timely news and information from Shulman Rogers, please click HERE to subscribe.
Stay up to date with all the latest news and events.
