AI still needs human assistance to make its case in court.
Data analytics and legal research company LexisNexis continues to involve very real lawyers in testing its AI models, Greg Dickason, CTO of legal and professional aspects at LexisNexis, said in an interview with InformationWeek.
Digesting large amounts of information is an area where AI excels, but if legal research goes wrong, lawyers could be summoned to court for interviews. There are more and more examples of AI fabricating lawsuits and being mentioned in court. Judge’s dissatisfaction.
AI legal advisor
Dickason called AI a transformative technology for the legal profession, noting its ability to reduce the time needed to investigate cases. However, he acknowledged that there is a lot of hype in the field, with model providers often pushing the praises of AI very strongly.
Because legal and professional users need context and accuracy in results, Dickason said LexisNexis does most of its AI development in-house, working with about 4,000 engineers who continue to evolve their knowledge along with the technology. “A significant number of them upskilled in the data science stream.”
In addition to changing the way lawyers investigate cases, AI has also changed internal engineering. “We’re doing more coding and building using agents like Claude Code, Codex, and those kinds of platforms,” Dickason says. AI currently accounts for approximately 30% of the LexisNexis ecosystem. He said he expects the number to rise to 70% eventually.
A big part of what LexisNexis does is take the equivalent of an open-book exam on tens of thousands of legal documents, Dickason said. This will be 1-2 pages for your model to use. He compared this to the more common large-scale model, where you might go through a million documents but lose that much content because you can’t digest it and use it effectively.
Impact of AI Dickason said the legal expert analysis is important because it produces summaries that help lawyers interpret vast amounts of information. For example, he said, AI can produce findings in minutes that previously took several junior lawyers an entire week to complete. “Large language models clearly help improve language performance, but they are not perfect.”
Efficiently finding relevant cases is not enough. The results must be examined in context to ensure that the precedent is true. To this end, LexisNexis is using AI models to test each other and asking hundreds of lawyers to evaluate the AI’s output, Dickason said.
The importance of context and accuracy
He said AI will make mistakes, and catching them early is part of preventing them from becoming systemic. These mistakes can be subtle if the model gradually introduces misinformation and then falls into its own fabrications.
The double-checking process that LexisNexis employs has the potential for AI to unstuck, reconsider answers, and recognize context that might have been overlooked, Dickason said. “In some ways, they’re incredibly smart. In other ways, they’re completely stupid… It’s like having a great mechanic who can fix any car perfectly, but often forgetting where the garage is or not knowing where the car keys are.”
LexisNexis was already using AI He said the search system needs to be improved before ChatGPT gains worldwide attention. After the public became interested in AI, LexisNexis introduced its own customer-facing AI models that could be used for research and drafting directly in the interface, he said.
Although AI can perform investigations in a tenth of the time compared to lawyers, that does not mean that all AI can be trusted for legal work. Dickason said lawyers who found themselves citing cases of AI-induced hallucinations may have turned to familiar AI and treated the findings as gospel. “They are using [same] These are the tools I’m using to create tonight’s recipe. “Then they will raise legal issues and take advantage of it, which is not very cool,” he said.
