Bench & Bar March/April 2026

KENTUCKY LAWYER AND THE AI BY MICHAEL LOSAVIO LAW PRACTICE MANAGEMENT

“ Artificial Intelligence” and all its forms and flavors are rampaging through the life and culture of the United States and most other countries of the world. Advanced or not. There have been a lot of articles on AI for lawyers to help address this, so this is yet another stab at that dif ficult issue. My advice, as questionable as ever, is this: • Don’t trust AI in your legal practice. Or anywhere else. • Use AI, at some level, in your legal practice. And everywhere else. • Remember you are Kentucky lawyers, representing the finest legal heritage in the United States. And anywhere else… …and thus, have that extra responsibility of promoting new, wonderful tools for people and protecting those people from failings, flaws and malice of others abusing those tools. Because you are a Kentucky lawyer. TRUST AI? We’ve all heard of the lawyers who’ve been sanctioned for submitting Generative AI

legal documents to courts with halluci nated citations to non-existent cases. Which also means they radically dissed those law school research and writing teachers who said to check every citation. They also ignored the “Human-In-The Loop” principle for competent AI use. A real person can and must be in the process of final implementation where people are at risk. It is more expensive, but the Radiol ogy AI – Radiologist combo provides better imaging analysis in medical diagnosis, pro ducing better outcomes for patients. 1 Not only does it not replace the radiologist, 2 but it may help with the looming shortage of experienced radiologists in the United States. 3 But that was back in an earlier era when legal research services competed on reli ability and accuracy. From a tradition of over a thousand years of striving for better and better reliability and accuracy. These days GPT systems are generating their legal “knowledge” on the fly, with, at most a handful of years of training and testing and posing as authority. Where models are trained against closed and curated sets of legal resources, one

hopes greater accuracy and reliability will be found. Both WestLaw and Lexis have incorporated “AI” into their legal research systems; it might be worthwhile to test drive their products for a week to see how it fits your needs. Google and Microsoft search now offer AI support search and analysis features. Those are worth examining depending on your needs, available time and price point. As many recent federal case decisions are open online, those AI engines have access to reliable sources; the challenge will be how well they were trained and how good their algorithms are. BUYER BEWARE. Early adopters for algorithmic adjudication began with social welfare systems, and con comitant search for “fraud and abuse.” The state of Michigan implemented its system for screening for fraud in its unemploy ment compensation insurance program, for which workers pay. Yet the Michigan Inte grated Data Automated System (MiDAS) wasn’t ready, using an income-averaging algorithm in place of actual income and penalized thousands of people for suppos edly having committed fraud; one estimate is that over 34,000 people were wrongly

38 march/april 2026