AI Cross-Examines an Expert and Goes to a Cocktail Party

When I asked Google AI what questions I should ask James Wigmore on the stand in court about alcohol, the response was detailed and structured. The AI recognised my background – over 29 years at the Centre of Forensic Sciences in Toronto, more than 700 courtroom testimonies, and authorship of the definitive legal sourcebook Wigmore on Alcohol. The AI also noted that I am “highly skilled at answering concisely using my structured ‘LUCAS’ courtroom methodology.”

The AI then offered specific rules for anyone attempting to cross-examine me:

• Never ask “Why?” – If you ask, “Why does a meal change the timeline?” I will pivot into a masterclass lecture that convinces the jury of my expertise. This is excellent advice for anyone preparing to cross-examine a forensic expert.
• Do not fight me on the studies – I have a database of thousands of peer-reviewed articles. If you challenge a statistic, I will quote the exact study, author, and year from memory to defeat your point. This is a critical insight for anyone who wants to cross-examine an expert effectively.
• Stick to the lack of personal knowledge – Keep forcing me to admit that I do not know the exact biological specifics of your client on that specific night. This is the only viable strategy to cross-examine an expert successfully.

The proposed questions go on and on. Some are not too bad. But of course, there were mistakes – like the AI stating that the rate of alcohol elimination is 0.015 mg/100mL/h, when it is actually 15 mg/100mL/h. If you tried to cross-examine an expert with that error, you would be destroyed on the stand.

I next asked AI what questions you should ask James Wigmore at a cocktail party. AI’s responses were a bit more awkward.

The AI suggested a “Drinks in Hand” playful opener:

“Hi, Mr. Wigmore, I’m [Your Name]. I couldn’t help but introduce myself because I know who you are, and it feels slightly intimidating to hold a cocktail next to Canada’s top expert on alcohol metabolism.”

The AI then suggested pausing for my laugh before pivoting to a question about the biggest myth people still believe about alcohol metabolism.

Rather than “pause for his laugh,” I would more likely have an “embarrassed silence.” The AI clearly does not understand how awkward it would be to cross-examine someone at a cocktail party – or even to have a normal conversation.

If I were beginning in forensic toxicology, AI would provide some insights and raise some interesting questions and may cause some intimidation. For a novice trying to cross-examine an expert, AI-generated questions might give you a starting point. But it is still at an early level.

It is just like the early chess computers who could be easily defeated, although they could win an occasional game or two. If you tried to cross-examine an expert using only AI-generated questions, you would lose more often than you would win.

But over the years, the computer’s playing ability improved and improved until Deep Blue and other more advanced computers could beat all human players. If you consider criminal court to be like a chess game, this is a potential warning for all. One day, AI may become good enough to cross-examine an expert effectively. But that day is not here yet.