A human mock jury went 7 to 4 for the defense. Our simulated jurors agreed
In March 2026, attorneys at a DRI seminar in Nashville watched the first public human vs. machine mock trial. A live mock jury heard a dram shop case and returned a defense verdict. While the humans deliberated, two Viewpoints.ai scientists built a simulated panel on the same materials. Its most likely outcome, out of 10,000 combinations, called the case the same way, with the same themes on top. The account was published in DRI’s For The Defense.
The case
Penny Plaintiff traveled to Nashville for her bachelorette party. At Bull’s Bar she signed a waiver, rode the mechanical bull, and was thrown off, suffering an ankle injury that worsened over the evening. Later, on a pedal tavern headed to the next bar, she suffered a far more serious leg fracture when David, a patron who had been served two beers at Bull’s despite obvious signs of intoxication, sideswiped the vehicle.
At trial, Penny sought pain-and-suffering and punitive damages. The jury had to decide whether Bull’s Bar unlawfully served a visibly intoxicated patron, whether the bar’s conduct supported punitive damages, and whether her inability to walk unassisted at her wedding traced to the bull ride or to the later crash. The case is fictitious, written for the competition, and both juries heard it presented live by lawyers and an actor playing the plaintiff.
The verdict, side by side
The human mock jury was recruited in advance and deliberated on camera while the audience watched. The simulated panel was created after the presentations ended, while that deliberation was still running.
| Human mock jury | Simulated panel | |
|---|---|---|
| Who sat | Nashville citizensRecruited and moderated by JuryConnect | 100 simulated jurors10,000 twelve-juror combinations drawn from the pool |
| When the panel was built | In advance of the live event | While the human jury deliberated |
| Verdict | Defendant7 to 4 on the initial vote | Defendant7 to 5, the most common outcome |
The authors’ takeaway: the humans and the AI were generally in agreement, despite a difference of roughly 100 times in the hours and 10 times in the dollars needed to reach each result. The simulation’s most likely outcome, found in 23 percent of combinations, was the one that most closely matched the humans.
What counsel would have known before trial
The simulated panel returned its verdict and its themes the same day. Here is what each side would have taken from it.
For the defense
The case was already theirs. Across 10,000 combinations the plaintiff carried 23 percent, 20 percent deadlocked, and the defense took the rest.
The intervening act was the card to lead. More jurors cited David’s crash than any other factor, and jurors who credited the bar’s warning signage moved a further 35 percent toward the defense.
For the plaintiff
The case lived on one theme. The re-serve after the spill and service while visibly intoxicated were the reasons jurors crossed over. Proving the service claim was the whole game.
Then run it again. Add the surveillance video, recut the service timeline, drop the waiver, and watch the split move. The live event scored one version of the case. A panel this fast to rerun can score them all.
How the head-to-head was run
A live, public benchmark on the same case, scored in the open and published by DRI.
Tested against live juries
The DRI head-to-head put a simulated panel against a human mock jury in public, and the two agreed. Book a demo to see more of the benchmarks and to run a venue-matched panel on one of your live matters.
Book a demo