What happens when researchers ask AI to perform the tasks traditionally carried out by jurors?
I have recently seen the question of AI replacing jurors appear repeatedly on LinkedIn, and to be honest it petrified me. Every time another article appears suggesting that AI could take over another area of law, it makes me think: what does this mean for those of us preparing to enter the profession?
Questions like these are the reason I began Redline. Not to scare future or trainee lawyers, but quite the opposite. The purpose is to step away from the predictions and look at what has actually happened. By looking at the evidence, I hope to provide clarity and knowledge rather than confusion and fear.
Before answering the question, it is important to understand why the question exists in the first place.
Imagine two juries are given the exact same case.
The first has twelve people. They have spent weeks listening to the evidence, reading documents and discussing what they believed happened. One juror might be influenced by personal experience. Another might misunderstand an important piece of evidence. A third may simply miss their children and be tired after a long day in court. With human judgment comes human feeling, and decisions can end up resting on emotion rather than fact.
Now imagine an AI jury. It processes thousands of pages in seconds, compares every piece of evidence and reviews the previous cases on which similar rulings have been based. It does not get tired. It does not carry personal bias. It presents the case as it is.
Framed that way, it is easy to see why some argue that AI should at the very least be involved in the jury process.
There is real evidence that human juries are influenced by more than the facts in front of them.
In 2009, Nicole Waters and Valerie Hans studied nearly 3,500 jurors who had decided felony cases.1 They asked jurors what verdict they personally would have reached if they had been the only juror, then compared that answer with the verdict their jury actually returned.
More than one third said they would have voted differently from their jury's final decision. The power of the group had shifted their position, even though privately they believed something else.
This is the strongest version of the argument for AI. If group pressure can move a third of jurors away from their own honest view, why not use a system that cannot be pressured at all?
This is where the evidence becomes genuinely interesting, because someone finally tested it rather than speculating about it.
In 2025, Hayley Stillwell and Sean Harrington of the University of Oklahoma College of Law published a study with the memorable title Michael Scott Is Not a Juror: The Limits of AI in Simulating Human Judgment.2 They asked a very specific question: can large language models reproduce the way real human jurors interpret evidence and reach decisions?
Using a mock trial scenario, the researchers tested models including GPT, Claude and Gemini, and compared the results against real human juror data.
The models did not reliably reproduce the reasoning of human jurors.
What makes the study particularly valuable is what happened next. When the researchers used real human juror data to fine-tune the models, the results improved. The machines got closer, but closer is not the same as equivalent.
| The Claim | What the Study Found |
|---|---|
| AI can reason like a juror | Not reliably, in its standard form |
| AI performance is fixed | Improved when trained on real juror data |
| AI can replace jurors | Not supported by the evidence |
So where does this leave us?
The argument for AI juries rests on a genuine problem. Waters and Hans showed that human juries are shaped by group pressure in ways that should concern anyone who cares about fair verdicts. That problem is real and it has not gone away.
But the response to a flawed system is not automatically a machine. The most direct empirical test we have found that large language models could not reliably replicate how human jurors actually reason, and that matters enormously. A jury is not simply a device for producing a verdict. It is a process of reasoning, deliberation and applied human judgment, and the research suggests that this is precisely the part AI struggles to reproduce.
For those of us entering the profession, the takeaway is not that our future is secure, nor that it is doomed. It is that the loudest predictions online are running considerably ahead of the evidence. The question of AI and juries is being tested carefully by researchers, and so far the answer is far less dramatic than LinkedIn would have you believe.
That is worth knowing before you let a headline petrify you.