AI blindness in the courtroom

Connectez-vous pour écouter l'article

Something has been changing in the way self-represented litigants arrive in court. I do not have numbers, but I have a pattern I watch for.

A decade ago, I started calling these litigants Google lawyers. They appeared in court having clearly done their own research. They typed their problem into a search engine, found something which looked like an answer, and tried to apply it. The translation was often imperfect. Sometimes badly so. But the effort had a quality I respected. They had found a piece of the map, and they were trying to read it themselves. They knew they were navigating, even if they did not always know the destination.

What I see now is different in kind, not degree. The litigant who arrives today has not navigated anything. An AI tool did the navigating for them. It took their problem, processed it, and produced a confident, coherent, legally flavoured conclusion. The person before me believes they have the answer. What they do not have is any understanding of how they arrived at it, any ability to interrogate the path, or any awareness that the path may have been wrong from the first step.

I have started calling this phenomenon AI blindness. The pattern is not new. Self-represented litigants have long arrived in court with confidently wrong information from authoritative-sounding sources. What is new is the polish and the speed. AI blindness is not a failure of intelligence. It is a structural feature of how these tools work. It creates a problem in the courtroom that the profession has not yet named clearly enough.

The closed loop

Here is the pattern—confident arrival at a conclusion, with no foundation beneath it. The confidence is not incidental. The tool bakes it in. These tools do not produce tentative answers. They produce complete ones.

From the bench, I see two closed doors in sequence. The first is the confidence. Confidence, even misplaced confidence, closes the listening ears. The person before me is not waiting to be corrected or educated. They are waiting to be agreed with. Getting past that is slow, careful work, and it does not always succeed.

The second door is the comprehension gap. Even when the confidence gives way, the litigant often cannot follow the correction. They may have misapplied the law, applied the wrong law entirely, or built an argument with a structural flaw which requires baseline legal literacy to see. The tool did not provide that literacy. It skipped that step.

The cruelty here is older than the tool. The technology appeared to offer something to people who could not afford a lawyer. What it actually offered was a conclusion without the reasoning behind it. And yet, the reasoning was exactly what the person needed.

What the bar owes

My first piece on this subject was about a verifiable failure: fabricated citations. This piece is about something different. It is a more subtle failure which can sit beneath the same use of these tools.

AI blindness is not only a problem for self-represented litigants. The same pattern can show up across the counsel table.

Verification is necessary but not sufficient. A lawyer who can confirm a case citation exists has not necessarily understood the case, the line of authority it sits within, or whether it actually supports their argument. Verification checks the output. It does not reconstruct the reasoning. When counsel can explain the path from problem to conclusion in their own words, without the tool, the work has been done. When they cannot, it has not. The answer is not the analysis.

The Rules of Professional Conduct already speak to competence. Whether they need to speak more specifically to this is for the bar and the law societies to decide. I have said what I see.

What the bench owes: recognition

The bench carries obligations here, too. The first is recognition.

A judge who cannot recognize AI blindness in the person before them cannot respond to it. That sounds obvious, but it is not yet universal.

This is institutional learning of a familiar kind. The bench has done it before, and with self-represented litigants generally. The litigant before me is still a self-represented party. What has changed is how they got here. The existing work the bench has done on self-representation is the foundation. AI blindness is a new feature on top of it.

What does it look like? The confident arrival. The inability to explain the foundation. The closure that descends when anyone questions the answer. The recognition has to come first. Everything else flows from it.

What the bench owes: in the courtroom

Sometimes, a self-represented litigant has trusted a tool which has led them astray. The obligation is to make sure they get a fair hearing. That obligation is also old.

The duty to ensure, so far as possible, that a self-represented litigant receives a fair hearing is not a new development in response to AI. It has been part of this work for as long as there have been unrepresented parties in courtrooms. What is new is the reason they need assistance. While the reason has changed, the obligation has not.

In practice, this means investing the time and the space. It means slowing down. It means finding a way past the closed door of confidence, carefully, without humiliating the person who trusted a tool because they had no other option. It means explaining, and then explaining again, in plain language, why the answer they arrived at may not be the right one for the matter before them. This is not efficient work. It is necessary work. I know what I am asking, and I know the system does not have the slack to absorb it. The obligation does not wait for the slack to arrive.

What counsel owes: in the courtroom

A narrower question: I raise it and leave it for the bar.

When counsel can see, from across the counsel table, that a tool has led the other side somewhere wrong, what does counsel owe? The adversarial system does not require lawyers to make the other side’s case. It does require them to act as officers of the court. The Rules of Professional Conduct already address candour to the court and conduct toward unrepresented parties. Whether they cover this situation specifically is for the bar to say.

Where are we now?

This piece will not close the access-to-justice gap that creates self-represented litigants. Neither will anything else the bench and bar can do alone. The people who come before me, who have trusted an AI tool because they could not afford a lawyer, are not the problem. They are evidence of a problem much larger and much older than any technology. AI blindness is the symptom. The gap is the cause.

Our obligations do not wait for systemic answers to arrive. The bar already knows competence. The bench already knows fairness. AI blindness is new language. What it names is old, and its obligations are ours.