It has been two weeks for stories about AI in the courtroom. You may have heard about the victim of a road rage incident whose family created a Appetite One of the effects of this as a statement (possibly for the first time it has been done in the United States). But more than that, there is a more fruitful conflict, according to legal experts. Legal filing is increasing more and more AI frauds. And it is starting to provoke the judges. Just look at all three matters, each of which is glimpse a glimpse of it, which we can expect when lawyers embrace AI.
A few weeks ago, a California judge, Michael Wilner, some lawyers took interest in a set of arguments made in the filing. He went to find out more about the articles he cited. But articles were not available. He asked the lawyers’ firm for further details, and he gave a new brief answer that exists Even more errors Already, Vilner ordered the lawyers to test the oath -taking testimony, explaining the mistakes, in which he found that one of them, from Elite Firm Alice George, used to help write a document to AI models related to Google Gemini, which created false information. As in detail in A Filing On May 6, the judge fined the firm, 000 31,000.
Last week, another California -based judge arrested another fraud in the court filing, this time was presented by the AI ​​company’s anthropic in a case that recorded labels brought against copyright cases. An Anthropic lawyer asked the company’s AI model cloud to cite a legal article, but the cloud contains the wrong title and the author. Anthropic’s lawyer admitted that no one who reviewed the document had made a mistake.
Finally, and maybe most of all, a matter is coming up in Israel. After police arrested a person on money laundering charges, an Israeli prosecutor submitted a petition seeking permission from a judge to keep the individual’s phone as evidence. But he cited rules that are not available, the defendant’s lawyer accused him of adding AI fraud in his request. Prosecutor according to Israeli News OutletsAdmitted that this was the case, scolding the judge.
Together, these matters point to a serious problem. Courts rely on documents that are correct and support these references. The two features that AI models, despite being adopted by lawyers seeking time, often fail miserably in supply.
Those mistakes are being caught (for now), but it is not to be imagined that at any point, the judge’s decision will be affected by something that is completely made by AI, and no one will be able to catch it.
I talked to Maura Grossman, who teaches Water Low University as well as in Osgod Hall Law School’s School of Computer Science, and has been the initial critic of problems that produce AI courts. They Is written About this problem in 2023, when the first incidents of deception began to appear. He said he believes that the current rules of the courts demand lawyers to attract the issues that will stop the issue with bad advertising of what they present to the courts. It’s not out of it.
She says “it seems” it doesn’t look like “her pace has slowed.” “If anything, they are faster.” She says these are not cases with unclear local firms. These are big -time lawyers who are making important, shameful mistakes with AI. He fears that such errors are also crushing more in documents that are not written by lawyers themselves, such as expert reports (in December, a professor at Stanford and AI expert Enrolled Including AI-generated errors in its testimony).
I told Grassman that I was surprised. Lawyers, who are mostly, have a diction obsession. They choose their health words. Why are many people stuck while making these mistakes?
“Lawyers fall into two camps,” she says. “The first is afraid of death and does not want to use it at all.” But then the early adoptions are. These are strict lawyers to provide a brief help on time or other lawyers without cadre. They are anxious for technology that can help them write documents under a tough deadline. And their checks on AI’s work are not always complete.
The fact is that high -power lawyers, who have to test the language, are being caught by making mistakes introduced by AI, how many of us treat this technology. We have been repeatedly told that AI makes mistakes, but language models also feel like magic. We have a complex question and receive what seems like a thoughtful, intelligent answer. Over time, the AI ​​model develops a hidden progress of the authority. We trust them.
Grassman says, “We assume that since these large models of language are so fluent, it also means that they are right.” “We slip into all kinds of reliable moods because it looks authentic.” Lawyers are used to examine junior lawyers and interns, but for some reason, Grassman says, do not apply these doubts to AI.
Ever since the Chattgupat launched almost three years ago, we have known about this problem, but since then the recommended solution has not been prepared: Do not rely on what you read, and consider what the AI ​​model tells you. Since AI models we emphasize many different tools that we use, I quickly feel that this is the unsatisfactory counter of one of the most basic flaws of AI.
Massive language models are suffering from ways to work. Nevertheless, companies are selling Generative AI tools made for lawyers that claim they are reliable. “Feel confident that your research is accurate and complete.” Precision on WestlaAnd for the website COCCOUNT Its AI promises that “authentic material is supported.” He did not stop his client, Ellis George, from imposing a fine of 000 31,000.
Fast, I sympathize with people who trust AI more than them. We, however, are living at a time when people are telling people who build this technology that AI is so powerful and should be treated as such. The nuclear weapons. Models have learned from almost every word that humanity has ever written and are infiltrating our online life. If people should not trust everything in the AI ​​model, they probably deserve to be reminded that they should be reminded a little more often by their companies.
This story originally appeared AlgorithmOur weekly newsletter on AI. To get such stories in your inbox first, Sign up here.