In a First, Two AI Systems Negotiate Contract By Themselves

In a first, an AI system recently autonomously negotiated a contract with another AI system completely independent of human intervention. The agreement was completed in minutes, and human involvement was only necessary for adding a signature.

Developed by legal technology company Luminance, the systems negotiated the specifics of an actual non-disclosure agreement (NDA) between the firm and proSapient, a client of Luminance.

In a demonstration of the software, lawyer photos appeared on two screens while Lumina’s AI analyzed and suggested edits to the agreement. NDAs, crucial for their strict privacy rules, often need a lot of legal checking. Luminance’s software helps by first highlighting problem areas in red. The AI then changed these parts to better terms, recording the edits during the negotiation and keeping in mind the company's usual contract preferences.

“By putting the day-to-day negotiations in the hands of an AI that is legally trained and understands your business, we’re freeing lawyers up to focus their creativity where it counts,” Luminance’s chief of staff Jaeger Glucina said in a news release. The company employs its own Large Language Model (LLM), trained on a database of over 150 million legal documents.

The new system is part of a growing surge of AI that could transform the legal profession.

Related:‘ChatGPT’ Lawyer Faces Sanctions for Fake Cases

The negotiating software has yet to be tested in real life, but AI is improving at law tests. A recent study discovered that  GPT-4 surpasses the performance of many prospective lawyers on the legal ethics examination mandated by almost all states for law practice. On a simulated Multistate Professional Responsibility Exam (MPRE), GPT-4 achieved a 74% accuracy rate in answering questions, outperforming the estimated 68% national average for human examinees.

An earlier investigation showed that the previous version of GPT-4 managed to get just enough marks to pass law school finals but didn't score highly. A newer study, however, found that GPT-4 can successfully pass the bar exam. Also, recent research discovered that GPT-4 helps law students write faster, but it doesn't improve their work.

Hallucinating in the courtroom

Chatbots might be getting good at exams, but recent attempts to use AI in real cases have not ended well. According to a recently filed brief requesting a retrial for Prakazrel “Pras” Michel, a member of the Fugees hip-hop group, his defense attorney improperly used an experimental generative AI program to compose the closing argument in Michel’s criminal trial.

Related:Meet Harvey: AI Chatbot and Legal Assistant

A new legal team representing Michel stated that the AI-generated closing argument crafted by Michel's former lawyer was an abject failure. According to the brief, "Kenner's closing argument presented baseless claims, misunderstood the necessary legal elements, confused the schemes, and overlooked key flaws in the government's case."

The Fugees case was not the first time AI has gone wrong in a courtroom. Earlier this year, a lawyer in Colorado thought he was using real court cases to help his client's case but found out that ChatGPT made up many of them. During a hearing, the judge could not locate the cases referenced in the motion and, as a result, rejected it because of the inaccurate citations. The judge also warned that he might file a complaint against the attorney. The same happened to a New York attorney, who was fined $5,000.

At the moment, AI's most significant impact is on the grunt work that lawyers do. AI is enhancing legal research by sifting through vast legal databases to deliver pertinent insights, according to AI legal expert Ceschino Brooks de Vita, head of evangelism at Sirion. The technology can parse extensive legal information, distilling crucial insights and precedents that bolster legal arguments, saving time and elevating the precision and quality of legal research.

AI doubters

AI may be helpful for routine legal matters, but one major concern is its lack of transparency, de Vita noted.

“As AI software becomes more advanced and autonomous, it becomes increasingly hard to decipher its inner workings,” he said. “This fuels skepticism among some legal professionals, their clients, and the public.”

Another drawback is AI's limited ability to understand context. While AI excels at processing data, it may need help fully grasp complex legal concepts. With these limitations, some observers say it is not likely to replace human lawyers anytime soon.

“AI lacks the ability to solve complex challenges, think critically, and even possess the empathy that’s so critical to client relationships,” trial lawyer C.L. Mike Schmidt said in an interview. “While AI can automate tasks and analyze massive amounts of data, the practice of law remains highly complex and intricate. AI is seen more as a tool that complements the skills of lawyers, ultimately saving them time and money, and making them more effective.”