Innovative AI Tool Aims to Bridge Legal Access Gap
In response to a growing crisis in legal accessibility, Yale Law School professors Scott Shapiro LAW ’90 and Ruzica Piskac are developing a cutting-edge AI model designed to provide reliable legal assistance. Housed within Yale Law School’s AI Lab, the project seeks to empower individuals who lack the resources to hire attorneys.
According to the Legal Services Corporation, low-income Americans receive inadequate or no legal help for 92 percent of their civil legal issues. Shapiro and Piskac aim to change that using artificial intelligence to democratize legal support.
Combining AI with Logic for Accuracy
The professors are acutely aware of the limitations inherent in large-language models (LLMs), which rely on predictive algorithms to generate responses. These models can be prone to error if not carefully guided. To mitigate this, the AI Lab has developed a hybrid system that integrates traditional LLMs with logic-based tools known as theorem provers.
“What we’re hoping to do is create AI tools that are not only accurate and explain themselves but can be used to bridge some of this chasm between people who need legal services and reliable providers of those services,” Shapiro told reporters.
The system is designed to eliminate incorrect answers by relying on formal logic rather than probabilistic predictions. Piskac emphasized that the AI does not generate answers through an LLM alone. “You will never be able to get a wrong answer for a simple reason, because the answer is not produced by an LLM,” she noted.
Understanding How the Model Works
The AI model begins by allowing users to upload legal documents, such as laws, regulations, or corporate policies. The LLM then translates the legal text into a formal representation known as first-order logic, which employs variables and Boolean operations to structure information logically.
While LLMs may struggle with complex reasoning, they excel in translation—an asset when converting natural legal language into logic. To ensure accuracy, the system uses multiple validation methods that automatically test the translated rules.
In one example, the professors used the Foreign Intelligence Surveillance Act (FISA) as a case study. Manually converting the document into logical form took the professors over four hours. In stark contrast, the AI completed the task in just 30 seconds. “It’s kind of upsetting that it took so long for us to do it and it’s so quick,” Shapiro remarked.
Answering Legal Questions with Precision
After the legal text is formatted, users can ask specific legal questions. The model translates the query into the same logical structure and evaluates it using a theorem prover—specifically, Microsoft Research’s Z3 Theorem Prover.
The theorem prover acts as a logic engine, verifying whether the question aligns with the established rules. If the query complies with the regulation, the AI confirms it. If not, it identifies the exact rule or clause that the inquiry violates.
“Even through hundreds and hundreds of pages of text, it will isolate which two rules are indeed contradictory,” Piskac explained, highlighting how the tool saves significant time and improves clarity.
Transparency and Explainability
A major concern with AI technologies is the “black box” nature of neural networks. Policymakers and legal professionals alike worry about the lack of transparency in AI decision-making. Shapiro and Piskac’s model addresses this by offering built-in explainability through logic-based reasoning.
“The black box,” Piskac said, referring to opaque AI systems, “we are gladly opening, because there is nothing secret.” The use of theorem provers allows users to trace how each conclusion was derived, ensuring accountability and trustworthiness.
Expanding Legal Support for All
Ultimately, the professors envision a future where AI can provide meaningful legal support to underserved communities. Shapiro describes the current landscape as a crisis, citing the urgent need for accessible legal tools.
“Helping people access public housing and other types of benefits, or to protect themselves against eviction,” Shapiro said, is the kind of impact they hope their AI model will achieve.
Looking ahead, the pair plans to co-teach a course in the spring titled “Law and Large Language Models,” which will explore the intersection of legal processes and artificial intelligence.
Their work stands as a promising solution to a pressing social issue, blending technological innovation with a commitment to justice and equality.
This article is inspired by content from Original Source. It has been rephrased for originality. Images are credited to the original source.