On Zoom, Gerald Abila turns his computer screen around, unblurs the background, and asks, “Do you see? There’s Winnie.” Abila sits in Kampala, Uganda, where he runs an NGO, BarefootLaw, trying to reimagine how to deliver legal access to remote populations across his country. Behind him is a desk mounted with a digital tablet, a corded cradle telephone from the 1930s, and a book, The Prophet by Kahlil Gibran, which Abila says is there to provide the “wisdom that can act as a guide for our collective humanity.” Pushed under the desk is a wooden chair with a white sign on it that reads “WINNIE.” Winnie’s chair is empty because Winnie is an AI lawyer.
Winnie was first developed to help Abila’s team at BarefootLaw more effectively deliver legal help in a setting where resources are already constrained. But Abila’s idea that an AI lawyer might help mitigate the access-to-justice gap does not exist in a vacuum. Others have proposed similar scenarios, even if Abila was one of the first to act on it. Around the world, roughly 1.5 billion people confront issues where legal help might be the answer—but those issues often go unresolved. In the United States, the justice gap stands at roughly 50 million people. But both scholars and practitioners in the access space have concerns about how AI is endlessly proposed as a solution, without thinking through its suitability or safety.
In this article, we think through how the use of AI in the access-to-justice space requires larger systemic thinking around governance, policy, and risk. We speak to Abila about how his initial excitement around AI has worn down; to Themba Mahleka and Nate Edwards, applied researchers in the Pathfinders for Peaceful, Just and Inclusive Societies program at the NYU Center on International Cooperation (CIC), about why AI governance has to be a multistate, multilateral solution; and, finally, to Drew Simshaw, professor of law at the William S. Boyd School of Law at the University of Nevada, Las Vegas, about the need for interoperability and a proposal for a national sandbox for innovation.
An AI lawyer
In 2014, before anyone had even heard of ChatGPT, Winnie was just a dream. As a lawyer trying to serve a distributed population without consistent internet access, Abila went on tour, speaking to tech companies in Johannesburg, The Hague, and San Francisco, asking, “Could we build Siri for lawyers?” The answer, at the time, was no. Siri was built on too-rudimentary technology.
Abila founded BarefootLaw around 2013 with just a Facebook page and a texting platform. Born in war-torn Uganda, he realized early on that “the law is not perfect, but it’s still the best tool to build conflict-free societies.” In law school, he became determined to make the law accessible to as many people as possible. His ultimate goal, he says, is to build a world where everyone “shall have access to the law to use it to empower themselves and their communities to thrive.” Originally, people would direct message or text Abila, and he would provide information. But he imagined he could go broader. Today, BarefootLaw maintains multiple solar-powered shipping containers outfitted with internet and screens, allowing rural Ugandans to access legal help via remote pathways. On any given day, Abila says, his team receives over 100 legal inquiries through the boxes and other various means. They’ve helped over 1 million people since they started.
An AI lawyer felt perfectly in line with this mission—it could help scale and deliver even more services to more people. In 2018, when Abila was about to give up, he received an email to meet a group of technologists in Dagstuhl, Germany, where they had set up a computer science research institute amidst castle ruins. From there, Winnie finally began to take shape. By 2019, Abila had a working prototype—a much simpler version of what people now understand AI to be capable of, designed to reduce response time for his team but never operating without human interaction and oversight.
AI is the future but it should be regulated like we regulate nuclear weapons.
Gerald Abila, founder and CEO, BarefootLaw
But when ChatGPT premiered in 2022, he took a step back: he realized the power and potential harms of AI and resolved to keep using Winnie internally but wanted to focus on promoting ethical AI development.
Today, when someone reaches out to BarefootLaw for help, they don’t interact with Winnie, but she is there. Winnie helps the organization with triaging and routing, research, and other back-office tasks, but the client ultimately interfaces with a human. What started as a dream of an external-facing tool to spread legal information has been restricted to an internal tool with guardrails. And while many are pushing AI throughout the legal ecosystem, Abila has cooled on it: what he is focused on now is “doctrinal innovation.” AI holds enormous opportunity but also enormous risk, he says, and he wants to define both before innovating further. “It’s the future,” Abila says, “but it should be regulated like we regulate nuclear weapons.”
Governing the gap
The regulation Abila wants has to be written somewhere, by someone. In 2023 a UNESCO report noted that 44 percent of judicial actors in 96 countries were already using AI informally in their work, without oversight or training. At Pathfinders for Peaceful, Just and Inclusive Societies, an initiative of the NYU CIC, Mahleka and Edwards work on the question of who writes the rules and what those rules should be. They advise countries and organizations tackling the questions at the heart of Abila’s project: How can AI be used in justice systems to increase access to justice while also being cognizant of harms? What is the role of justice systems in providing accountable and enforceable AI governance across diverse sociopolitical contexts?
While AI governance in a corporate headquarters centers around what internal guardrails the company might place about who can use it, what data goes in, and how the output is verified, AI governance for this international think tank is about global and national governance systems: How can states design policies for AI in their justice systems that are additive and outcome oriented rather than about using a technology for technology’s sake without thinking through how it might support ordinary people in accessing justice? And, on the other hand, what role should justice actors, working on the front line of rights protections in practice, play in designing national AI regulations?
Around the world, the justice system is already strained to capacity, and adding in new roles and responsibilities could be an additional burden, Mahleka and Edwards warn. To that end, Pathfinders is supporting states that want to share work on strategies and solutions collectively rather than having parallel solo workstreams. With this, they bring Pathfinders’ “people-centered approach” to bear on the problem. This means “designing services that are holistic in meeting people where they are, and then identifying where AI can be additive to that versus just the primary goal being to use AI,” says Edwards. “We really don’t feel that there needs to be unfettered application of AI just because it exists. It needs to actually be targeted and purpose-driven to address these different gaps.”
What’s evident from some of our research is that AI is not creating new categories of harms; it’s scaling existing ones.
Nate Edwards, applied researcher, Pathfinders for Peaceful, Just and Inclusive Societies program, NYU Center on International Cooperation (CIC)
Mahleka and Edwards highlight AI-driven services in countries like Brazil and India designed to tackle the often years-long backlog of cases in courts—timeliness in hearing cases is one component of access to justice. They point to how the use of unregulated, untested technology in any government proceedings can cause harm, as in the Robodebt scandal in Australia, where an incorrect algorithm demanded individuals pay back debts they never owed; in this vein, Pathfinders is also interested in situations where procedural rights to remedy and accountability are carefully outlined. Ultimately, the responsibility of providing remedy and accountability for any AI-related harms will fall to domestic justice systems.
“What’s evident from some of our research is that AI is not creating new categories of harms; it’s scaling existing ones—it’s diluting responsibility, blurring liability, and weakening transparency on what’s actually happening if you can’t understand how agents are making decisions,” says Edwards. “What we’re finding is that legal systems are often relying on existing laws—around consumer protections, anti-discrimination, data privacy, product liability, children’s online safety, copyright infringement, these types of things to regulate AI-related harms. But with agentic AI, especially where multiple agents are interacting, these existing laws are going to become harder to apply.” Policy needs to evolve, they argue.
Mahleka and Edwards are cognizant of the two-tiered system of unequal justice. There is a future scenario where the wealthy still have access to lawyers and the more vulnerable have a chatbot. Governance can help. “We need to be able to understand whether AI is going to become a bridge to justice or a substitute for justice,” Mahleka says. “And that’s not about the technology itself; it’s about the policy choices that we have, the funding decisions we make, and the institutional values that surround it.”
An opportunity and a stress test
“We think we’re the custodians of the law,” Abila says, “and yet, with this technology, everyone is potentially a lawyer.”
In addition to needing to ready judicial actors and governance systems with proper policy and training around AI, there is a world where AI also levels the playing field for pro se litigants. In the United States, a large majority of civil cases filed are debt-collection cases, in which large corporations with expensive lawyers are filing suit against resource-poor individuals who end up representing themselves, if they’re able to show up to court at all. (Stanford Law School has partnered with the LA Superior Court system to use AI to review debt-collection claims, in case such filings were actually done without evidence or in error.) But for those without a lawyer, could AI help?
Some believe that too will cause issues. “Courts are woefully unprepared for a future where anyone with a chatbot can become a high-volume filer, or where ordinary people might rely on chatbots for desperately needed legal advice,” Keith Porcaro, assistant clinical professor of law at Duke Law School, writes in Wired. Moreover, one slightly muddled prompt could grossly lead someone down the wrong path. Porcaro partnered with the Legal Aid of North Carolina in his algorithmic auditing class to help improve their legal chatbot. When the Nevada Administrative Office of the Courts similarly launched a chatbot for pro se litigants specific to family law, Porcaro’s class audited the technology to understand how it was helpful and when it was harmful. “The audit’s recommendations included improving the chatbot’s ability to detect high-risk situations and deliver more specific resources to the user, such as suggestions to call 911 or child protective services.”
A sandbox for innovation
Simshaw has studied how incorporating AI to solve justice problems requires bigger systemic changes. “I think the thing to not lose sight of is that access to justice is in a lot of ways an ecosystem problem,” says Simshaw. “A groundbreaking individual technology operating inside a fragmented ecosystem can still produce disappointing results.”
Simshaw looks at experimentation in states like Utah and Washington as potential avenues for improvement to restrictive innovation around legal technology. He proposes a national regulatory sandbox, akin to the state sandboxes that allowed experimentation with innovation around legal advice and law firm ownership and investment, that would allow legal organizations to propose potential new AI-driven tools to attack long-standing problems in a highly regulated atmosphere. “Our legal services market is increasingly technological and capable of operating across borders, while professional regulation remains strictly state specific,” Simshaw says. A national sandbox that states could opt into might centralize resources while allowing legal tech companies to operate in more markets, and more states would mean more data about what works and what doesn’t. At the end of the day, participating states could still decide on their own rules.
“If we want regulation to be responsive to actual consumer harm rather than hypothetical fears on one side or extreme technological optimism on the other side, broader experimentation can give us better data about outcomes and risks,” he says. “A well-designed sandbox is not a zone without regulation. Its purpose is active supervision and empirical learning about what regulation is actually necessary.”
The objective really should be to determine what combination of the technology and human assistance is appropriate for the particular person, the particular problem, the particular process involved.
Drew Simshaw, professor of law, William S. Boyd School of Law, University of Nevada, Las Vegas
Moreover, as experimentation takes hold, it is critical that all these systems be able to speak to each other or be interoperable. If an AI tool helps someone prepare a filing but the court can’t accept the form that the filing is delivered in, what then? Simshaw points to Brazil as a case study that has thought through holistic interoperability. With a backlog of cases and enormous pressure, Brazil’s judiciary noticed that some courts were starting to innovate while others weren’t, creating coordination issues. To solve for this, an academic team from Columbia University partnered with a Brazilian nonprofit research institute to develop a framework for the National Council of Justice that would “inventory and assess the AI tools already being developed so that they could be integrated and standardized, design a collaborative governance structure, and then align that structure with international best practices,” says Simshaw. The framework would create a quasi “national factory for AI models.”
Many countries are trying to figure out the rules around “who should provide legal services and how lawyers and technologists might work together. How should this be regulated?” says Simshaw. It’s worth observing how different jurisdictions identify and solve for risks and opportunity. But Simshaw agrees that technology is not a Band-Aid. “The objective really should be to determine what combination of the technology and human assistance is appropriate for the particular person, the particular problem, the particular process involved,” he says. He calls this “calibration for access.”
Problem-solvers first
Taken together, these proposals sketch what governing AI for access to justice would actually require—procedural rights to remedy when automated systems fail, the multilateral coordination Pathfinders is building, supervised experimentation of the kind Simshaw’s sandbox imagines, and the infrastructure of interoperability—so that tools, courts, and jurisdictions can speak to one another. But nothing in the legal profession and the regulations that govern it is happening as fast as the technology advances.
Abila says that legal AI should force a reckoning in the legal profession. If every phone can provide legal advice, the lawyer must be more—a problem-solver, he says. He tells his team, “We are first and foremost human beings. We are secondly problem-solvers, who use the law as the primary tool to problem-solve.” Not every problem that comes to BarefootLaw requires the law, but “we are problem-solvers who know the law and we can use it to resolve problems more effectively than others.” If technology can help, Abila says, then great. But for them, they’ve taken a step back from Winnie and AI in general. “We have descaled it until we understand what it means from a doctrinal perspective,” he says. “Until we can develop it in a way that amplifies our work but doesn’t take away from it.”
For now, Winnie’s chair is under her desk in Uganda, the simple sign still on it. Whether the empty chair is a placeholder or a promise remains to be determined, in no small part by the humans writing the rules.
Image Credit: Shutterstock // Tamas Gabor
