Of all the intractable debates we face as a society, whether AI helps or hinders access to justice probably rates second after “is a hot dog a sandwich?” On the one hand, handing people a cheap and accessible magic answer button opens up previously unaffordable opportunities for redress. On the other hand, it makes stuff up. People with legal problems have “done their own research” for years, but AI presents an added psychological risk: most folks knew that their self-directed Google research might be wrong, but AI delivers output with confidence and finality that can lead people into disaster.
At Everlaw Summit in San Francisco, former Legal Services Corporation president Jim Sandman suggested that we might have the wrong framing for the dispute. Instead of placing the bot as the alternative to legal help, consider what it can provide someone in need that the legal system is historically bad at providing.
Sandman, who now teaches at the University of South Carolina’s Joseph F. Rice School of Law, explained how he challenges his students to write the kind of prompt a normal person would type — “as if you had never set foot in a law school” — and run it through two frontier models and two legal industry research tools. One student typed: My apartment is freezing. What can I do?
ChatGPT and Gemini came back with plain-language numbered steps. Philadelphia (until this year, Sandman taught at Penn, and this anecdote clearly dates from those days) requires landlords to provide heat from November through March. So the bot told the user to first tell the landlord. If that doesn’t work, send a demand letter… and here’s a draft that cites the relevant regulations. Which, in practice, is probably good enough to make the landlord assume the tenant has a lawyer. If that fails, the bot explained, call the housing authority and let them fight the battle for you. “And way down the list,” Sandman explained “is sue.”
By contrast, the legal tools got the law right, but were “terrible at communicating.” Because the legal tools began from the premise that the tenant needs to run to court. It was the first suggestion those tools offered! As if someone without heat in December wants to launch a months-long litigation as their first resort.
What if AI’s greatest gift to the access to justice problem is keeping people out of court? Years ago, Sandman spoke at a different conference about the court system’s hostility to pro se litigants. Right there in the name, we’re already forcing them to deal with a dead language. Everything about the process is alienating, and if anyone from the court system tries to help… they’re just going to get fired.
Sandman conceded that frontier models sometimes get the law wrong. Ideally, he would like to see something that marries clarity and common sense with legal-grade accuracy. “We’re not there yet,” he said.
For most people, a lawyer is just never going to be on the menu. Landlords have counsel in 84 percent of eviction cases, while tenants have lawyers in about 4 percent. Only 1 percent of America’s roughly 1.3 million lawyers are paid legal aid attorneys. Vanderbilt Law’s Cat Moon, also on the panel, noted that there are fewer than two attorneys statewide for every 2,000 people eligible for Legal Aid in Tennessee. The most recent appropriation for the Legal Services Corporation was $540 million. Sandman noted that this Halloween, Americans were projected to spend $920 million on costumes for their pets. That would sound disastrous until you remember that the Trump administration had proposed $21 million, for an “orderly closeout.” It’s proposing eliminating the agency again, though something tells me this next Congress will be less amendable to that.
To the extent AI does become involved in bridging the courthouse gap, Sandman reminded the audience that “If you automate a bad process, you end up with an automated bad process.” And most of the country considers the legal system to be a bad process. Verna Williams of Equal Justice Works noted that 79 percent of Americans perceive the justice system as unfair and 62 percent believe that income shapes outcomes. Without a significant reappraisal of the system’s design, AI just throws more “garbage out” on the pile of too many appearances, incomprehensible forms, and hearings for people who can’t get time off or child care. If the system is serious about introducing AI into the process, everything needs to be redesigned around the fact that most cases have at least one party without a lawyer.
Earlier that week, on the other side of the country, I spoke with Clio about the role AI can play in fixing the court system specifically. In the span of about five weeks, the company had hired Casetext cofounder Pablo Arredondo to build a judiciary product line, recruited former Michigan Supreme Court Chief Justice Bridget McCormack to run the business, and bought Learned Hand, an AI-for-judges startup already in use at L.A. County Superior Court. “The legal experience for all often runs through courthouses,” McCormack said. “Sometimes not, but often it does. And if they’re left behind, you’re just not going to deliver on that mission.”
As much as we all appreciate a good mission statement, even Canadian companies need to make money, and courts just don’t seem like an obvious revenue play for a major legal tech provider. The budgets are perennially strained and glacial pace of adoption befits a bureaucracy. But Clio seems to see an opportunity here that never existed before.
You’ve probably guessed that it revolves around AI.
If AI lowers the barrier to entry for people who can’t afford lawyers, McCormack regards that a net positive. But the catch is that courts face more pressure to deliver on their mission once that barrier drops. Sandman talks about redesigning the courts to deal with the reality of unrepresented litigants now, and Clio envisions the courts struggling against even more of that litigation in the near future. “And that’s going to require some of the same technology that has lowered the barrier to entry.”
Clio CEO Jack Newton’s keynote called the rise in filings hitting courts as “the agentic flood.” Arredondo upgraded the forecast to “a monsoon,” because the effort to file “has come down to near zero.” Some trial courts have seen filings double with zero clerks to absorb them. Most litigation happens in state courts, and state court judges aren’t bringing on multiple top-tier law grads every term to stay on top of the docket.
The fix Clio is pitching isn’t robot judges, so much as simple hyperlinks. Arredondo told me, “forget J.A.R.V.I.S. for the moment — spell check is my muse.” His approach to bringing AI into courthouses attempts to model the success of spell check as what he calls “ambient error detection.” When a user misspells a word in a modern word processor, “the red squiggly” appears. That’s what Clio’s trying to model, but with more powerful outcomes — automatically giving the judge in-line assistance to get the legal research they need in front of them quickly and seamlessly.
A theme of my AI coverage has been my skepticism of speed. Tools that warp through the workflow don’t necessarily generate better output, and approaching the technology from the perspective of time forecloses human insight and produces sub-optimal work. I’ve been cautiously optimistic to see that, in practice, lawyers seem to be doing more work in the AI era. And I don’t think that’s just worriedly verifying every cite, but rather a sign that AI is being used like a teammate and enabling the human lawyer to chase down insights they wouldn’t have come up with otherwise.
For the courts, this might allow AI to become a real slowness enabler. Not that it would slow down the whole process, but that it can fulfill the role of a supercharged clerk to allow a judge to really sit with the fully fleshed out argument instead of feeling rushed to fire off an answer the moment the research is complete.
Arredondo’s breakout session included a video of former Massachusetts Appeals Court Chief Justice Mark Green describing the overwhelming burden of emergency motions to stay evictions that arrive at 3:30 in the afternoon, “the day before the moving truck was set to appear on the curb the next morning.” Green said that in “north of 90 percent” of those eviction-stay motions, there’s nothing to be done. But a judge still has to read every motion to find the 10 percent where the courts need to intervene, usually by wading through “an undifferentiated mass of handwritten materials submitted by a self-represented litigant.” With a tool an outside vendor built for the court, four and a half hours of review dropped to 45 minutes, and the judge got citations back to the record so nobody had to take the summary on faith.
That may be a tool for the courts, but the beneficiary is the tenant about to be thrown on the street. When a court can efficiently process the deluge of filings, the judge has time to deliberate and take action.
AI should help self-represented parties achieve justice without the courts, but there are two sides to the equation. If the court system is woefully unable to address the needs of those who need court intervention, then what good are the courts?
At the Everlaw panel, Williams described the access to justice crisis as an “all-hands moment.” With the necessary resources unlikely to arrive, part of the present emergency will require better triage. We need more Legal Aid funding and pro bono effort, but we also need bots that help people get justice without lawyers. And bots to help the courts deliver justice when faced with a mountain of paper and too little time.
No one wants to replace lawyers — well, some people do, though let’s forget about them for a second — but the reality is that there are already people with legal problems who will never have a lawyer and there are already courts rummaging through a growing number of filings from unrepresented parties. “The risk I see is the mistake of comparing an AI tool to a lawyer and concluding that if a lawyer would be better, you shouldn’t use AI,” Sandman said. “The proper comparison is between AI and no help at all.”
Joe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter or Bluesky if you’re interested in law, politics, and a healthy dose of college sports news.
The post For Access To Justice, The Perfect Lawyer Is The Enemy Of The Good Chatbot appeared first on Above the Law.
Of all the intractable debates we face as a society, whether AI helps or hinders access to justice probably rates second after “is a hot dog a sandwich?” On the one hand, handing people a cheap and accessible magic answer button opens up previously unaffordable opportunities for redress. On the other hand, it makes stuff up. People with legal problems have “done their own research” for years, but AI presents an added psychological risk: most folks knew that their self-directed Google research might be wrong, but AI delivers output with confidence and finality that can lead people into disaster.
At Everlaw Summit in San Francisco, former Legal Services Corporation president Jim Sandman suggested that we might have the wrong framing for the dispute. Instead of placing the bot as the alternative to legal help, consider what it can provide someone in need that the legal system is historically bad at providing.
Sandman, who now teaches at the University of South Carolina’s Joseph F. Rice School of Law, explained how he challenges his students to write the kind of prompt a normal person would type — “as if you had never set foot in a law school” — and run it through two frontier models and two legal industry research tools. One student typed: My apartment is freezing. What can I do?
ChatGPT and Gemini came back with plain-language numbered steps. Philadelphia (until this year, Sandman taught at Penn, and this anecdote clearly dates from those days) requires landlords to provide heat from November through March. So the bot told the user to first tell the landlord. If that doesn’t work, send a demand letter… and here’s a draft that cites the relevant regulations. Which, in practice, is probably good enough to make the landlord assume the tenant has a lawyer. If that fails, the bot explained, call the housing authority and let them fight the battle for you. “And way down the list,” Sandman explained “is sue.”
By contrast, the legal tools got the law right, but were “terrible at communicating.” Because the legal tools began from the premise that the tenant needs to run to court. It was the first suggestion those tools offered! As if someone without heat in December wants to launch a months-long litigation as their first resort.
What if AI’s greatest gift to the access to justice problem is keeping people out of court? Years ago, Sandman spoke at a different conference about the court system’s hostility to pro se litigants. Right there in the name, we’re already forcing them to deal with a dead language. Everything about the process is alienating, and if anyone from the court system tries to help… they’re just going to get fired.
Sandman conceded that frontier models sometimes get the law wrong. Ideally, he would like to see something that marries clarity and common sense with legal-grade accuracy. “We’re not there yet,” he said.
For most people, a lawyer is just never going to be on the menu. Landlords have counsel in 84 percent of eviction cases, while tenants have lawyers in about 4 percent. Only 1 percent of America’s roughly 1.3 million lawyers are paid legal aid attorneys. Vanderbilt Law’s Cat Moon, also on the panel, noted that there are fewer than two attorneys statewide for every 2,000 people eligible for Legal Aid in Tennessee. The most recent appropriation for the Legal Services Corporation was $540 million. Sandman noted that this Halloween, Americans were projected to spend $920 million on costumes for their pets. That would sound disastrous until you remember that the Trump administration had proposed $21 million, for an “orderly closeout.” It’s proposing eliminating the agency again, though something tells me this next Congress will be less amendable to that.
To the extent AI does become involved in bridging the courthouse gap, Sandman reminded the audience that “If you automate a bad process, you end up with an automated bad process.” And most of the country considers the legal system to be a bad process. Verna Williams of Equal Justice Works noted that 79 percent of Americans perceive the justice system as unfair and 62 percent believe that income shapes outcomes. Without a significant reappraisal of the system’s design, AI just throws more “garbage out” on the pile of too many appearances, incomprehensible forms, and hearings for people who can’t get time off or child care. If the system is serious about introducing AI into the process, everything needs to be redesigned around the fact that most cases have at least one party without a lawyer.
Earlier that week, on the other side of the country, I spoke with Clio about the role AI can play in fixing the court system specifically. In the span of about five weeks, the company had hired Casetext cofounder Pablo Arredondo to build a judiciary product line, recruited former Michigan Supreme Court Chief Justice Bridget McCormack to run the business, and bought Learned Hand, an AI-for-judges startup already in use at L.A. County Superior Court. “The legal experience for all often runs through courthouses,” McCormack said. “Sometimes not, but often it does. And if they’re left behind, you’re just not going to deliver on that mission.”
As much as we all appreciate a good mission statement, even Canadian companies need to make money, and courts just don’t seem like an obvious revenue play for a major legal tech provider. The budgets are perennially strained and glacial pace of adoption befits a bureaucracy. But Clio seems to see an opportunity here that never existed before.
You’ve probably guessed that it revolves around AI.
If AI lowers the barrier to entry for people who can’t afford lawyers, McCormack regards that a net positive. But the catch is that courts face more pressure to deliver on their mission once that barrier drops. Sandman talks about redesigning the courts to deal with the reality of unrepresented litigants now, and Clio envisions the courts struggling against even more of that litigation in the near future. “And that’s going to require some of the same technology that has lowered the barrier to entry.”
Clio CEO Jack Newton’s keynote called the rise in filings hitting courts as “the agentic flood.” Arredondo upgraded the forecast to “a monsoon,” because the effort to file “has come down to near zero.” Some trial courts have seen filings double with zero clerks to absorb them. Most litigation happens in state courts, and state court judges aren’t bringing on multiple top-tier law grads every term to stay on top of the docket.
The fix Clio is pitching isn’t robot judges, so much as simple hyperlinks. Arredondo told me, “forget J.A.R.V.I.S. for the moment — spell check is my muse.” His approach to bringing AI into courthouses attempts to model the success of spell check as what he calls “ambient error detection.” When a user misspells a word in a modern word processor, “the red squiggly” appears. That’s what Clio’s trying to model, but with more powerful outcomes — automatically giving the judge in-line assistance to get the legal research they need in front of them quickly and seamlessly.
A theme of my AI coverage has been my skepticism of speed. Tools that warp through the workflow don’t necessarily generate better output, and approaching the technology from the perspective of time forecloses human insight and produces sub-optimal work. I’ve been cautiously optimistic to see that, in practice, lawyers seem to be doing more work in the AI era. And I don’t think that’s just worriedly verifying every cite, but rather a sign that AI is being used like a teammate and enabling the human lawyer to chase down insights they wouldn’t have come up with otherwise.
For the courts, this might allow AI to become a real slowness enabler. Not that it would slow down the whole process, but that it can fulfill the role of a supercharged clerk to allow a judge to really sit with the fully fleshed out argument instead of feeling rushed to fire off an answer the moment the research is complete.
Arredondo’s breakout session included a video of former Massachusetts Appeals Court Chief Justice Mark Green describing the overwhelming burden of emergency motions to stay evictions that arrive at 3:30 in the afternoon, “the day before the moving truck was set to appear on the curb the next morning.” Green said that in “north of 90 percent” of those eviction-stay motions, there’s nothing to be done. But a judge still has to read every motion to find the 10 percent where the courts need to intervene, usually by wading through “an undifferentiated mass of handwritten materials submitted by a self-represented litigant.” With a tool an outside vendor built for the court, four and a half hours of review dropped to 45 minutes, and the judge got citations back to the record so nobody had to take the summary on faith.
That may be a tool for the courts, but the beneficiary is the tenant about to be thrown on the street. When a court can efficiently process the deluge of filings, the judge has time to deliberate and take action.
AI should help self-represented parties achieve justice without the courts, but there are two sides to the equation. If the court system is woefully unable to address the needs of those who need court intervention, then what good are the courts?
At the Everlaw panel, Williams described the access to justice crisis as an “all-hands moment.” With the necessary resources unlikely to arrive, part of the present emergency will require better triage. We need more Legal Aid funding and pro bono effort, but we also need bots that help people get justice without lawyers. And bots to help the courts deliver justice when faced with a mountain of paper and too little time.
No one wants to replace lawyers — well, some people do, though let’s forget about them for a second — but the reality is that there are already people with legal problems who will never have a lawyer and there are already courts rummaging through a growing number of filings from unrepresented parties. “The risk I see is the mistake of comparing an AI tool to a lawyer and concluding that if a lawyer would be better, you shouldn’t use AI,” Sandman said. “The proper comparison is between AI and no help at all.”
Joe Patrice is a senior editor at Above the Law and co-host of Thinking Like A Lawyer. Feel free to email any tips, questions, or comments. Follow him on Twitter or Bluesky if you’re interested in law, politics, and a healthy dose of college sports news.
