There is a growing belief that generative AI is going to produce so many new lawsuits that our courts will simply be unable to keep up. As the technology makes it easier for people to prepare and file pleadings, the thinking goes, we will eventually face a deluge of litigation that our existing judicial system cannot possibly handle. And increasingly, I hear suggestions that the solution will be to place smaller claims into a separate category and allow AI systems to decide them, leaving human judges to handle more complicated matters.
I understand the reasoning. Courts are already struggling with limited resources, and generative AI is changing how legal work gets done. Someone who might have struggled to prepare a petition a few years ago can now ask an AI tool to draft one, complete with legal arguments and supposed citations. But this is not just about self-represented litigants. Lawyers can use these same tools to take on more clients, prepare more petitions and pleadings, and move cases forward with considerably less time spent on routine work. Whether that increased capacity produces the anticipated flood of litigation that buries us remains to be seen, but it is certainly a possibility worth considering.
What troubles me is how quickly the conversation has moved from identifying that possibility to assuming we will need artificial judges to handle it. We seem to take for granted that AI will dramatically increase the number of cases coming into our courts without considering how that same technology might increase the capacity of the judges responsible for deciding them. We have only begun exploring what a properly trained, AI-competent judiciary might be capable of accomplishing.
I have spent much of the last several years traveling around the country teaching judges about generative AI and encouraging them to explore how these tools might be incorporated into judicial workflows. Some judges are already finding meaningful ways to use them, while others are still learning the basics. Most courts, though, have not yet developed the training, infrastructure, or workflows necessary to take full advantage of what this technology can do. Yet we are already discussing the need to build entirely new systems to decide cases without judges.
Before we get too far down that road, consider a dispute between a homeowner and a contractor over $4,500 in allegedly unfinished work. The contractor believes he is owed the money. The homeowner believes the work was defective and refuses to pay. Both have photographs, text messages, invoices, and their own versions of what happened.
To a court system processing thousands of cases, this might look like an insignificant dispute, precisely the kind that some envision assigning to AI. But for the contractor, the money may be needed to pay his employees. For the homeowner, it may represent what she needs to make her home habitable. Neither necessarily considers the dispute small.
The homeowner also did not choose to enter the justice system. Someone sued her, and the government has established a process that requires her to respond. Under the proposed approach, she might discover that because the amount in controversy falls below a particular threshold, her case will be decided by an AI system rather than a human judge.
Of course, we already treat smaller claims differently. We have developed simplified procedures and specialized courts to resolve less complicated disputes more efficiently. Depending on the jurisdiction and type of proceeding, we may limit discovery, restrict access to juries, or assign matters to magistrates or other judicial officers. But these procedures generally preserve a human decision-maker exercising authority within an established system. Simplifying the process is different from replacing the person responsible for the decision with AI.
I am not opposed to AI systems that help people resolve disputes. Those systems already exist. The American Arbitration Association has introduced its AI Arbitrator, which assists a human arbitrator in resolving disputes, and AI-powered mediation services are becoming more widely available. I welcome these developments, particularly when the parties knowingly consent to using them. But there is a meaningful difference between voluntarily choosing an alternative method of resolving a dispute and being sued in court, only to discover that the government has decided your case is too small to warrant a human judge.
For me, that distinction goes directly to the social contract underlying our justice system. We have spent generations developing a system in which human beings are entrusted with the authority to resolve disputes and exercise the power of the state. We have established rules governing how judges obtain their positions, how they must conduct themselves, and how their decisions can be challenged. The system is certainly imperfect, and human judges bring their own experiences, perspectives, and limitations to the bench. But the relationship between the public and the judiciary involves more than simply producing an opinion.
That is why the suggestion that AI should begin deciding smaller cases deserves more than a discussion about whether the technology is capable of doing so. Using AI to assist a judge with judicial work is one thing. Having the government select an AI system to exercise judicial authority is quite another. Someone still has to choose the model entrusted to decide the homeowner’s case, and that choice involves much more than selecting the most capable technology.
I have written before about whose singularity we are building toward and whose values and priorities will shape the increasingly intelligent systems we rely upon. Every AI model has its own bend, influenced by the information on which it was trained and the choices made by the people developing it. We are only beginning to consider what accountability would look like when the decision-maker is a machine. And even if AI eventually becomes capable of deciding cases as well as, or perhaps better than, human judges, that does not necessarily mean we should restructure our institutions around it.
Rather than replacing human judges, maybe we should spend our time and resources exploring how we can use AI to make the judges we have substantially more capable. In our homeowner’s case, AI could help prepare the judge and thus shorten the time from suit to decision. The time saved could also enable judges to better address the rest of their dockets.
In fact, that possibility is shaping how we are approaching AI at my own court. We recently purchased two AI servers and began developing applications around the work our judges and staff perform every day. We are starting with some of the less glamorous tasks surrounding the administration of justice, including filing compliance, document processing, and organizing records. We want to see what these tools can help us accomplish before exploring more ambitious applications.
I recognize that even a judiciary equipped with the best AI tools may not be able to keep pace with every increase in litigation. We do not yet know how much additional capacity these systems will provide, and there will always be aspects of judging that require time and careful deliberation. But it seems premature to conclude that we need an entirely new system of AI adjudication when we have barely begun investing in the capabilities of the judges we already have.
Courts have modernized many times without fundamentally changing the relationship between the public and the judiciary. We moved from handwritten records to typewriters and laptops, from physical law libraries to electronic research, and from paper files to digital records. More recently, we expanded the use of remote appearances during the pandemic. Each advancement changed how judges performed their work, sometimes dramatically, while leaving the responsibility for judicial decisions with the people entrusted to make them.
Generative AI may change judicial work far more profoundly, but realizing that potential will require more than purchasing subscriptions to commercial products. We need meaningful judicial education, practical training, secure infrastructure, and tools designed specifically for judicial work. That will take time and investment, but so would building an entirely new system of AI adjudication.
Perhaps someday society will decide that certain disputes should be adjudicated by AI without human judges. I am not prepared to rule out that possibility. But I also do not believe that outcome is inevitable simply because the technology becomes capable of performing the task. There is a difference between recognizing where technology may be heading and deciding where we want our institutions to go.
If generative AI is going to make it easier for litigants and their lawyers to bring disputes to court, perhaps we should begin by using that same technology to make our judges better equipped to receive them.
And who knows, we may be able to modernize the social contract without having to rewrite it.
