The Future of AI in the Justice System: A Principled and Practical Approach

Helen Davies KC: Tonight’s lecture is The Future of AI in the Justice System, a topic which sits at the intersection of law, ethics, technology and one which will shape the future of how justice is delivered. To provide us with his insight into this important topic, we’re honoured to have with us the Right Honourable Sir Robert Buckland, KCMP, Master of the Bench. Throughout his career, Sir Robert has been a forward-thinking advocate for justice reform, committed to making the legal system more accessible, fair and effective. As we all look ahead to how AI may reshape legal processes, access to justice and even judicial decision-making, there are few people better placed to guide us through both the possibilities and the challenges.

Sir Robert Buckland KBE KC: On an almost hourly basis, we seem to be bombarded with more news about the latest developments in machine learning, quantum computing and agentic artificial intelligence. It is very tempting to let things just wash over our heads and plod along as we’ve always done, hoping for the best. I don’t think that such an approach will do.

As lawyers, judges and administrators of justice, we are in danger of taking a passive role, assuming that we will only be asked to address AI issues reactively as and when the administration of justice processes themselves deploy the emerging technology. The reality is that the world around us is already full of AI. An increasing amount of material that we consider as lawyers is AI-generated in whole or in part. Some of it will be misinformation, disinformation or deep fakes. How well-equipped are we really to deal with these developments? And do we lawyers know what we want, what we need and how we would like to work with AI?

In my first Harvard paper, AI Judges and Judgments: Setting the Scene, I decided to focus not on the technology itself, but on the very essence of justice – namely the question of judgment itself. Precisely what is it that lies at the heart of justice and judgment? And what is it about that human element that makes us have trust and confidence in our system? AI offers the potential to enhance efficiency, reduce backlogs and even correct certain human biases. But these technologies do not exist in a vacuum. They are created, trained and deployed by people. As such, they inherit our values, our assumptions and our failings.

We’re seeing technology that can process vast amounts of data, identify patterns and even predict outcomes with remarkable speed. In our system that could mean faster case resolution, reduce backlogs and improve access to justice. AI has the potential to enhance fairness, at least in theory, by removing some of those inconsistencies and biases that human judges might bring to the bench. But we must tread carefully. Judgment in its truest sense is not just about applying rules to facts. It’s about understanding context, exercising moral reasoning and sometimes showing empathy.

Bias is another critical concern. The systems are only as good as the data they’re trained on. I identified four sources of bias in AI: (1) The algorithmic bias from flawed or prejudiced data; (2) A coding bias, where laws are misinterpreted or oversimplified; (2) A misuse where systems are deployed for political or commercial ends; and (4) A human AI interaction bias where judges either over-rely on, or reject, AI input without proper scrutiny.

Transparency is therefore vital. Many of these systems are proprietary, meaning that their inner workings are hidden from us. This undermines due process. We need explainable AI systems that can justify their decisions in terms that we humans can understand. Then there’s the threat of deep fakes. These technologies can already fabricate convincing audio and video evidence, casting doubt on the authenticity of real evidence. I concluded that first paper by stating that the starting principle should be that AI should assist not replace human judges.

In my second Harvard paper, which is the title of this lecture, I sought to come up with some solutions that will be a framework for the use of AI in our justice systems. A growing and real question now is: Will people still want to use conventional court litigation systems if they can access private dispute resolution processes that are cheap and fast? Does increasing familiarity with AI mean that more and more people will readily consent to automated decision-making in justice?

I think that the answer is a resounding ‘yes’, but that the consequences for the existing system and our rule of law do not have to be a zero-sum game. Instead, state systems of justice can at their heart enshrine principles of fairness, human rights and independence of decision-making that will be the gold standard of a justice system that has true integrity. But here in England and Wales, reaching that gold standard is proving difficult.

The justice system is under strain, and traditional methods are no longer sufficient. I’ve advocated the use of agentic AI, with the automation of routine administrative tasks and assistance in evidence reviews that can ease the burdens for court clerks and judges. I think that agentic AI can help to create a culture of compliance with court orders, but efficiency must never come at the expense of fairness.

We know that the digital age has transformed the nature of evidential materials with investigations now involving vast amounts of digital data. AI can help, but its deployment must be carefully governed. And we’ve got to distinguish between mere assistive technologies, which support human decision-making, and automated adjudication, which can replace it. Now, the former is already here. The latter demands rigorous ethical scrutiny. But that does not mean that we should baulk at its introduction to the system.

In England and Wales, the Ministry of Justice published its action plan on AI just over two months ago, and a chief AI officer and justice AI unit have now been created. And already we’re seeing initiatives developing. But it was in the underlying principles of the action plan that I gained most encouragement.

Firstly, that AI and justice must work within the law. It must protect individual rights and it must maintain public trust. Secondly, they say that AI should support not substitute human judgment. Because the independence of judges, prosecutors and oversight bodies will and must be preserved. Thirdly, they want to design AI tools around the needs of users: victims, offenders, staff, judges and citizens. And finally, no duplication. Where there is a common solution that can be used across the system, avoid each arm of government building its own technology.

So, what could be guiding principles for the use of AI systems in the administration of justice? I set out six rules in my paper.

Number one I call ‘algorithmic humility’. AI must recognise its limitations, and it must defer to human judgment in complex or sensitive cases. This principle stipulates that any AI system deployed in a judicial context must be programmed with an acute awareness of its own limitations. This ensures that AI will complement rather than compromise the integrity of judicial processes.

The second rule is the principle of opt-in consent or informed choice. In the initial stages of AI integration, participation in automated judicial processes should be on a voluntary basis. Defendants should be given a clear and informed choice between traditional human-led proceedings and AI-assisted adjudication.

The third rule in the framework is the principle of contextual sensitivity. AI systems must be capable of recognising and flagging cases where there are broader societal or systemic issues. This ability to identify potentially systemic issues is crucial in ensuring that AI systems do not inadvertently perpetuate or exacerbate existing inequalities or flaws in our system.

The fourth rule is one of continuous human oversight. Whilst AI systems may be entrusted with certain decision-making processes, there must always be a clear chain of human responsibility and the possibility of human intervention.

The fifth principle is one of ethical transparency. Any AI system that is deployed in justice must be open to scrutiny, with its decision-making processes explicable in clear non-technical language.

And then the final rule in that framework is the principle of adaptive learning. Whilst AI systems must operate within strictly defined parameters, they should also have the capacity to learn and improve over time, based on feedback from human oversight.

So, those six rules give us the first robust framework for assessing the suitability of AI integration into our system. So, how then to actually deploy it? I think that we should be developing a tiered framework approach.

The first tier is one of human-only adjudication that would be reserved for complex or precedent-setting cases. The second tier would be AI-assisted human adjudication, supporting perhaps cases of a less significant importance by aiding research and analysis, with humans making final decisions. The third tier, which I describe as human overseen AI adjudication, allows AI a much greater role in minor matters like traffic or small claims, always with human oversight for unexpected complexities. And then finally, fully automated processing without a human to handle routine uncontested cases, boosting efficiency. Even here, safeguards ensure that appeal rights and regular audits maintain fairness.

Now, integrating AI into the system offers benefits, but it also presents challenges. Ensuring transparency and accountability in AI-driven legal decisions is crucial, but too much disclosure has the danger of allowing individuals to potentially manipulate outcomes. This raises questions of legislative intent and fairness. Ethical concerns about AI-assisted judicial processes highlight the need therefore for robust oversight.

So, I propose a system of calibrated transparency which will offer varied levels of explanation, detailed breakdowns of the algorithmic process for judges and legal professionals, but more general explanations for defendants and the public, all published in accordance with the vital principle of open justice. I think regular updates to the AI parameters will help to prevent gaming, and ongoing audits and adaptive measures are vital as new forms of gaming or imbalances emerge. These techniques can clarify AI decision-making by showing the importance of certain factors and using standardised explanation formats. Now increased explainability might lead to even more sophisticated gaming, and an undue deference to AI by judges. To mitigate that we should firmly maintain the principle that AI must remain advisory, with a human rationale required when there is a divergence.

Now, I just want to bring us back to some of the current challenges. We’ve got the dangers of synthetic media and deep fakes. We’ve got these realistic fake images around us all the time. They pose real risks to legal proceedings. They make the authentication of evidence difficult. They can lead to potentially unfair trials or reliance on what should be wholly inadmissible evidence. I believe that some law reform is needed to ensure that we criminalise the use of harmful synthetic content in courtrooms.

Another challenge posed by AI is future reliance on the court system itself. As AI becomes widespread in daily life, public trust and reliance grow, but this might lead to private dispute resolution via entirely automated systems, making state courts less relevant, less central for private cases. Now, this shift should raise concerns in all our minds about how we evolve the law itself through accessible public judgments.

So, whilst AI is increasingly used for administrative and now some civil legal tasks, justice systems should proactively establish ethical and professional standards. And in England and Wales, we must ensure that the ‘do no harm’ principle is followed. As a member our Inn’s IT and AI committee, I will do all I can to ensure that The Inner Temple is at the forefront of thinking on this most exhilarating, yet bewildering, of challenges with a sense of humility and a willingness to engage in genuine collaboration. Then AI can expand access to justice whilst upholding the rule of law – a true alignment that will retain that vital human element of justice.

The Rt Hon Sir Robert Buckland KBE KC
Foundry Chambers

For the full video recording:
innertemple.org.uk/futureofai