The new dilemma: AI and public law

Can artificial intelligence replace professionals in courts and write judgments in public law? Richard Clayton KC looks at the issues involved.

When discussing AI’s role in replacing advocates and writing judgments in public law cases, it is important to recognise that this involves speculation about AI’s future development, and is not rooted in AI’s current technological capabilities.[1]

Secondly, even assuming the technology enables AI for advocacy or judging in the general run of civil cases, my experience as a King's Counsel and Deputy High Court Judge suggests a cautious approach to whether AI can replace advocates or judges in public law cases, because of the particular nature of public law work.

But I shall begin by saying something about what I mean by AI.

What I mean by AI

“Artificial Intelligence’ refers to computer systems which are intended to replicate human cognitive functions. This includes ‘machine learning’, where algorithms detect patterns in data, and apply those new patterns to automate certain tasks.[2]

Typically, an AI system is provided with a ‘training set’ of data about the subject and its algorithms then identify relationships within the data. This ‘training’ can be based on humans correcting machine responses , or simply by the system responding to feedback from its environment. For instance, the AI system developed to play the game “Go” was trained on over 30 million moves and was able to come up with moves which no Go experts imagined (Knight 2016) when it beat the world champion five games to nil.[3]

When considering the impact of AI it is instructive to reflect on how long computers have affected chess.  The first computer chess was invented in 1950 by Alan Turing, the English mathematician and computer scientist who was featured in the 2014 film, The Imitation Game, starring Benedict Cumberbatch as Turing, who decrypted German intelligence messages for the British government during World War II. 

Currently, systems like Siri and IBM’s Watson[4] can follow simple spoken or typed commands and answer basic questions. But they cannot hold a conversation and have no real understanding of the words they use - which highlights the difference between learning and understanding in these systems. Furthermore, language is difficult, and linked with common sense, For example, if the current systems are told that some people had to wait ‘forever’ for a table in a restaurant, it will automatically decide that the literal meaning is improbable, and the people probably just hang around for a long time and become annoyed: see Knight[5]

Therefore, AI has not as yet surpassed human decision making. A more long-term objective is ‘General AI’, where human intelligence is mimicked so that any task can be undertaken. This, in theory, could lead to systems that are more intelligent than humans, a point known as ‘The Singularity’. Advances in AI will evolve not just from the development of new algorithms and algorithmic combinations but from the increasing power of computers themselves.

As of 2018 areas of AI application included:

  • Document analysis;
  • Contract intelligence to scan contracts and other legal documents and present the information;
  • Document delivery;
  • Legal adviser support IBM’s Watson develops legal advice by asking the computer a research question in natural language, like a person. The computer then reviews the relevant law stored in its system, gathers evidence, draws inferences and returns highly relevant, evidence-based candidate answers. It also monitors the law around the clock to notify users of new court decisions that can affect a case. The programme continually learns from the lawyers who use it to bring back better  results each time.
  • Clinical négligence analysis;
  • Case outcome prediction- Researchers at University College London, the University of Sheffield and the University of Pennsylvania applied an AI algorithm to the judicial decisions of 584 cases that went through the European Court of Human Rights (ECtHR) and found patterns in the text.[6]  Having learned from these cases, the algorithm was able to predict the outcome of other cases with 79% accuracy. It found that rather than legal argument being predictive of case outcomes, the most reliable factors were non-legal elements: language used, topics covered and circumstances mentioned in the case text.
  • Public legal education

At the moment, however, human expertise dominates at high skill complex tasks, whilst machines lead on low skill, routine tasks which are administrative or process based- although this is likely to change over time when technology changes rather than replaces the work of lawyers. 

Professor Richard Susskind whose positive views on technology and legal practice are very well known, argues that it is important to differentiate between custom made or bespoke legal work and low level commoditised work. He rightly points out that most lawyers insist that their own work as  bespoke.[7] He also argues that a second wave of AI will result from three developments: systems that can analyse huge amounts of legal materials, those that answer questions and solve problems in an apparently intelligent way (like the Go game I described earlier) and affective computing which deliver systems which can detect and express emotions.[8] He rejects the AI fallacy that the only way to get machines to out perform the best human lawyers is to copy the way human lawyers work. He says that human lawyers will be outgunned by brute processing power and remarkable algorithms which operate large amounts of data.  Nevertheless, Professor Susskind acknowledges that there is no computer programme that can generate legal argument;[9] nor can a computer produce a reasoned judgment, even in easy cases where there are no complex issues of law.[10]

The role of AI in public law cases

Even if it were technically feasible for AI to replace advocates or judges giving reasons at some stage in the future, there are particular reasons why AI would not suit public law cases .

As a UK barrister and part time High Court judge who regularly acts in in public law and commercial cases in both the UK and the Caribbean, I take the view that there are special features in public law cases which limit scope for AI:.

  • Public law focuses on the exercise of discretion by public bodies.
  • Public law cases do not entitle a Court to overturn decisions by public bodies with which it disagrees. Instead, the Court must decide that the public body has acted unlawfully. Administrative law “is not intended to take away from those authorities the powers and discretions properly vested in them by law and to substitute the courts as the bodies making the decisions”.[11]
  • Public law cases involve claims based on administrative law principles (that the public body has asked the wrong legal question, has operated an unfair procedure or has acted irrationally) or breached the Constitution (in the UK, a breach of the Human Rights Act 1998).
  • Therefore, public law cases require a clamant to show that the public body has acted unlawfully. So public law cases raise legal issues, sometimes tricky ones which raise general issues of public importance.
  • In deciding whether the public body acted unlawfully, it is vital that the Court considers how the legal principles apply to the particular facts of the case when viewed in context.[12]
  • In public law cases orders for discovery or disclosure of documents by the parties are rare,[13] and witnesses are very seldom cross examined[14]

Public law cases are, therefore, very different from judgments of the ECtHR which led to impressive results of the UCL study, as previously mentioned. The ECtHR is an international court. Its purpose is not to sit as an appeal court which overrules domestic courts. Instead, the ECtHR finds facts and decides the application of European Convention principles, firmly recognising that domestic courts are the primary decision makers, by virtue of the principle of subsidiarity.[15]  As a result, very few applications before the ECtHR succeed, about 5%.[16]

The upshot is that it is difficult to see how AI can help in cases which depend on pure arguments of law where the submissions and judgments have to be sensitive to the context and involve evaluative reasoning. 

I would therefore suggest that, even if AI became technically feasible to replace advocates or judges, it more likely to assist in civil cases which involve straightforward factual disputes between witnesses than in public law cases which have a high legal content.

Richard Clayton KC is a barrister at Exchequer Chambers.

This paper is based on a seminar paper given to the Young Commonwealth Lawyers Association Virtual Conference on 8 November 2023.  Richard can be contacted at This email address is being protected from spambots. You need JavaScript enabled to view it.

[1]  This distinction is highlighted in Harry Sudin’s illuminating article Artificial Intelligence and Law: An Overview Georgia State University Law Review 2019.

[2] See the very instructive analysis of the Law Society of England in Wales in Horizon Scanning: forward thinking Artificial Intelligence and the Legal Profession May 2018 

[3] Professor Richard Susskind in Tomorrow’s Lawyers (3rd edition, 2023) at pp 22-23 repeats this story in saying that it is his favourite example of the progress made in the field of machine learning.

[4] IBM’s Watson is a question answering computing system that IBM built to apply advanced natural language processing, information retrieval, knowledge representation, automated reasoning, and machine learning technologies. It is being used to develop applications in healthcare, the pharmaceutical industry, publishing, and biotechnology, as a teaching assistant and for weather forecasting. It is also available to third parties to develop their own applications.

[5] Knight, Will (2016), ‘AI’s language problem’. MIT Technology Review.

[6] “Artificial intelligence 'judge' developed by UCL computer scientists"”, The Guardian, 24 October 2016  

[7] See Professor Richard Susskind in Tomorrow’s Lawyers (3rd edition, 2023) Chap 4

[8] See Professor Richard Susskind in Tomorrow’s Lawyers (3rd edition, 2023) Chap 22

[9]  See Professor Richard Susskind in Online Courts and the Future of Justice (2019) p 156 ” we are not yet there, not by a long way”. 

[10] See Professor Richard Susskind in Online Courts and the Future of Justice (2019) p 280-281.

[11] Chief Constable of The North Wales Police v Evans [1982] 1 W.L.R. 1155, Lord Hailsham at 1161A. 

[12]  In the human rights case R (Daly) v Secretary of State for the Home Department [2001] 2 A.C. 532 Lord Steyn stressed [28] “even in cases involving Convention rights. In law context is everything”.

[13] Tweed v Parades Commission for Northern Ireland [2007] 1 A.C. 650 

[14] Save Guana Cay Reef Association Ltd  v The Queen  [2009] UKPC 44, Lord Walker at [47] “orders for discovery and cross-examination are still exceptional in judicial review proceedings, for good reason. Such proceedings are essentially a review of official decision-making, and need to be determined without any avoidable delay”.

[15] As stressed long ago by the ECtHR n Handyside v United Kingdom (1979-80) 1 E.H.R.R. 737 [48], the “Court points out that the machinery of protection established by the Convention is subsidiary to the national systems safeguarding human rights. 27 The Convention leaves to each Contracting State, in the first place, the task of securing the rights and freedoms it enshrines. The institutions created by it make their own contribution to this task but they become involved only through contentious proceedings and once all domestic remedies have been exhausted (Art. 26 )”.

[16] The Annual Report 2022 of the European Court of Human Rights states that the ECtHR decided 39,570 cases of which 1,992 succeeded. 
4-5 Grays Inn Square Governance