How will AI Change the Law?
There has been a lot of speculation about how AI will change the law. Helena Kennedy, Nicolas Mostyn and Charlie Falconer recently interviewed Professor Richard Susskind on the issue in their Law & Disorder Podcast and Nick Abrahams (Norton Rose Fulbright) wrote an article in the Law Society Gazette.
One of the issues in both those contributions was what would happen to lawyers once a lot of legal work is done by AI. Professor Susskind said doctors do not ask that question. They ask how they can care better for patients using AI. Nick Abrahams points out that technology opens up new markets and there will be complaints and class actions where there were none in the past because it will be so much easier to bring them.
When we talk about AI in the legal context, we are probably thinking about large language models like ChatGPT (there are others that may be able to do specific tasks of course). Professor Hannah Fry points out that they are not encyclopaedias – or brilliant barristers or judges for that matter. She says you need to think about them as actors who are really good at playing a role and who will then use what they have gathered from the internet or other resources to respond in coherent language. So if I send a brief to a brilliant barrister, they will represent my client in court using their own legal knowledge, experience and oratory skills. If I sent my brief to AI, it may well come out with a good opening and closing speech, but it is not a barrister, but an actor who has played a barrister in films and uses those skills to perform (together with an assistant who has gathered legal information from the internet etc.). This may or may not come down to the same outcome. However, who will listen to those speeches – judges or AI?
If we are going to get a multitude of complaints and class actions because AI is making it easier, as Nick Abrahams says, who will decide on all those cases? While it might take a while for courts to use AI judges, no doubt complaints and ombudsman services are quicker to adapt. However, will they produce a fair result?
Let’s look at work, which I think is often underappreciated: drafting written witness statements. In family and civil litigation in England & Wales witness evidence is for the most delivered by way of written statements well in advance of a hearing. When as a solicitor I draft such a witness statement for my client I do not just act as a scribe taking down everything they say. I will advise my client to leave out irrelevant points, probe them on relevant issues, sort the points into a logical order, and use succinct language which a judge can read fast. I may also include words which I feel will speak to judges in general, but I do not know who the judge at the final hearing will be, what mood they are in on the day of the hearing or what my client (and indeed other witnesses) will say when giving evidence.
Often litigants in person do not appreciate how important witness statements are and what can and cannot go into them. They may think everything they may ever want to say should go into a witness statement and include submissions and their opinion (which is not strictly speaking allowed); they may ramble and include a lot of irrelevant points, not only leaving judges frustrated when faced with reading a long statement, but also risking that judges will skim-read them and overlook important points.
If cases were to be decided by AI alone, those drafting statements, written submissions and ultimately even the consumer contracts on which the case may turn, will know the AI judge and will know what key words or points may tip the AI judge’s decision one way or another. This is probably only something someone can do with an equally or more powerful AI which analyses submissions etc and the AI judgements. So if the budget airline uses such AI, it will be at a massive advantage because their contracts, delay and cancellation justifications, complaints handling etc. will be drafted in such a way as to have a high chance to get a favourable outcome from the AI decision maker if the passenger makes a claim. This puts consumers making claims at a disadvantage and we may end up with a two class system where the economically more powerful party can pay for the AI to deal with their claims and massively increase their chances of winning. Unless consumers have access to lawyers or claims handlers who are also using equally powerful AI, their chances of success could be minimal.
We could end up in a system where passengers tick a box on a third-party booking platform which will automatically make a claim in case their flight is delayed, airlines tailor all their written work towards winning a claim and an AI judge decides on the claim – all without humans having anything to do with the process at all. Whether that results in fair outcomes remains to be seen.
Tagged AI
AI is no use for legal research
I tried out AI to do legal research and here is the result.
I was looking at child maintenance the other day and the way the Child Maintenance Service (CMS) calculates child maintenance in specific cases. That is set out in Schedule 1 to the Child Support Act 1991. So far so good.
This has been amended several times and the government helpfully has the amended latest version on its legislation website. However, there is a paragraph that provides that some people do not have to pay child maintenance at all:
5 The rate payable is nil if the non-resident parent—
(a) is of a prescribed description; or
(b) has a gross weekly income of below £7.
How do I find out what “prescribed description” means? This is set out in secondary legislation, also called statutory instruments or SIs. They are also online on the government legislation website, but they are not cross-referenced in the act which they define further, nor are they updated. So an SI may have been amended by another SI, but the older SI will appear on the government website in its original wording. So unless I get a learned book on the topic (and even then finding this is not straightforward), I would have to traipse through dozens of SIs to find the definition. I did find the answer (by chance) in regulation 45 of the Child Support Maintenance Calculation Regulations 2012 SI 2012 No. 2677: prisoners, children and some students and residents in care homes etc.
I wondered whether AI could get there quicker. Since all this information is publicly online and it is just a question of reading all of it, cross-referencing it and finding the most recent version if there is more than one, it does not really take intelligence to find it, just the ability to read a lot of information fast and put it in date order.
I started with bing.com AI. I asked:
Hi. I’m looking for the statutory instrument which sets out the “prescribed purposes” for paragraph 5 of schedule 1 of the child support act 1991.
It told me:
I’m sorry, but I couldn’t find the specific statutory instrument that sets out the “prescribed purposes” for paragraph 5 of schedule 1 of the Child Support Act 1991. The information available on the legislation’s website1 does not provide the exact details you’re looking for. You may want to consult a legal expert or a law library for more specific information.
Polite but wrong. I tried Gemini from Google next:
Unfortunately, there isn’t a single statutory instrument that directly defines the “prescribed purposes” for paragraph 5 of Schedule 1 of the Child Support Act 1991. Here’s why:
-
- No specific instrument: Paragraph 5 itself doesn’t mention any specific statutory instrument for prescribed purposes. Instead, it references regulations made by the Secretary of State. This means the definition of “prescribed purposes” depends on the specific regulations in force at the time.
- Dynamic regulations: Child support regulations are regularly updated and amended. Past regulations might not be valid now, making it difficult to pinpoint a single instrument.
- Multiple regulations: Several different sets of regulations could contain relevant definitions depending on the specific scenario you’re interested in.
Right, yes, if the SI was mentioned in the Act, I would not have asked! So I thought: “Third time lucky.” and tried ChatGPT (free version):
The statutory instrument you’re referring to is likely “The Child Support (Information, Evidence and Disclosure) Regulations 1992 (SI 1992/1812)” which sets out the prescribed purposes for paragraph 5 of Schedule 1 of the Child Support Act 1991.
And which paragraph in that SI do I need to look at?
You’ll want to look at paragraph 3 of Schedule 1 of “The Child Support (Information, Evidence and Disclosure) Regulations 1992 (SI 1992/1812)” for the prescribed purposes related to paragraph 5 of Schedule 1 of the Child Support Act 1991.
Not only did ChatGPT randomly pluck out an SI (admittedly relating to child support), but it also referred to a paragraph of a Schedule which didn’t even exist. It simply made it up.
So when Elon Musk said that we would all be out of a job in a few years’ time because AI can do it, I am not getting concerned. (Despite having 10 surviving children, Musk also doesn’t seem to have cared for a child, older person or anyone else in his life or he would know that AI is hardly going to be able to do that job.)
Posted in Children Cohabitation | Tagged AI Child Maintenance Musk

