I feel like everything I read or hear is about AI these days. It’s everywhere. This week it is the brilliant and terrifying report from Dr Natalie Byrom (more excellent stuff from Nuffield Family Justice Observatory): Delegated by default? Privately owned technology in the family justice system – opportunities and challenges for access to justice.

It should act as a wake up call, but frankly, I’m not holding my breath. One of the things the report mentions is the need for some work on the LawTechUK website. That was intriguing as I hadn’t heard of LawTechUK so I went to find out what it was. Guess what? It’s an MoJ thing, and its been mothballed. It shows only a short message indicating ‘update pending’. One of the other things the report mentions is ‘pilot-and-discontinue cycles’, of which it appears this may be an illustrative example (other pilot-ad-discontinues are available).

Separately, someone alerted me to this equally alarming blog by Sam Freedman: ‘Drowning in Slop‘, which looks at the skyrocketing numbers of complaints to various regulators, tribunals etc, apparently driven by AI (e.g. increases of approaching 40% for employment tribunals and the Legal Ombudsman. This mirrors a reference at the annual open meeting of the Family Procedure Rule Committee on Monday to the threefold increase in the numbers of appeals from the Family Court to the Family Division of the High Court – sufficiently impactful to prompt adjustment to the rules in respect of appeals (the number of Family Division Judges has not risen by a multiplier of 3). I’ve made a Freedom of Information Requestabout this to see if its possible to get a bit more clarity about that. Although the Freedman piece suggests that FOI requests have doubled, apparently because of the ‘assistance’ of AI, mine is not one of them. I just hope that my request will not be slowed down by any AI generated backlog. Perhaps HMCTS will deploy AI to help them calculate the answers to my questions…?

As Freedman observes:

What officials are receiving is also much longer, harder to parse, more complex and often wrong. LLMs can help polish or improve an application or complaint from someone who broadly understands the situation and may just want some help with their writing or a point of detail. But confused prompts lead to inaccurate responses, and LLMs try to please, so will fulfil a request and find ways to try and make an incoherent case sound compelling. By throwing in everything, they also create lots of unnecessary work. […]

It can also make conversations with citizens recursive, as they feed an official’s answer back into an LLM and send back the answer they get almost straight away:

Overall the piece makes a compelling argument for how AI drives a breakdown in trust and relationships with professionals. Schools, social workers and local authorities report a worsening in relationships with parents. Much of what Freedman is worried about is obviously applicable to the families involved in family court cases. What it tells us is that litigants may choose to use AI to help prepare their submissions, meaning that

…they often do not understand the submission they have made, it is hard for them to follow the response or get proper support on their case. […] Social workers report similar issues. They are increasingly seeing vulnerable young people and families communicating at a level of sophistication well beyond their actual understanding, which can lead agencies to communicate back at the same level, creating confusion.

Read with Dr Byrom’s report, there is a real cause for concern:

Where high-quality provision exists, the growing use of general-purpose GenAI tools for search risks displacing it, disrupting decades of investment aimed at improving access to authoritative, accessible online provision. In family justice as elsewhere, there is emerging evidence that litigants are turning to these tools directly for advice and document preparation, despite the fact that GenAI tools frequently get the law wrong. Several features of GenAI tools create risks that are especially serious in the context of family disputes – which are relational and emotionally driven. GenAI tools are prone to sycophancy, affirming users in their beliefs even where they are incorrect – research demonstrates that using these tools even once can distort social judgment and discourage behavioural repair. Accordingly, the use of these tools risks transforming good-faith litigants into intransigent ones, undermining ongoing attempts by policymakers and professionals to de-escalate conflict and divert parties towards conciliation. This propensity may also affirm vexatious litigants and domestic abusers, while simultaneously enabling them to produce superficially convincing court documents at scale, weaponising litigation and overwhelming the already overburdened courts.

And that’s before you start to digest what Dr Byrom says about the use of AI by judges…