AI is here to stay, but be aware of its limitations

The legal world is no different. In May, the Solicitors Regulation Authority (SRA) in England and Wales authorised the first legal firm, Garfield.Law, to provide legal services purely using AI. The core offering is an interface set up to allow automated pursuit of small debt claims. It is thought this can be a cost-effective way of increasing access to justice.

The SRA has provided its own detailed views on the risks posed by the use of AI. However, it was satisfied that there were appropriate processes in place to check quality, ensure client confidentiality and identify conflicts of interest. The solicitors behind the business would be held accountable for any breaches.

The SRA emphasised that: “We have also checked the firm is managing the risk of ‘AI hallucinations’. The system will not be able to propose relevant case law, which is a high-risk area for large language model machine learning.”

This risk has already materialised in a number of examples. The most publicised case involved a lawyer in New York in 2023. Steven Schwartz used ChatGPT to prepare submissions in a personal injury claim, which resulted in reference being made to judgments which did not exist. In his defence, Mr Schwartz showed that he had sought confirmation that cases were real, which ChatGPT had duly provided.

In other cases, it has been concluded that the consequence of using AI large language models in a way which results in false authorities being cited at court, is likely to be referral to a professional regulator, as it is a lawyer’s professional responsibility to ensure that checks on the accuracy of citation of authority or quotations are carried out using reputable sources of legal information.

This issue has now arisen in the Scottish Courts. In the recent decision of Jennings v NatWest Group, the Pursuer sought to appeal against a decision to dismiss an action brought under the Equality Act 2010.

In refusing the appeal, Sheriff Principal Ross noted that the Pursuer had prepared submissions by utilising ChatGPT. This resulted in “free-form legal propositions with only limited links to the facts” being advanced and at least three cases which were relied upon appearing to be non-existent.

The power of AI, when harnessed correctly, can be remarkable. Complicated and lengthy tasks can be completed in moments.

It is here to stay and those not utilising it will quickly be at a disadvantage.

However, its limitations remain and many examples of this have been reported upon.

It is readily apparent that there is very significant risk using “open source” AI products, which cannot provide assurances regarding quality.

Those involved in legal processes are well advised to remain diligent in scrutinising what they, and indeed their opponents, produce in legal claims to avoid falling foul of these problems.

Steven Smart is a Partner, Horwich Farrelly