Jane Colston discusses the use of AI in the courts under the new disclosure pilot scheme’s push for cheaper tech-assisted review
Lord Burnett of Maldon has set up a new advisory board made up of senior judges and artificial intelligence (AI) experts to consider and advise on the issues arising from the use of AI in the courts. In setting up the group, Lord Burnett seeks to ensure that judges are at the forefront of governing how increasingly capable AI technology may be used in and by the courts of England and Wales.
The growing use and impact of machine-learning AI technology in and around civil litigation seems inevitable and the new disclosure pilot scheme for the Business and Property Courts in England and Wales is giving litigators a good push in this regard. The pilot scheme came into effect on 1 January 2019 and will for the next two years apply to both new and existing civil proceedings where no order for disclosure has been made. It has been characterised as a ‘wholesale cultural change’ to disclosure in court proceedings in England and Wales, not least because it asks litigators to accept they may miss out getting all of the relevant documents from their opponents. The scheme’s antidote to any fear of missing out is the cost efficiency it promises clients.
Under the scheme, from the very outset of proceedings, the parties and their legal representatives are subject to obligations regarding disclosure. For example, known adverse documents must always be disclosed while ‘initial disclosure’ (a concept borrowed from international arbitration proceedings) requires that each party serve the key documents on which they rely at the same time as they serve their statements of case. Under the scheme there is no presumption that a party is entitled to disclosure beyond this. I suspect in complex litigation this will rarely just be ‘it’ as the court can make an order where it is persuaded that it is appropriate to do so to fairly resolve one of the ‘issues for disclosure’. What I think is likely to make a substantial difference is the pilot scheme putting squarely on the table the need for the parties to use technology-assisted review software (TAR) and techniques, with a view to reducing the burden and cost of disclosure, ie the application of new machine-learning AI technology to do the actual review of data. Some TAR systems can be used on data sizes of 50,000-plus and actively learn what is relevant and what is not from initial human review. This learning can then be applied by the algorithm to the balance of the data through a process of ‘predictive coding’ to identify relevant documents.
Under the pilot scheme, the court may give directions in regard to TAR’s use. The scheme’s requirement that parties discuss and seek to agree the use of analytical tools, including TAR, has come a long way since Pyrrho Investments Ltd v MWB Property Ltd & ors  EWHC 256 (Ch) when the court said: ‘…No English court has given a judgment which has considered the use of predictive coding software as part of disclosure in civil procedure. At all events, a search of the BAILII online database for “predictive coding software” returned no hits at all, and for “predictive coding” and “computer-assisted review” only [Irish Bank Resolution Corporation Ltd v Quinn & ors  IEHC 175].’
It remains to be seen whether the pilot scheme’s aims of reducing the length, complexity and, ultimately, cost of disclosure will be achieved in the long term. The recent case of Omers Administration Corporation & ors v Tesco plc  EWHC 109 (Ch) shows there is still ‘life in the old dog’ disclosure.
In Omers, Tesco was ordered by the civil court to disclose documents in their control obtained from the Serious Fraud Office (SFO). The materials included interview transcripts and witness statements that the SFO had obtained from third parties, using its powers to compel the production of information/documents under section 2 of the Criminal Justice Act 1987. The SFO had provided the documents to Tesco for negotiating a deferred prosecution agreement to settle the SFO’s investigation into Tesco’s subsidiary. In correspondence with Tesco, the SFO had stated that the SFO documents were provided ‘in confidence for the purpose only of providing legal advice in relation to the criminal investigation being conducted by the [SFO]’. Tesco was therefore caught between its disclosure obligations in the litigation and its obligations to the SFO. Omers highlights the tensions at play for companies facing criminal investigation and parallel civil litigation. In Omers the court had to consider the public interest in preserving the integrity of criminal investigations and protecting those who provide information to public authorities from wider dissemination of such information. Ultimately, the court held that the public interest in confidentiality, though usually of particular weight in the context of documents obtained by compulsion, must give way to the public interest in ensuring in civil claims that all relevant material is before the civil court.
Omers highlights how, despite the pilot scheme’s aims, disclosure disputes will likely rage on. In our experience so far, the pilot scheme is front loading substantial costs as the litigators’ fear of missing out is resisting a forced change in habit. For example, the parties’ obligations to agree the nature and scope of disclosure, is giving rise to a substantial amount of work and inter partes correspondence on negotiation of the issues and models for disclosure. What I think is likely to make a dramatic difference, therefore, is the judges and the pilot scheme getting behind the use of TAR, giving it the court’s endorsement and positively mandating parties to consider it and discuss its use.
The pilot scheme’s push for TAR’s use requires a philosophical shift, not least because the TAR technology is not perfect. However, the bench mark is not perfection but what is proportionate and reasonable. In Pyrrho, the English court was comfortable that the use of TAR/predictive-coding software did not lead to less accurate disclosure being given than, say, manual review alone or keyword searches. The civil courts here and elsewhere have been dismissive of the argument that TAR has to be held to a higher standard than human review.
AI algorithms can power through a review of around 600,000 documents per day faster and more accurately than any human team can manage. Once you are working on a large case, the volumes of data go beyond the realms of practicality and costs for a human to review. There is little doubt the pilot scheme’s aims will be achieved if TAR becomes commonplace in litigation and the courtrooms of England and Wales. The civil courts’, clients’ and public agencies’ (like the SFO) TAR nudge will be welcomed by those who understand how TAR, effectively deployed, can reduce the length, complexity and cost of disclosure, and understand the need to partner with tech-savvy professionals who know how the ‘AI black box’ works.
Jane Colston, partner, Brown Rudnick.
Jane Colston is a civil litigator specialising in commercial banking, contract, tort and shareholder disputes. She acts as a supervising solicitor regarding search orders.
The Legal 500 2018 names Colston as a Leading Individual: ‘A dynamic and creative practitioner’ who ‘brings a fabulous intellect, credibility and command of law along with incredible strategic thinking and client-facing skills’. Colston is part of the team which won The Legal 500’s award for Firm Specialist of the Year in 2018 and again in 2019 for Crime and Civil Fraud.
Who’s Who Legal Asset Recovery 2019 names her as one of the Global Elite Thought Leaders and states she: is ‘one of the best and brightest in the field’ according to peers who consider her ‘a top player’ in the asset-tracing space. They regard her as ‘an excellent litigation partner’ known for her ‘clear and concise’ approach.