Last week the International Legal Technology Association (ILTA) announced its Generative AI Guide, in an effort to provide much needed certainty to litigators over ‘the acceptable use of GenAI in Court disclosure’. The Guide, co-authored by a cross-firm team of litigators, is an addendum to ILTA’s successful Active Learning Best Practice Guide which is currently under review by the Master of the Rolls. It is hoped that both Guides will receive the judicial stamp of approval in the coming months.
Its self-proclaimed purpose is ‘to provide practical industry guidance on the use of Generative AI tools… and approaches in preparing for and responding to court-ordered document review exercises’.
Underpinned by fundamental principles of transparency, proportionality and defensibility, the Guide embraces the use of GenAI in e-disclosure workflows, but stops short of setting out hard and fast rules around its implementation, the group explained.
Jamie Tomlinson of UK-based law firm DAC Beachcroft, a co-author, said: ‘The aim of the Guide is not to tell litigators what they can and can’t do with GenAI. The rate of technological advancement in this area is so fast that any prescriptive rulebook would very quickly become redundant, and we’d be back to square one. Instead, we have tried to outline the basic principles which should govern any proportionate disclosure exercise and extrapolate those out to fill the current regulatory gap in PD57AD and other regimes.’
The Guide’s central focus is the modest and auditable deployment of powerful GenAI technology, preferably in the context of existing Active Learning Workflows, they added.
It targets the ‘accountability and responsibility’ of parties using GenAI to streamline their review process; promotes ‘appropriate transparency’ through early-stage party discussions about why and how GenAI should be used; and suggests techniques for keeping detailed audit trails. Specific recommendations include:
Keeping a ‘separate audit log of prompts’ detailing which prompts were used and why, over which part of the dataset, and how those prompts were refined iteratively as the review progressed.
Consistent validation and quality assurance, including through testing, random sampling, and ‘calculating the precision and recall’ of the GenAI tool.
Identifying ‘documents or issues not suitable for GenAI’ at an early stage, both to reduce wasted costs and efficiencies, and to get the best out of the tools used.
Recording agreement between parties in ‘procedural documents’ like the Disclosure Review Document
‘We want litigators to understand the purpose and scope of their GenAI workflow and, importantly, be prepared to explain it to their opponents and the Court on demand. That first requires some demystification around what GenAI actually is, how it works, and what it can do – which we hope the Guide does. After that point, keeping clear and detailed records of the steps taken to build and validate a GenAI workflow will be really important,’ Tomlinson added.
Imogen Jones, also of DAC Beachcroft, and another co-author, concluded: ‘The Guide is by practitioners, for practitioners. The cross-firm collaboration behind the Guide allowed the drafters to pool a range of experiences – we very much hope those efforts encourage a collaborative approach by parties using GenAI in disclosure.’
The Guide will be officially launched at an event at Fieldfisher’s London offices on May 6th which several of the drafters will be attending.