I love AI, I
should because we have been using it in eDiscovery technology for more than a
decade, but of course Gen AI is different, newer, exciting BUT has risks, none
more so than in civil litigation and eDiscovery.
We have all seen
incidents in cases in various parts of the world, including more than our fair
share in South Africa, where the use of Gen AI has resulted in false citations
or fictitious legal refences. Quite rightly, the lawyers or law firms involved have
been criticised and, in some cases, sanctioned. Make no mistake, this is
serious and the consequences for law firms and individuals are far reaching. It
is not just the publicity and reputational harm, but the financial implications
and of course damage with clients.
Given all of the
above it is nor surprising that some regions are introducing laws or rules
governing the use of Gen AI and I have been particularly drawn to a Practice
Direction on the subject in the NSW State of Australia.-
This became
effective in February 2025 providing clear guidelines on the use of Gen AI in
proceedings and as such it has implications for eDiscovery and AI driven tools.
A very brief summary of the guidelines is as follows:-
·
Gen AI can be used for drafting chronologies,
summarizing documents, indexing briefs and reviewing documents but NOTE its use
for generating affidavits or witness statements is prohibited without strict conditions
·
Warnings concerning hallucinations e.g. false
citations or legal references and the need for independent verification.
·
No information subject to non-publication, including
for example subpoenaed documents, to be entered into Gen Ai tools without
proper confidentiality safeguards
·
Gen AI can only be used with the Court’s prior
approval and any Gen AI involvement must be fully disclosed in experts reports
·
Given that Gen AI is useful for document review,
partes must be cautious of confidentiality risks and ensure the accuracy of
outputs. This Practice Note stresses responsibility and transparency in AI
usage.
I was so
impressed with this Practice Directive that I have forwarded the link to the
Rules Board of the DOJ in South Africa. For me it is difficult to disagree with
anything it says as I become more and more concerned with non-qualified or
experienced people thinking Gen AI is their saviour.
Furthermore, it
makes me so comfortable adopting the use of Gen AI within a responsible
software solution such as RelativityOne just so long as its use is restricted
to those who understand the solution, and who are experienced and qualified. I
have written previously on the longstanding and documented principles regarding
AI adopted by Relativity so I won’t repeat them here but happy to provide
further information if required. However, I want to make some detailed points about the privacy
and confidentiality aspects, subjects which are dear it my heart in eDiscovery.
Relativity
utilises its partnership with Microsoft Azure OpenAI for the use of Gen AI in
its products such as aiR for Review, Privilege and others, and I put the
following words in emboldened caps for emphasis – NO DATA IS STORED IN AZURE
OPENAI – when using RelativityOne’s Gen Ai tools. Furthermore, in
RelOne, the prompts (inputs) and completions (outputs), the embeddings and
training data:-
·
are NOT available to other clients
·
are NOT available to Open AI
·
are NOT used to improve OpenAI models
·
are NOT used to improve any Microsoft or third party
products or services
Following the
NSW Practice Note and using RelativityOne means full defensibility and accountability
It will be
interesting to see if SA ever adopt a similar Practice Directive but whether
they do or not, my message is a simple one – Avoid the risks and leave it to
the professionals