Linklaters has outlined the early work that it is doing with generative AI technologies, which includes running a proof of concept of Google Cloud’s Vertex AI platform for search; launching a chatbot using Microsoft’s Azure OpenAI Service; and trialling AI assistant CoCounsel from Casetext, which was acquired this year by Thomson Reuters.
The magic circle firm has published details of its POCs to date on its website and speaking to Legal IT Insider, co-head of client solutions and innovation, Shilpa Bhandarkar said: “Generative AI is definitely a transformational piece of technology and people across the firm are excited about it and wanted to engage with it, so we thought about how to do that in a structured way.
“We have always embraced technology in the way we work but we needed to get our governance right and we also needed to understand what people want to use it for. No-one centrally is going to know the value of this on something like an M&A deal.
“There’s not enough time to do one after the other so we put together a steering group which includes our firm managing partner, senior partner, members of our executive committee and tech sector practice, our CTO and a senior risk lawyer. Marc Harvey, a senior lawyer in our LAI practice and I chair this group. We wanted to include people from our practice and business teams to cover all the different ways we need to be thinking about this technology.
“We have had the Ideas Pathway for a while so put that to good use in asking how people will use the technology. We had so many ideas, which is good for engagement but also gave us tangible use cases from our practices, not just us centrally telling people ‘this is what you can do.’
“We triaged the ideas and the thing that a lot of people asked for is something like ChatGPT, because a lot of use cases can be solved with a chatbot, so that was our first product.
“We produced and ringfenced the chatbot using Microsoft Azure OpenAI Services. Everything you can do in ChatGPT you can do using our chatbot, so you can upload a document and get it to summarise it or create a first draft, but you can do it in a safe secure place. The main thing is that we get to test it, to see what it does well and less well.”
Much like ChatGPT, Linklaters’ chatbot is capable of hallucinating and Bhandarkar said: “Nothing goes out without being checked.”
Bhandarkar points out that the need for accuracy is on a sliding scale, commenting: “If you are creating speaking notes for a marketing presentation, you don’t care so much about accuracy. If you’re drafting a clause for a contract, you very much care.
“The thing is that we’re all learning. We have innovation forums and have all the people who are using the chatbot talking to one another about what they find useful and not useful.” Linklaters is also training people across the firm on how to give prompts (aka prompt engineering).
In terms of the prominent use cases so far, Bhandarkar says: “People are using it for all sorts of different things. A lot of people are using it for drafting and content creation. Some are using it to build macros in Excel.”
The next big idea was search and Bhandarkar says: “We’re an institution that does so many deals across so many offices and our USP is our knowledge. The other possibility that this technology presents is the ability to surface knowledge at the point that people need it. If we have done 50 similar deals and you need something done quickly, no matter people to say it is really hard to find the thing you need. You know someone has written it, but it’s really hard to find. So that is the next POC that we’re running, the Legal Knowledge Search.
“When you think of search, you think of Google, so we’re collaborating with a boutique consulting firm called Kin + Carta and Google Vertex AI. We did a quick POC in the summer and the results were positive. It’s hard to find the document that you know exists even if you have the best filing system. If there is a document you need from that transaction, in that year, between those parties, and you get to that information quicker, that is incredibly valuable. Even if the results are 10% better, that is a huge result”
Vertex AI is a development platform that gives customers access to Google’s foundation models such as PaLM 2 so they can test and deploy them in their own applications.
Bhandarkar says: “This was just a quick test and we’re now taking it to the next step. We’re going to do a phase two to drill down further.”
She adds: “We’re not trying to boil the ocean but we know that technology is changing and we can take all of that learning.”
With regard to CoCounsel, Linklaters has run a global pilot to see how it can help lawyers with tasks such as summarising documents, preparing witness statements, extracting contract data and completing legal research.
The challenge now is to collate the feedback across these various POCs and work out which use cases to take forward. Bhandarkar said: “Part of our challenge is volume and complexity. What works for litigation in America might not work for a corporate lawyer in Japan. The challenge is to identify consistent use cases across the world and the next phase is to aggregate all the intelligence.”
caroline@legaltechnology.com