Episode 4 – Guest Post – Benefits of AI in legal research (written by Martina Domenicali, Co-Founder at Lexroom.ai)

Ep. 4 image

Legal research is one of the most intriguing and, at the same time, painstaking activities for lawyers.

As a legal tech startup with a focus on continental Europe and civil law countries, we have spent a considerable amount of time asking ourselves which activities added low value and could be disrupted or seriously affected by the implementation of generative AI.

This entailed months of conversations with lawyers in law firms, institutions, and corporate legal offices, along with countless hours of internal conversations. After ruling out several alternatives, we opted to focus our efforts on legal research. Despite technological advancements, it is surprising that most lawyers and trainees in continental Europe still rely on keyword searches, which often return hundreds, even thousands, of out-of-context documents.

Did you know that, according to our market research, each lawyer spends, on average, 80 hours a month on legal research? And that a lawyer spends at least four hours to complete a legal research task, and at least another hour to formalize the research into a legal opinion?

This is one of the most practical examples of how generative AI could impact your legal work and your business. Indeed, generative AI platforms, when correctly implemented, could make it possible to complete hours of legal research in minutes, maybe seconds.

Two aspects of this statement require a deep dive.

First of all, how is it possible to complete hours of legal research in a short amount of time, and what does this imply from a technological perspective?

The answer is quite clear. With the implementation of natural language models, lawyers can ask legal queries in natural language, providing the necessary details in the question, and the model will take this context into consideration for a more precise semantic search. Moreover, the legal sources that are found can be summarized by AI in a legal opinion draft, which can provide a first answer to the query that will be further developed by the legal professional. This technological shift enables a revolution summarized by two elements: the ability to ask legal queries in natural language and to generate a first draft of a legal opinion from the synthesis AI provides based on the sources found.

Secondly, and most importantly, how can we limit hallucinations?

To address this question, we decided to focus on the starting point. Reliability and use of AI for professional purposes depend on the legal data used. As we say in business, “garbage in, garbage out”. For this reason, we carefully selected the legal sources that will be indexed for the model to work. We did this by building the product through modules specialized for practice areas.

This also means considering another element. The generated legal opinion needs to be structured in a way that cites the relevant legal sources that have been used to take the information. Otherwise, as recently seen in the US with the famous Michael D. Cohen’s case, the risk of hallucinations presents a problem for lawyers using these drafts.

As a legal tech startup, our challenge is not merely to demonstrate the benefits of AI for research. Our collected data shows us that customers save on average 70% of the time compared to traditional legal research methods.

The hard job, indeed, is to benefit from this great efficiency without losing the reliability of the information used. And this is only possible if there is clarity regarding the legal sources the model uses and if the model can precisely cite the sources in the generated draft.

Quite a challenge, isn’t it?