The Role of AI in Disputes

While lawyers have had various forms of AI available to them for years, it is generative AI and the development of large language models (LLMs) which is likely to represent a fundamental shift for dispute resolution. This technology now offers language capabilities that have never been seen before, and is likely to transform the way lawyers conduct proceedings.

The key: a machine that understands language

The key characteristic of LLMs that is relevant to dispute resolution in any context is their natural language processing capacity, i.e. their ability to "understand" words and concepts. This is the result of the initial training process that LLMs have gone through, where – in highly simplified terms – the models were trained to create a map of human language. This enables them to understand words, sentences and paragraphs and also draft the same. Fundamentally, this is a characteristic that does not require any further training, rather it works "out of the box", although further finetuning is of course possible.

Tedium no more? What LLMs can already do for disputes lawyers

This significant technological advance means that dispute resolution lawyers will be able to make use of technology that is able to independently carry out document-related tasks such as summarising documents, answering questions on a document set, categorising documents based on which issue they relate to, extracting names or figures from a document, preparing chronologies, etc. LLMs are also able to assist with drafting any kind of text with suitable prompting. LLMs are not limited to language – products that work with sound and are able to provide meeting summaries (not transcripts) of calls are already commercially available. LLMs can equally work with images, both still and live footage, if needed.

The advantages are obvious as time-intensive and tedious tasks are plentiful in the litigation process. Especially in relation to disclosure, the heavy lifting could be taken on by LLMs. They could also assist with summarising calls, selecting documents for bundles, drafting timelines or interrogating sets of documents for specific issues. Drafting is another area where LLMs will be able to assist by providing first drafts of correspondence, questions for witnesses, or other documents. Generally, immense productivity gains can be expected from using this kind of technology, freeing up human time. Separately, LLMs that have been connected to legal research databases are coming onto the market now, which promise to fast-track complex legal research tasks. 

Real risks, and real mitigation strategies

The risks in relation to this technology are very real, but possibly over-emphasised in a traditionally risk-averse profession. Much has been made of the risk of hallucinations, i.e. the propensity of LLMs to occasionally provide answers that do not correspond to reality – they are making things up, or more accurately, providing an output based on a prediction that happens to not match reality. Risk mitigation strategies may well be sufficient to address this problem. The main ones are to force an LLM to cite documentary sources for every answer it gives, or to use so-called grounding techniques to check the output against a source of truth. Issues of bias in the training data will remain a problem, although guardrails attempt to mitigate this issue. In terms of security and data protection, special enterprise versions of LLMs promise a secure environment for client data where nothing is used to train the underlying LLM. 

What is now and what lies ahead

The advent of LLMs should be seen as a pivotal moment for dispute resolution lawyers. This is not a prediction of the far future – rather, everything discussed in this article is technology that is available right now. It is also likely that even more complex tasks will be performed by LLMs at a later point in time, although it is currently difficult to predict what this will look like exactly. LLMs are constantly being improved and developed at breakneck speed, and lawyers should take note that major new developments tend to appear in this space as a matter of weeks or months, rather than years.

Discover more insights on the AI guide