could ask a question and the AI would provide an answer by finding the best candidate documents in the repository or from the external legal knowledge sources? This would be useful if, say, a lawyer is asked to list the different rules for an employee’s right to work in California. Rather than scouring the numerous documents in the system, the lawyer could directly ask the chatbot the question and get the correct answer, complete with details of the reference used to formulate the answer. While this is precisely the type of task that generative AI is happy to tackle, can its answers be trusted? Again, we come back to grounding, and the importance of making sure that the large language model is being fed a diet of quality input on which to base its answers and generate content. Grounding ensures that instead of just asking the language model for its opinion, based on its ‘worldview’, it will provide an answer based on the text found in specific documents that the tool has been pointed towards as 48 LAWYER MONTHLY JUNE 2023 resources. This capability gives end users more control over the quality of the AI’s results – and thus, more confidence in its outputs. Less Drudgery Around Drafting What if generative AI could help not just with searching and finding legal knowledge, but also with helping to draft legal agreements? This is another potential use case for generative AI. The key, of course, is to make sure that the generative AI leverages what the organisation considers to be the best standards when it comes to writing. This is where firms will want to really lean on the knowledge within the organisation and tap into it. Is there a template for a share purchase agreement for mid-size tech startups that has been endorsed by the subject matter experts in the firm? Consider that as the gold standard that generative AI should be using to ground itself when helping to draft that particular type of share purchase agreement. Now, does this assist from AI mean that the human is taken out of the equation when it comes to drafting legal documents? No – it just means that more of the grunt work is taken out of the process of writing. While drafting a legal agreement, the lawyer could ask the AI chatbot to compare a specific clause against the current ‘market standard’, as determined by the organisation’s in-house experts. In some use cases, the legal professional could even go as far as starting off with a set of bullet points for a legal agreement and then have generative AI tap into the proper internal or external resource to help build out the various clauses with the appropriate language. At the end of the day, the human is still formulating the foundational intellectual underpinning of the legal agreement; they are just getting some help in fleshing out the wording. Enhanced eDiscovery eDiscovery has always been one of the more AI-driven processes in the legal world, which makes it an intriguing potential area for the ‘new kid on the block’, generative AI. Traditionally, eDiscovery has relied on supervised machine learning models to do automatic tagging and classification of data while combing through reams of potential evidence. Humans train machine learning models by showing them examples of what type of evidence should fall into what particular bucket, and eventually the machine learning model can tag and classify items with a high degree of accuracy. Generative AI could help out here in several ways. For starters, imagine a scenario where a file has been unearthed and automatically tagged a certain way during the eDiscovery process, but there is a question as to whether or not it would actually be useful or relevant to the case at hand. Using a ChatGPT-type interface, a legal professional could use perfectly natural human language to ask the AI whether the file meets the criteria or not, and
RkJQdWJsaXNoZXIy Mjk3Mzkz