Editor’s Note: The recent article, Seven Questions Lawyers Should Ask Vendors About Their AI Products, authored by Maura Grossman and Rees Morrison, provides a must-read overview of questions and considerations for lawyers to examine as they move from “asking about” artificial intelligence to “asking for it” as part of vendor products and services.
Seven Questions Lawyers Should Ask Vendors About Their AI Products
An extract from an article by Maura Grossman and Rees Morrison
The frenetic and much-touted world of artificial intelligence (AI) has poured into the legal industry like a storm surge. Lawyers who lack technical expertise or feel overwhelmed by jargon and arcane mathematical concepts are at a distinct disadvantage in this technology-oriented new world. Vendors can make assertions with little risk of cross-examination.
If your law firm or department has invited a vendor to explain or demonstrate its AI software, you likely already know the foundational questions to ask about the vendor’s company, competitive position, pricing, support, and user base. These days, you likely also know to ask about the vendor’s data protection and data security practices. However, you are probably on less solid ground concerning the questions to ask about the underlying machine-learning software. This article proposes seven basic questions – and a framework for understanding the answers to those questions – that are specifically targeted at vendors that offer AI and machine-learning products and services.
Question #1: What Do You Mean When You Say Your Software Uses “Artificial Intelligence” or “Machine Learning?”
A subcategory of artificial intelligence, machine-learning software finds patterns in data, and the software improves its performance (i.e., “learns”) as it processes more data. Data can include the words in documents – such as those contained in emails in electronic discovery or in word-processing files in contract analytics – which are analyzed using natural language processing or statistical methods. Data can also include figures from time and billing systems, where regression and neural networks can provide insights. Or data may be derived from human resources files, where classification methods, such as support vector machines or decision trees, can help identify records of interest or improve the quality of predictions.
The vendor should explain whether their software uses supervised or unsupervised learning. If supervised, your data will need labels (corresponding to classes or categories of interest, such as whether the client is a public or private company, whether the documents are privileged or not, or whether the practice group of a lawyer is corporate, litigation, or tax). In unsupervised learning, such as k-nearest neighbor classification, the software detects patterns on its own, based on the numbers in the variables.
What you should not hear from the vendor are grand, vague assertions, or that they cannot answer your questions because their software is based on proprietary methodologies.
- AI and Automation: From Removing Workers to Improving Quality
- Meeting with Vendors: Three Quick Qualification Questions