Last week’s Everlaw Summit in San Francisco offered a stage for Everlaw’s CEO and founder, AJ Shankar, to discuss foundational aspects of responsible AI development. Shankar dedicated a significant portion of his speech to elucidating three main principles crucial for responsible AI advancement and for attaining optimal long-term results for clients. These principles are privacy and security, control, and confidence.
Privacy and security are central to Everlaw’s AI approach. Shankar stressed the company’s stringent data retention policies, ensuring large language model (LLM) providers cannot store user data beyond immediate queries or use it for model training. “We ensure that they apply zero data retention to your data,” Shankar affirmed, underscoring a commitment to user privacy.
On the aspect of control, Everlaw emphasizes transparency and user autonomy. Features have been designed that clearly notify users when AI tools are in operation and what models are in use. This provides the transparency necessary for informed tool use, aligning with administrative capabilities that manage access and usage at various levels within an organization.
The third principle of confidence proves more challenging to instill. The potential pitfalls of generative AI, like hallucinations, are acknowledged. Everlaw approaches this by focusing on domain-specific uses of LLMs, embedding AI in the existing workflows, and thus enhancing precision and reliability.
Shankar additionally introduced the notion of changing the user’s mental model regarding computer interactions—likening AI to a “smart intern” that is both capable and fallible. Users are encouraged to regularly verify AI outputs, treating it as a learning process wherein they can discern where an AI tool excels and falters.
Notably, an attorney who beta-tested these AI solutions, Cal Yeaman of Orrick, Herrington & Sutcliffe, shared his initial skepticism towards AI tools in legal reviews, only to find the AI’s Coding Suggestions feature exceeding human review accuracy and offering significant cost reductions notwithstanding its remarkable speed.
The Everlaw Summit also saw the unveiling of other features alongside AI. These include multi-matter models for predictive coding, integration with Microsoft Directory for legal holds, and enhancements to clustering and data visualization tools. Beyond the technical discussions, the event’s atmosphere was charged by a workers’ strike at The Palace Hotel, advocating for fair wages and appropriate staffing. To learn more about the workers’ campaign, visit UnitedHere! and find endorsed hotels at FairHotel.org.
The summit featured keynote speakers like Shankar Vedantam and Kevin Roose, promising a well-rounded insight into AI and its evolving role in the legal landscape. For more details on the summit and its takeaways, visit the full article on LawNext.