skip to content

A Q&A With Everlaw Founder and CEO on ChatGPT and Generative AI in Legaltech

by Colleen Haikes


ChatGPT from Open AI has truly captured the imagination of the legaltech community. Launched Nov. 30, 2022, this tool marked a rapid acceleration in the development and implementation of generative AI — with the latest iteration of GPT coming out just yesterday.

We spoke with Everlaw Founder and CEO AJ Shankar to hear his thoughts on this generative AI tool. As a PhD in computer science, AJ enjoys delving into new technology. The Everlaw engineering and product teams have been testing ChatGPT since it launched. Our respect for users means as a company we are providing due diligence in our analysis of large language models like ChatGPT to understand the potential impact, opportunity and risks for our customers.  


What is your big picture impression of ChatGPT? 

Large language models (LLMs) like the ones used in ChatGPT are a promising innovation in the world of AI for generating content that could have meaningful applications across business, education, government and the legal field. Here’s an article I found to be helpful in explaining how it works. LLMs are a significant step forward from prior AI in many ways: fluency of interaction using natural language, ability to synthesize information in novel ways, and what I would consider an impressive level of sheer creativity. At the same time, by their nature, they are constrained in what they can reliably do (more on that below). Still, they’re an amazing breakthrough, and ChatGPT has really set the standard here, though of course there are other tools in the works, such as Google’s Bard. We should expect to see lots of innovation as a result of this competitive environment.


We are always proactively looking at innovative ways to take advanced technology – such as ChatGPT and Google Bard – and bring them to the litigation and investigation market. A great example is how we used new algorithms based on a hierarchical density-based model to implement our concept clustering technology. With ChatGPT-like technologies, we've actively been looking at a number of areas where we can help legal teams better understand the corpus of digital evidence they work with. For example, LLM models are very strong at summarizing documents and we are looking at some unique approaches to using that strength to help our customers.


What are some of the challenges you and the Everlaw team have seen? 

There are issues around data privacy, which is core to our customers’ work. For example, in the case of US federal customers or EU-based data, there are restrictions on where data resides based on FedRAMP and GDPR, respectively, as case materials would be transferred out of their repository for processing by ChatGPT. At some point if there were an open source version, we could fine tune and host it ourselves. Additionally because LLMs can so easily “hallucinate” false information, we’d want to make sure that any usage was strictly additive to the user experience, and that users could easily see and corroborate or iterate on any output from such a tool. Everlaw is a truth-finding machine – so truth is not one of the areas we could compromise on for our customers.

Additionally, ChatGPT is hard to apply to the scale of ediscovery work, which can number millions of documents and diverse evidence – including video.


Speaking of the state of AI today, what kinds of AI innovation does Everlaw have now and how could that evolve with or without ChatGPT?

The good news is that AI has already opened up a new world of possibilities to legal teams. Everlaw Clustering (unsupervised machine learning) and predictive coding (supervised machine learning) are game changers. For example, with concept clustering, legal teams can sort through and understand millions of documents for full review or early case assessment, without any human input required. Everlaw Clustering presents findings in an intuitive, visual format that encompasses both a 30,000-foot snapshot and a granular, down-to-the-document view.  Everlaw is dedicated to delivering the most impactful AI advances and will continue to integrate it on our single platform to make it intuitive and integrated with all of your ediscovery work.


In the world of unintended consequences for new technology, what could be some of the sticking points? 

There is no thinking or conscious understanding going on in these tools, and at the same time they’re incredibly open-ended so you can get them to go off the rails without too much difficulty. By now, we’ve all already seen the results of many pundits and experts’ tests of ChatGPT in the wild, such as New York Times reporter Kevin Roose's experience. We think the best way to get real value without unintended consequences is to bound the interactions by providing lots of clarity to the tool on exactly what task we want it to do, and also likely constraining the number of back-and-forths as well.


How does Everlaw’s value of “Respect for Users” play in this process? 

Everlaw believes that creating a product that users actually love is a fundamentally honest and satisfying way to build a product and company. A corollary: the only reason to add a feature is if it meets a real, pervasive user need; making Everlaw easier to sell or market is not a sufficient reason. That means we are not jumping on the generative AI bandwagon to get a quick hit or promote ourselves. We are looking for substantive improvements to the user experience that are consistent with the rest of the platform. We’ll be thoughtful and careful as we evaluate customer value. 


Do you feel any legaltech company that integrates ChatGPT into its platform gains a first-mover advantage?

We’re at an interesting time: there is tons of potential in the space, but also high variance from one experience to the next. We also think that the standard for quality and correctness in our industry is (or should be) higher than most. So we think that it's critical to think through innovation – specifically on the integrity of doing the right thing and the discipline of doing it the right way.That’s how we’re approaching this exploration. So while I do think there is some value to being a first mover, we are more interested in being the best mover, so to speak. We won’t take forever to do this, either, but we’re focused more on great than fast. Fundamentally, we think LLMs are here to stay, and so we’re thinking about the long-run here as well.


With the data haystacks growing 100x in the last 15 years, it’s clear that the legal teams using AI in their litigation and investigative work will gain substantial advantages over those who don’t. We are entering a new era of AI and advanced analytics in the legal field that help not only with generative tasks but with the unprecedented scale of data visualization, textual analysis, diverse data types and prediction – from hot docs to winning arguments. Everlaw is well positioned as the most advanced, future-proof platform for legal teams. 

[Ed. note: This post was not composed by ChatGPT]