Legal AI Tech Startup Puts Humans, Not LLMs, at Center of Strategy
Many tech startups developing AI tech before ChatGPT burst onto the scene are seeing some benefits of generative AI.
But one AI startup focused on the legal sector. EmotionTrac, relies on human input for its AI product. The company develops emotion-tracking technology for mock jury selection and trial preparation.
"The difference that we deliver that other AI companies don't offer -- we're delivering a combination of AI with data from real people. That's a big bonus," said the company's founder Aaron Itzkowitz.
The company has found a niche in helping lawyers judge possible jury reactions they may encounter when developing legal strategies.
To be sure, EmotionTrac isn't dismissing LLMs completely and is using them where the technology offers value. It also offers the technology to customers requesting it.
But Itzkowitz believes that human expertise and interpretation remain crucial, even when using advanced AI tools.
"There's still an element of social science and interpretation," Itzkowitz said.
That makes sense -- humans will remain at the center of the legal sector. Robots won't be in juries that decide cases, and hallucination is a big issue with large language models.
Lawyers turn to the company's core AI technology, which involves tracking emotions of people using video cameras and facial recognition technology.
EmotionTrac asks participants to fill out a survey, which includes demographic information, and then watch a video about a case. The facial recognition tech specifically zeroes in on emotion tracking, and the company collects real-time data on how people are responding. The users then fill out a post-survey, which the company then aggregates and provides to lawyers.
"We screen everybody before they participate in the study, and depending on what … demographics the client wants to test against, we'll determine what questions are being served," Itzkowitz said.
Users get rewarded for participating in the studies, and they have to give EmotionTrac permission to access the front-facing camera. The participants are usually a 50/50 male-female split, and span all races and ages.
"We're certain about the data we deliver to the customer," Itzkowitz said. "Our clients who have used us have achieved over a half a billion dollars in recovery on behalf of their clients."
The technology is time-tested in providing accurate information required by lawyers, who make plans for tackling their cases. It is also generating revenue.
"We're on a trajectory now to be hopefully about 400% over last year," Itzkowitz said.
The company is using LLMs -- in most cases as a backup -- to analyze and interpret data and metrics, generate reports, and identify trends.
EmotionTrac isn't developing a large-language model for the legal sector, as other heavily funded companies like Harvey AI are working on that.
"You hear stories about LLMs in the legal sector where ... draft agreements generated were inaccurate. There are people working on it," Itzkowitz said.
Human oversight is still needed in the sector.
"I say … proofread everything you get out of [LLMs]... because it's not going to be 100% accurate all the time," Itzkowitz said.
EmotionTrac is exploring the possibility for clients to directly query the data using natural language processing via APIs that connect to LLMs.
"I think we might just incorporate that and give [customers] the added value," Itzkowitz said.