Tech startups focusing on AI development in various industries, prior to the emergence of ChatGPT, are witnessing the advantages of generative AI. However, one AI startup called EmotionTrac, which is specifically targeting the legal sector, relies on human input alongside its AI product. EmotionTrac specializes in developing emotion-tracking technology for scenarios such as mock jury selection and trial preparation. Founder Aaron Itzkowitz emphasizes that their unique offering lies in the combination of AI technology and real human data, a significant advantage over other AI companies. The company has carved out a niche by assisting lawyers in gauging potential jury reactions when formulating legal strategies. EmotionTrac acknowledges the value of AI tools, including language models (LLMs), and employs them selectively based on their usefulness. These technologies are also made available to customers upon request. Nevertheless, Itzkowitz firmly believes in the continued importance of human expertise and interpretation even in conjunction with advanced AI tools. He contends that social science and interpretation continue to play a crucial role. Itzkowitz's standpoint is grounded in the fact that humans will remain central to the legal sector. Juries, for instance, will continue to comprise humans, as reliance on robots in decision-making processes is unlikely. Moreover, concerns around hallucination pose significant challenges for large language models. In the legal sector, lawyers turn to EmotionTrac's core AI technology, which leverages video cameras and facial recognition to track emotions.

Participants are required to complete a survey, providing demographic information, before watching a case-related video. Through facial recognition technology, EmotionTrac captures real-time data on participants' emotional responses. Subsequently, participants complete a post-survey, and the company aggregates the data obtained to provide insights to lawyers. To ensure accurate representation, all participants undergo screening, and the company tailors questionnaires based on the desired demographics requested by clients. Participants, who span diverse races, genders, and ages, are rewarded for their involvement. Itzkowitz asserts the company's confidence in the data it delivers to clients, citing successful recoveries totaling over half a billion dollars for their clients. EmotionTrac's technology has stood the test of time, providing lawyers with precise information required for case strategies. Additionally, it has proven to be a revenue-generating venture for the company, projected to achieve a growth rate of approximately 400% compared to the previous year. While EmotionTrac utilizes LLMs in certain cases for data analysis, metric interpretation, report generation, and trend identification, the company is not actively developing a large-language model for the legal sector like other well-funded companies such as Harvey AI. Itzkowitz points out the drawbacks of relying solely on LLMs, recounting stories of inaccuracies in the legal sector, such as generating faulty draft agreements. He acknowledges ongoing efforts in addressing these issues but underscores the necessity of human oversight in the industry. Itzkowitz recommends thoroughly proofreading all outputs from LLMs since complete accuracy is not guaranteed. EmotionTrac is exploring the possibility of allowing clients to directly query data using natural language processing via application programming interfaces (APIs) connected to LLMs. Itzkowitz believes this could provide added value to customers and is considering incorporating this feature into their offerings.