Every Single AI Model Must Have a Human in the Loop — Great American Insurance Group Data Science Manager

(US & Canada) Nasser Mooman, Data Science Manager at Great American Insurance Group, speaks with Derek Strauss, Chairman, Gavroshe, in a video interview about his background and interests, AI and technology use cases, and AI challenges and ways to mitigate those.

Great American Insurance Group is engaged primarily in property and casualty insurance, focusing on specialized commercial products for businesses.

Mooman mentions computer science as his first interest and reflects on how the field has evolved from the development of computer chips to now generating AI. He further mentions that having a solid electronic engineering background has helped him understand the inner workings of the computer.

After earning his master’s degree at the University of Virginia in Canada, Mooman leveled up in computer science. His early research was focused on synchronization between images and sound, leading him to work in imaging, sound, and technology for years.

Next, he pursued a Ph.D. at the University of Waterloo, specializing in multi-agent systems and reinforcement learning. That is when Mooman began to understand the broader potential of AI and its mathematical foundation.

Over the years, he has worked with various tech companies, including Xerox and Blackberry, and with startups during the dot-com era. In addition to industry experience, Mooman has also taught at the Rochester Institute of Technology.

Apart from academia, Mooman has also worked with Blue Cross Blue Shield and started his own biomedical devices company in Waterloo. However, his venture was heavily impacted due to COVID-19, and that made him explore new opportunities, he later joined Great American Insurance Group as a Data Science Manager.

While Mooman joined the Great American Insurance Group before generative AI (GenAI) came into practice, he believes that it is turning theoretical research into tangible results that benefit everyone. He advocates that it is best to apply GenAI to tap into and utilize historical legacy data in the insurance domain.

Moving forward, Mooman shares two interesting use cases. He mentions building a key product during the biomedical device startup, which was an intelligent mattress designed to sit on top of hospital beds, capable of detecting and classifying pressure ulcers for bedridden patients.

The technology developed was driven by IoT, and IoT-driven AI was implemented, which was before the rise of GenAI. He mentions designing sensor-driven microcontrollers with AI models capable of classifying and predicting different conditions. The sensors in the smart mattress were designed to measure humidity, temperature, and pressure, he says.

Since cameras are often not allowed in hospital rooms for privacy reasons, the sensors were used to generate a heat map of the patient’s position in bed. Also, each microcontroller was equipped with AI to classify and predict humidity or pressure metrics.

Later, the technology was expanded to detect other conditions, such as whether the patient was in bed, out of bed, or how their posture changed over time.

The second use case also precedes generative AI and is related to social media, says Mooman. During the emergence of Twitter and Facebook, he recalls working on a graph database to detect influencers based on activities such as tweets, retweets, followers, and comments.

By using Twitter’s early data, Mooman developed the knowledge graph to identify influencers, enabling companies like Nike and Starbucks to target advertisements. Additionally, he shares that the IoT-driven AI for the smart mattress even led to a patent, publications, and medical trials.

Delving further into challenges, Mooman discusses how, with the internet revolution, there was a challenge in gaining social acceptance as society moved from paper-based systems to the digital world of mobile and internet technology.

Similarly, AI is set to become the next essential framework, much like the internet and mobile, says Mooman. With the emergence of internet and mobile technology, the challenges revolved around limited hardware capabilities and data access. These issues, however, have been resolved through hardware advancements like GPUs and the vast availability of data collected through social media or business.

Circling back to the insurance industry, Mooman confirms the collection of massive data, from routine claims to more intricate information. He adds that while AI offers significant benefits in the process, such as usability optimization, speed, and accuracy, it also generates challenges.

One of the biggest challenges with AI is ensuring ethical use, says Mooman. At Great American, this challenge has been addressed by incorporating robust data governance and ethical frameworks, ensuring responsible AI usage. More importantly, every single AI model must have a human in the loop, he says.

Elaborating, Mooman says that the job is to simplify life and not replace humans. Rather, it is designed to give them a cushion to make better-informed decisions. By addressing ethical guardrails, one can manage biases within AI models, knowing whether it is a gender bias or bias in the model itself.

This is further mitigated by ensuring that AI models undergo regular audits, security assessments, and continuous integration and evaluation, says Mooman. Trust is another significant hurdle, he says, along with enabling writers, agents, members, and clients to adopt technology without the risk of overwriting.

Concluding, Mooman says that it boils down to enhancing life safety without causing harm and compromising ethical standards by using data in a bad way.

CDO Magazine appreciates Nasser Mooman for sharing his insights with our global community.