AI can Nix Bias, but Humans can Help Too

WCI 2025

AI isn’t nearly as new as you think, and you probably use it more than you realize, but does it have a role in bringing equity to workers’ compensation claims?

According to Ngozi Nnaji of Ako Brokerage Services, LLC and Nikki Jackson of Gallagher Bassett in the Diversity and Talent Alliance breakout at the Workers’ Compensation Institute, the way AI makes decisions can help build fairness into the claims process and focus on cultural competency along the way.

More than 50 Years Ago at Dartmouth

Nnaji started the presentation by providing the history of AI, noting that it dated back to "summer camp for nerds" (Nnaji's a self-proclaimed "insurance nerd," so she relayed that the term was one of endearment) at the Dartmouth conference in 1956 and then evolved into rule-based programming in the 1980s and later into deep machine learning in the 2010s before growing into what we now call artificial intelligence.

"We act like we fear AI, like it's coming to get us, it's coming to get us," Nnaji said, pointing out that every time Netflix suggests a title you might like or your GPS suggests a different route, AI's at play.

"People will say, 'Oh I don't use AI,'" Nnaji said. "Oh, yes you do, you just don't know it."

Why AI?

Jackson pointed out that AI has shown versatile capabilities for:

+ Safety ("How far away should the ladder be?")

+ Reporting claims

+ Adjusting

+ Predictive modeling

+ Nurse case management

+ Settlements and reserve amounts

+ Claim closure

+ Healthcare (e.g., medical records and patient journey maps)

Nnaji explained that AI holds strong potential for:

(1) Decision-making. "Experience helps you make the decisions, but there's a lot of factors that may be beyond your experience," Nnaji said. "What AI offers is consistency, effectiveness, and uniformity."

(2) Injured worker satisfaction and inclusivity. "If you're an injured worker, it's stressful, and you want to feel like you're being seen," Nnaji said. "You want to feel like you have overall equity in that you're getting what you need when you need it."

(3) Removing bias. "How do we remove bias?" Nnaji asked, highlighting the ability to prompt AI to limit the influence of bias in decision-making.

On the point of bias, Jackson presented two hypothetical adjusters who make decisions based on their experience -- and biases.

"But if we have historical data, we can make more informed decisions, so the AI can look through it and land on a decision," Jackson said.

Human Touch

Nnaji and Jackson cautioned that the technology can't do it all and advised that overreliance on AI can be just as problematic as under reliance.

"How do you balance automation and human judgment?" Nnaji asked.

For Jackson it's about asking questions and making sure there's still room for humanness.

"I want to make sure there's an out," Jackson said. "I want to be able to look at it and ask how did AI get the answers it got, how did you get to it?"

Jackson explained that AI doesn't know unique factors at work in a worker's claim.

"Maybe there's a family situation or a divorce or the person was getting a promotion," Jackson posited. "Maybe AI said this, but we need to change it to something else based on these factors."

Communication

While it's not usually the case that a bot is a better form of communication than a human's voice, Jackson noted that AI can provide culturally sensitive reminders as well as give an option to "people who don't like talking to people" and would rather interact with chat bots.

Additionally, AI could notice missed appointments or poor follow-up compliance and flag cases for intervention rather than let them go unseen.

Culturally Competent

Nnaji pointed out that humans working hand in hand with the technology can ensure that AI is responsive to cultural differences through:

(1) Training data. What data do we need to be representative of the population?

(2) Audits. Is there unintended bias?

(3) Transparency. Can the AI platform explain back to you what went into its decision? Can AI explain how it got to a wrong answer?

(4) Community. "You know the principle 'garbage in, garbage out," Nnaji explained. "Who in the community is providing input and making sure its as inclusive as possible?"

Jackson explained that insufficient data could result in "garbage out" results.

For example, AI might suggest lower reserves for female clients but fail to account for underreporting. Similarly, reports on Black and Hispanic workers often undervalue pain, leading to less aggressive treatment options.

"But an audit could address those broader options, not just historical values," Jackson said. "If you don't have broad enough data, it's going to get skewed."

The Future's Coming

Nnaji highlighted that human involvement in AI is part of eliminating bias.

"It's not engaging AI blindly," Nnaji said. "We're trying to eliminate bias, so we have to be proactive in accounting for it."

She also emphasized that it's a productivity tool.

"How do you lead the team?" Nnaji asked. "Embracing AI instead of creating an environment of fear is leadership."

And just like Netflix suggestions or directions in a maps app, AI and human experience can make each other better, according to Jackson, with each having a role.

"It's like having a super GPS and still needing someone to drive the car," Jackson said.

Share This Article: