Revolutionizing Benefits with AI

Technology is the future of benefits. Check out this recap from one of our top sessions at Vision 2023 where Rae Shanahan, Chief Strategy Officer, and Sony Sung-Chu, Head of Science and Innovation, discuss the future of AI in benefits technology and HR efficiencies.

Editor’s note: This is an abbreviated transcript from our live Vision session: AI-Driven Insights. You can also listen to the recording to hear Rae and Sony discuss the value of AI in benefits and HR technology.

Key takeaways from this session:

• The integration of advanced generative AI technologies with Sofia, our AI benefits assistant, will bring about significant advancements in the Ben Admin space overall, and enhance Sofia’s overall efficiency.
• With the addition of Microsoft’s HIPAA-compliant service for ChatGPT, Sofia’s processing has improved by better identifying what members need to know, how best to respond, and how members can utilize their benefits in ways that are cost-effective for both them and their employers.
• AI “is not magic” and will not replace humans. Still nervous about AI? Simply try it yourself!

Rae Shanahan

Let’s ask Bing for a quick joke to get us started: what do you call a person who understands their employee benefits, like healthcare? …A unicorn! Because they are so rare and magical.

Alright. Sony, please introduce yourself.

Sony Sung-Chu

I’m Sony, the co-creator of Sofia and I am head of Science and Innovation Teams here at Businessolver, I’m happy to be here today to talk about what we’re doing with ChatGPT and Sofia.

Rae

Let’s level set to provide some context around Sofia. Can you talk about her journey over the last seven years?

Sony

Sofia was the result of just two people on a team with a hackathon idea. She’s a machine-learning, artificial intelligence, personal benefits assistant. And as we saw early success, we started growing the team to support the different areas that we wanted to develop, like natural language processing, and information search and retrieval, up to the point where today we have a sizable team that’s supporting all those different programs for Sofia.

Rae

So, Sony, what are you most proud of in regard to Sofia?

Sony

I think what’s really been impressive is her growth over time. Over the millions of questions she’s answered, every new thing that she learns is really us wanting to fulfill someone’s need.

Rae

Let’s jump into the future of AI, what’s going on and what’s exciting, and then we’ll talk about the alternative for that—what we could be concerned about.

Sony

I was super excited when all these headlines came out. 2023 has really been an unprecedented year for AI Technologies in the news. So much happened in Q1, including Open AI, Microsoft announced they would use GPT-4 to power Bing chat, Google launched Bard, Meta launched LLaMA, Baidu launched ERNIE, and so on. All of these are examples of generative AI, which are basically tools that are really good at generating content, like responding to a question, writing articles, or writing code. And they get especially more powerful when paired with existing technology.

So for example, in the case of GPT-3.5 Turbo, it has ~175 billion parameters, but GPT-4 has ~170 trillion parameters, and the takeaway here is that the more parameters you have, the better they are at identifying complex patterns in the data that they’re exposed to.

And as we recently announced, we are in the midst of evolving Sofia using these technologies, which we believe are going to be incredibly transformative in the Ben Admin space.

Rae

Sony, I know that there’s some people that have raised some concerns about AI. So how are you factoring that in as we make plans around Sofia’s continued learning?

Sony

The best way to think about the dangers of AI is to look at it from the perspective of bounded systems.  And what I mean by that is, if you have an AI system that isn’t bound by anything, meaning you let it go and learn without guidance, it may not learn the values that you want to instill in it. So, it’s like if you have a child with free, open access to the Internet, they might pick up things that maybe aren’t safe or appropriate for them.

The second thing to consider is that if the bound itself that you’re providing is misguided or somehow nefarious, going back to the analogy of the child, it’s similar to if the child met someone who had a compromised moral compass. There may be unintended consequences. It’s really our responsibility to make sure that we ourselves are guiding AI towards serving our members the best way it can be.

Rae

I know that we already factor this in, but I’d also like to hear you speak briefly about how we make sure that Sofia isn’t bringing bias into her conversations?

Sony

We have several ways to ensure that there isn’t a state of bias within Sofia, to make sure that when she’s responding to people, the responses are fair, and they’re not biased in any way. We continue to teach Sofia that diversity is a part of her code.

Rae

We have some new opportunities to use ChatGPT to make Sofia even better, talk to us about that.

Sony

The value of ChatGPT rises exponentially when you pair it with applications that can do other things. For example, if you asked ChatGPT “give me a recipe for Thanksgiving,” not only does it give you the recipe, but when integrated with a grocery shopping app, it can add the ingredients to your preferred grocery store cart and notify you when to pick them up at the store, and that’s much more impactful.

When we apply this to our industry, we know that people struggle to understand their benefits. That being said, there are some small steps we can take for Sofia to improve her own understanding, to help our members.

First, we can help Sofia receive questions, and summarize the answers if someone were to ask a more complex question, like a question with 2 parts, or if there are grammatical errors, she can glean the intent and get an idea of what they really meant.

Rae

So how are we going to bring Sofia and Open AI together? How will it be Better Together?

Sony

We use Microsoft’s HIPAA-compliant service for ChatGPT to analyze the question that the member is asking, like relevant facts that could help Sofia answer the question. That information gives context, so that she knows that the information is relevant to something within Benefitsolver, and she can craft the answer to give back to the member.

For cases where Sofia herself may not know the answer, she can use ChatGPT’s power cognitive search service to allow Sofia to read documents that we have in the Reference Center and then summarize them for the user.

Rae

We have a poll question “are you concerned with employees using generative AI tools in the workplace?” with 44% of respondents saying “yes,” 56% say “no,”. I can understand that there’s hesitancy, but again, I keep coming back to what some of the other conversations today have been about, which is we’ve got to meet people where they want to be met.

So, let’s go a little bit deeper here. How is this going to change as we go forward?

Sony

Well, you’ll never see a response coming directly from ChatGPT that hasn’t been filtered through Sofia’s own knowledge base. And that’s what’s super important—ChatGPT alone would not be able to do this because it wouldn’t have Sofia’s knowledge.

Rae

We launched our executive summary of our Empathy report. So how are we infusing empathy into Sofia?

Sony

We believe that if Sofia has the ability to alleviate someone’s frustration, provide comfort, celebrate successes, we want to take that opportunity to do so. Sofia does this today to some extent. She shows empathy in certain cases where the member may seem to be in distress and, we connect a human, or live rep, to help the person.

We will be bringing forth the capability to understand this sentiment in almost every interaction that she has, and she can respond appropriately and convey that she understands and appreciates the feedback and will still offer additional resources if needed.

Rae

As technology evolves, what can we expect in the future?

Sony

I think that every virtual assistant will evolve beyond the chat window and become a primary interface between humans and applications using natural language.

And if you couple that with the fact that information is growing exponentially, the size of like the digital universe today is like, zettabytes, which is enormous.

Rae

Last question, how do you quell anxious thoughts surrounding AI?

Sony

I think through education and through experience. I don’t believe the headlines don’t pay attention to the news. Just go and try it, and I think the one thing that you’ll find out is that it’s not magic. It’s the culmination of the sheer amount of information and processing power that we have today that’s enabled this, and it’s really amazing technology.