The past year has been transformative for the world of AI. The developments around large language models (LLMs) and generative AI have put the spotlight on a set of challenges that expert.ai has been focused on for more than three decades: AI applied to language data.
Language data is ubiquitous in the enterprise, and it’s at the root of many critical business processes. The insurance industry, in particular, faces the challenge of managing an ever-growing volume of information and relies on its efficient utilization and analysis for so many processes.
With the arrival of generative AI, businesses recognize the need to stay competitive with early movers while balancing the risks for privacy and uncertainty around tangible outcomes and costs. But how to navigate the hype to understand how to move forward?
Hype aside, there is real value to be gained from AI right now. While having a LLM or using generative AI are not prerequisites to success, there are ways to leverage any AI technology safely and with intention.
This is the focus of our recent episode on the Insurance Unplugged podcast. In the first episode of the new season, expert.ai CEO Walt Mayo joined host Lisa Wardlaw to talk about this transformative age of AI and how insurers are dealing with new capabilities, how to break through the hype and how to move forward.
Below is an excerpt of their conversation.
The Impact of LLMs on Insurance
Lisa Wardlaw: How have LLMs changed the playing field and the AI landscape, especially for industries like insurance?
Walt Mayo: I would start by maybe shifting your framing from the language not keeping up with the technology to, historically, the technology not keeping up with the language. Insurance companies are very attentive, very, very purposeful around how they use language to frame a risk that they are covering to ensure that there is a clear understanding of mutual obligation.
However, the complexity of language is very difficult for technology. It needs software that can handle the ambiguity of language and the enormous compute power that this requires. These two things came together in LLMs. When OpenAI launched ChatGPT in November of 2022, they did two things. They devoted an enormous amount of compute power to it, and the dataset they used was by and large the English language content of the internet. So, when you think it, this is your training set and then the compute power that was required brought forth this extraordinary capability.
The more breathtaking thing is how rapidly that capability is commoditizing. It has come down unbelievably quickly, and the amount of money that has poured into driving down the cost of this LLM capability is enormous. Amazon, Google, Microsoft, venture capital has put tens of billions of dollars into these capabilities in the last 18 months to make it faster, safer, smarter, etc.
Now, what that affords is a really powerful new tool that can be used in combination with a thoughtful approach to solve a problem.
Deployment Thinking
LW: How are companies actually deploying this capability? Are they just adding another widget?
WM: I see insurance executives, leaders and the C-suite in general thinking about this across several layers.
At the highest level, they are thinking about the fundamental value they provide of pricing risk and doing it in a way that serves the reality of the incredibly varied conditions we face.
I think there’s a second level, which is how do you think about the foundational processes that are common and your desire to serve your customers most effectively?
I think where most folks start is at this third level, which is, ‘I have a process that I want to do faster, better, cheaper, etc.’
The interesting part is really moving up from that into a meta strategy where you’re saying, ‘look, we have an underlying data form that now appears tractable with technology, it’s language. Where is it? And how can we apply this technology in a way that enables us to serve our customers, to gain competitive advantage, to come up with creative new models for our offerings, etc.’
I don’t believe that’s happening yet. I think most folks are still in the thinking of, ‘we have an existing process, it’s language dense and we want to do it faster, smarter, better,’ and that’s probably fair for now.
LW: How are you seeing insurance companies address this?
WM: Senior leaders in the insurance domain are being thoughtful. I’m actually pretty impressed with the approach. If you think about it, insurance companies actually had some of the original data scientists, they were called actuaries: Someone who is taking a reasonable amount of structured data and trying to make mathematically sound predictions that enable the company to drive value. They were called actuaries back then, but there were data scientists fundamentally.
The way I would be able to distinguish folks who have a structured and thoughtful approach is if they have categorized AI capabilities in a reasonably thoughtful way.
If you think about it, there are three broad AI capabilities:
- Structured data is the ability to use very, very sophisticated mathematics to predict the likelihood of events and the relevance of those events. We’re talking about the structured data, how you collect it, and how you act on it.
- The ability to use images. Think about applications for auto insurance where you can take a picture of an automobile and the computer vision can render a pretty accurate representation of what occurred in a way that used to require a human being to actually look at the vehicle.
- The third bucket would be around language.
Just having those three capabilities is a pretty good indicator that a company is being very thoughtful around the capabilities that are emerging now.
Get Smart About AI
LW: What would be the one call to action for every listener to take away from this episode?
WM: Get after it. And by that I mean get smart enough to be smart enough to understand what’s happening right now with a really exciting technology development.
It’s not beyond the reach of any senior executive to have a serious, thoughtful conversation and ask really good questions. You’re going to have to invest some time. This is going to be the highest-value ROI activity that you can achieve in 2024.
Get to the point where you can have a very serious conversation that is grounded in the reality of where the technology is now, and get to the point where you’re just a little bit uncomfortable, maybe even saying some things where you’re not a hundred percent sure, but you’re going to go out on a limb.
It’s like speaking a new language. When you travel to a new country and you only use the phrases that you know really well, you’re not going to stretch yourself very much. So try some phrases that you’re a little uncertain about.
In the AI world, for the business leadership, push yourself and don’t fuss if you got something wrong.
Listen to the full episode with Walt Mayo of expert.ai on Insurance Unplugged with Lisa Wardlaw.
