From the editor:
Since the beginning of 2023, generative AI has been the topic of conversation across industries and the general public. It has captured the imagination of both futurists and pragmatists who extoll its benefits and raise concerns about its risks. But it has also become clear that despite the spotlight on generative AI, there are still many questions about what generative AI is, what it is not, and what it means for businesses trying to figure out how this innovation fits into their technology plans. We sat down with Eric Sibony, Shift’s Chief Scientist and Chief Product Officer, to get his perspective on this exciting technology and what it may mean for the insurance industry.
Shift: How would you define generative AI for someone who is not really all that familiar with the space and what's happening there?
Eric Sibony: At its core, generative AI is a new type of system, or more specifically new types of models, that are designed specifically to generate a distinct result based on the prompt given. And since I realize that I’ve effectively provided a circular definition, let’s look more closely at how generative AI actually works. Understanding that generative AI can be used to produce a variety of results, including art and music, for our purposes here I’m going to focus on text.
From our perspective, what’s so intriguing about generative AI is how it can tackle texts, and in particular, natural language. As such, generative AI can process virtually any kind of text that it can access. It's able to analyze news stories or articles, business or legal documents, or financial summaries or reports. Truly, all of the varieties of texts generative AI can analyze are simply too exhaustive to list here. And perhaps more important than the kinds of texts generative AI can ingest is the amount. The scientific and technology communities involved in generative AI have found ways for the models to be able to ingest huge quantities, and I mean really huge, quantities of data. We're easily talking about billions of data points.
What may be most interesting about generative AI is that a single model - for example the GPT model from OpenAI or the LLaMA model from META - is able to learn from these massive amounts of data. And while automated learning, the fundamental principle behind generative AI, has been in use related to AI for decades now, it’s never been applied at this kind of scale. These models have learned and been trained on billions and billions of data sources.
What makes this so intriguing is that upon deployment you have a model that has already been trained and that already knows how to do a lot of things. That’s its initial starting point, which is radically different from more traditional approaches to AI. With a standard AI machine learning based approach, you have a problem that you want to solve. You gather a data set and then set out to predict something about the data set you’ve brought together. But first, you need to manually label your data. So you end up with a labeled data set, a trend, and an algorithm related to the particular problem you wanted to solve. And when you have another problem to tackle, you repeat the process again.
With generative AI, the model is able to accomplish many, many tasks. And the real beauty of it is these aren’t tasks the model has specifically trained to do. We're talking millions of tasks that they haven't been trained in. These new large language models (LLMs) are fully capable of answering any kind of question that we ask them or solving any kind of problem that we have. You can really ask it anything!
Shift: What are some of the benefits you're seeing that generative AI can offer the insurance industry?
Eric Sibony: Before getting into the specifics of insurance, I think there's a good example that will resonate across industries. Generative AI is already proving to be quite adept at producing first drafts of emails and other correspondence as well as other written communications. I think quite soon we’ll see that ability applied to presentations and other forms of content. If Generative AI can eliminate the hardest part of communicating for some people - getting started when facing a blank screen - employees will be freed to spend their time on activities that generate far more value for the company. This is one of the real promises of generative AI.
But I believe for insurance companies, content generation is only the beginning. Every day, insurance professionals are tasked with combing through a lot of different documents to extract the specific pieces of information they need to do their jobs. These documents may relate to the underwriting process, the claims process, the renewal process, or any of the numerous processes critical to operating an insurance business.
Now, imagine if a claims handler, for example, could simply ask for a targeted summary of all the documents related to a claim they were working on. Or better yet, what if those relevant summaries were waiting for them when they sat down to their computer to start their day. Instead of spending hours reading through claims intake forms, adjuster notes, police reports, weather reports, or other associated documents, they can get right to the truly important work, deciding how to proceed with the claim in front of them. For insurers, generative AI holds the very real promise of providing the industry’s employees with the information they need, when they need it, to make the best decisions possible.
Eric Sibony: As I’ve previously mentioned, generative AI models can ingest amazing amounts of data and produce compelling results. At the same time, those results may lack context or even contain errors depending on the data source accessed. We’ve certainly seen this with some of the more consumer-friendly generative AI solutions which use the Internet to produce results.
For insurers exploring how generative AI could benefit their organization, I’d say the first big challenge centers on data. Insurance is a nuanced business. It has its own language. It has its own rules and regulations. As such, insurers must be diligent about the data its generative AI solution uses to produce results. Just because generative AI models can ingest billions of data sources it doesn’t mean it should.
Take the time to understand what you want to achieve from your investment in generative AI and which processes you want it to support. For example, a policyholder-facing chatbot requires a different set of data than an applicant-facing chatbot. You’ll avoid a lot of potential headaches if you effectively match your data sets to what you’re trying to accomplish.
Shift: How is Shift Technology helping the insurance industry adopt Generative AI?
Eric Sibony: Shift has been working with LLMs since 2020, and we’re on track to have the Azure OpenAI Service live in our solutions by the end of this quarter, so we’re quite familiar with the promise and the potential challenges associated with this AI approach. As we’ve discussed, the real power of generative AI comes from its ability to analyze natural language and produce a usable result, essentially right out of the box. However, as generative AI lacks context, it doesn’t necessarily know if the result it produces is “right” or “wrong” for a given prompt.
When producing the first draft of an email or memo, this isn’t that critical. It’s a different story if you’re talking about helping an insurance professional make a decision about what to do with a claim or a policy application. And that’s where Shift can help.
First off, we have a proven track record of applying artificial intelligence to the specific challenges associated with decision making across the policy and claims lifecycles. We’re experts in creating cleaned and mapped insurance data sets ready for use by generative AI models. Further, we can provide the insurance industry context that gives confidence in the results produced by generative AI. But perhaps what is more important is that not only are we experts at creating exceptional insurance data sets, but also in creating data sets specific to our customers’ unique needs. Not every insurer is the same, so it doesn’t make sense to offer a one-size-fits-all approach to their data.
For more information about how Shift can help you adopt generative AI to meet the unique challenges facing the insurance industry contact us today