Using AI To Help Both Sides, Adjusters And Injured Workers

Episode Notes

Transcript

Announcer:

Workers Comp Matters, the podcast dedicated to the laws, the landmark cases, and the people that make up the diverse world of workers compensation. Here are your hosts, Judd and Alan Pierce.

Judson Pierce:

Welcome everyone to another edition of Workers Comp Matters. My name is Judd Pierce in Lovely Salem, Massachusetts and we have a very interesting episode for you today with two guests who work in the sphere of artificial intelligence. An article that they co-wrote appeared in property casualty 360 last month and it was entitled AI Sparks New Era in Empathetic Workers’ Comp Claim Management. And because we recently had an episode with Dr. Claire Musselman on concerning that topic, I thought that this would be a nice bridge to the future, perhaps the present and see where we may be going as a workers’ comp model. So our guests, I’d like to introduce ’em briefly, are Tycho Speekenbrink head of ai and John Peters, co-founder and chief science Officer, they’re both a gain life, an organization that makes a claims communication platform that integrates behavioral economics with artificial intelligence. Gentlemen, thank you for being with us today.

John Peters:

Great to be with you Judd.

Judson Pierce:

Could you tell us a little bit about your professional journeys, both of you and how you came to focus on this intersection of AI and workers’ comp?

John Peters:

So my background is in healthcare. I’m a PhD biochemist actually, but I got into public health back in the 1990s with the epidemic of obesity and chronic diseases that have continued to rise since then and really trying to understand the behavior of people and how to get them to take better care of themselves. So that’s what got me into the sort of the behavioral side of things and when I started to look at other applications of what I had learned in that space, we got into this workers’ comp space and noticed that really there wasn’t a lot of attention being paid to the injured worker as opposed to the people who were running the business trying to manage the claims. It was not being looked at through the lens of the people who were injured. And so we thought there was probably some remedy we could apply there to help people get back to work sooner faster with less pain along the way.

Pain not only physical pain, but the emotional pain of having to go through the process and the tie to AI is really when you look at all of the wasted time and energy on really things in the process that don’t require a human, it doesn’t take a human to do all the pieces of paperwork and get approvals and do this and that and the next thing, what the human is needed for is to basically solve problems, talk to an individual, apply their people skills, use empathy, really understanding to help a person make it through the journey from an injury to getting back to work and taking all that routine stuff out of the way. So that’s how I got interested in ai. I’ll let Tycho talk about his entry.

Tycho Speekenbrink:

Yeah, so I commented from a little bit of a different angle. So I started as an actuary in big insurance carrier then did sort of the data scientist with all of machine learning and then sort of naturally evolved towards AI and then I think specifically using AI to look more at the empathy component. For me it’s just a very interesting use case where not just with the operational part, which I think is what AI is used for also to be able to leverage in another way. I think there’s a lot of opportunity there. So that has made me very excited to work in that specific space as well.

Judson Pierce:

Well, thank you both. Your article highlights the transformative potential of AI in workers’ comp. What are some of the key research projects or studies you’ve been involved with or have read that have explored this area?

John Peters:

Well, I think there’s a number of angles to this, which is really if you look at the process of an injured worker getting back to work and you try to understand what are the factors that influence that outcome, the speed of the outcome, then you can understand that there are so many of the pieces along the way where there’s dead time, whether it be waiting to get an authorization for treatment or whether it be just the process that many insurers still use, which is snail mail to send documents to be signed and so forth and so on. So much of that can be automated and you can have electronic signatures and you can have documents being sent really instantaneously for somebody to approve and get the process going. But I think the thing that really struck me was if you look at the social determinants of health as Tycho alluded to earlier, talking about AI is probably something like 70 or 80% of the changeable factors along a claim.

I’m not talking about the medical side of things, I’m talking about the factors that influence nonmedical outcome pieces. 80% of those could probably be related to the social determinants of health, which I think of as context. It’s like where you live, what are the circumstances of your living, what are the resources you have at your disposal? All of those things impact the outcome and impacts your ability to do what you need to do to take care of yourself and get better. So those are the things that we can look at as the entree. There’ve been massive studies done to look at the social determinants of health as indicators of outcomes and so if you can remove some of those factors, intervene early when you find that somebody needs transportation to get to the doctor or to the physical, something like that, if you can find out that in the moment that can be transformative.

Judson Pierce:

Did you look at it from the standpoint of the claimant representing his or herself or the workers’ compensation injured worker having an attorney doing a lot of that for them? Or did you sort of look at it from both angles?

Tycho Speekenbrink:

So I think we’ve been focused mainly on the injured worker perspective. If you think about the communication that the injured worker has with the claim adjuster, there’s a lot there. So it happens in different points in time. So obviously at the beginning of the claim, at the first notice of loss, there’s some interaction going on there, but even during a claim, especially within workers’ compensation, there’s a lot of interaction going on continuously. So that specific information, looking at injured worker perspective and also making sure that they are being taken care of for their specific needs. That’s where we’ve been really focusing on

Judson Pierce:

Absolutely because as an attorney representing injured workers for 25 years, what I’ve noticed is the amount of time out of my day in trying to bridge those dead spots as you call ’em, John, trying to get that MRI approved that signature sent over, I’m the conduit really once the claimant represented the adjusters and the insurer cannot speak with the injured worker any longer, they have to go through counsel. So I’m experiencing perhaps what the injured worker would experience if he or she were not represented and goodness my day, the claims adjuster’s day would just be so much more productive if they had these extra technological resources. And what are some of those AI resources as laypeople? Not really in the industry, just trying to get up to speed with it. We’re familiar with the words chat, DBT and some of the Microsoft and Google, they have their own. Could you tell us a little bit about the actual instruments to gain access to this technology?

Tycho Speekenbrink:

So there are a lot of different use cases that we can actually use AI for. A lot of it is in the operational part as well. So the nice thing about these l lms, so these large language models that are underlying these chat bots that you mentioned before is that are able to take a right range of information and then they’re able to translate that into something concrete. So for example, when it comes to a triage process, so once a claim comes in and you want to decide who to assign to that specific claim, that’s something that can be done by looking at all the information available. So you use AI to interpret that information, look at past cases to determine which type of justice would be assigned to that specific case, and that’s something that you can optimize. So that’s one example that we can think of. Another one using AI as well, which is having these LMS actually analyze a conversation. So in a similar, we build it in into the current workflow and whilst a conversation is going on, this AI is being able to pick up on this information and then actually give back some hopefully insightful comments about how the conversation went and what the adjuster potentially could do differently. So that’s a little bit how these tools can be integrated in different parts of the workflow.

Judson Pierce:

Interesting. So if an insurance supervisor said, oh, we have Bonnie here, she’s really good at picking up on client cues in terms of their vocal and their, but she could use a little bit of help in terms of what they might need going forward, she could benefit of that and let’s assign her to this claimant. Okay, exactly.

Tycho Speekenbrink:

It Comes both ways

Judson Pierce:

Yeah. Mike might not have that empathetic listening skill and might need some more of the empathy that chat GPT could pick up on a phone call between Mike and the claimant and the review of that conversation could be some learning and what to do next time for Mike.

Tycho Speekenbrink:

Exactly. So the nice thing of these ais is, and the way that they’re progressing is that they’re able to interpret things in a lot of different ways. So they’re able to use a lot of diverse information and then in the example that you mentioned, if you actually look at the injured worker at its characteristics and you can judge that from the information they provide, you can also connect a specific adjuster that would do well with that specific injured worker, but then also once the injured worker is actually talking with the adjuster, you can help the adjuster analyzing again the same type of communication to be able to do things better or in a different way, or maybe you should have said it in this way. Those are all use cases that are definitely possible.

John Peters:

One of the ways of thinking of AI is not artificial intelligence but assistive intelligence. So it’s really assisting the adjuster as opposed to taking them out of the equation. I think one of the other things that we often forget about is you can use some dumb AI I call it, to do some remarkable things. If you think about the sophistication of chatbots these days, what people crave when they’re looking for information is they want the information. Now if it’s two in the morning and they can’t sleep and they have a question about their claim, they want to find out something about the workers’ comp process. They can’t call their adjuster two in the morning, but they could call an AI and find out about what’s going on with this claim or they can find out about an answer to a specific question that might be, can I change my doctor? I don’t like my doctor, I want to change. Is it okay to change? And that may depend on what state they’re in and what jurisdiction they’re in, that sort of thing. And so AI can be there 24 7, 365 where the adjuster can’t.

Judson Pierce:

That’s a good point. Why don’t we take this moment for a break from one of our sponsors and when we come back we’ll talk with John and Tycho about trusting ai. Is it possible to, and whether there are some potential risks involved in relying heavily or somewhat on ai? We’ll be right back on Workers Comp Matters and we’re back. We were talking about the benefits to insurance adjusters in having the use of not artificial intelligence, but assisted intelligence. I love that phrase, John. I was thinking about the grand old question, can we trust it? Is it doable and usable for us to rely on? Are we at that point where we feel comfortable in trusting that that answer that the claimant asks, can I change my doctor? Is he or she going to get the right answer?

Tycho Speekenbrink:

So we’re getting there. So if you look at the AI as they are now, we definitely still have problems like hallucinations, which is that the model tends to make up something that doesn’t exist, but you actually see this being better as we go along. I think what’s important now but also in the future is that whenever you implement something, you want to have some safeguards around it. So safeguards can mean a specific engineering application, a human in the loop approach depending on how sensitive the information is. So that means that you would actually have to have someone else look at that information. So there are a lot of things that you can build around these AI systems to actually make sure that the security and the trustability is good enough so that it doesn’t hurt the trust rate, for example, your injured worker that you’re communicating with.

John Peters:

Yeah, I think a good example for the application of AI and the human in the loop would be suggested responses to questions that might come from an injured worker and it could save time for the adjuster if the AI would suggest a response based on the nature of the content, but that wouldn’t just be said automatically. The adjuster would see the response that the AI had constructed and say, that’s perfect send or No, that’s not perfect. I’ll change it. And so there’s a safeguard there,

Judson Pierce:

Will it steal these people’s jobs? That’s one of the questions that’s on my mind every time I hear about ai, whether it’s lawyer jobs, adjuster jobs, is it coming or can we prepare ourselves to just adjust our jobs?

John Peters:

I don’t think it’s going to take anybody’s job. I think it depends on how you use it though. In the article we pointed out that if you don’t use it correctly, you can say, wow, we’re saving all this time so we can pile more adjusters pile on top of what they already have as claimants. We can give them many, many more injured workers to take care of. Well, that’s not how you use the additional time you gain by using ai, you would want to use that additional time by having the adjusters spend more individual time with the injured parties, making sure they’re solving their problems and providing that empathetic advocacy approach. That’s just an example of how it can go wrong, but I think if you think of it as assistive intelligence as opposed to artificial such that you’re placing somebody, then that gives you a much better sense of what it’s designed, what it’s optimally designed to do to make you a super adjuster. I can take an average adjuster and make you a super adjuster because now I can offload all these things that you don’t need to use your brain for and we can save you for the stuff that really matters, the human interaction.

Judson Pierce:

What I’m thinking on that is that what if the super adjuster does what she usually does in an eight hour day, but in a two hour day, and so the employer then says, well, why are we paying you eight hours? We’re just going to pay you two hours a day. So isn’t that essentially losing one’s income, source of income?

Tycho Speekenbrink:

I think the way that we look at it is if you look at the workers’ compensation industry in terms of the work that the adjuster does, there’s such an amount of pressure to be able to deliver things. There’s a bunch of administrative things that need to be done. There’s a lot of complexity and then the things that actually matter, which is taking care of the injured worker and making sure that their needs are being served, which is something that really requires this human touch as well. We don’t think that that’s something that can be replaced. The efficiency parts hopefully helps the adjusters to be able to do their work in a better way. So given the fact that if you look at the workers’ comp industry, there’s already assured us of adjusters as well. I think there’s an opportunity to just make the work more interesting, more efficient, more rewarding, and in the end, the way that we think about it is that adjusters will benefit from this technology as well.

Judson Pierce:

What was one interesting thing you learned when you were writing this article that you didn’t already know?

John Peters:

The thing that keeps coming up is really understanding what the possibilities are. If this were implemented across the industry to solve some of the industry’s biggest problems, I think Tycho brought it up, which is the talent shortage. What kid coming out of college wants to go into a job where most of the correspondence is done by snail mail and email is the new thing on the block, and so things are still kind of in the last century and you have paper files everywhere, and so they’d want a job that’s basically kind of encouraging them to use their brain to solve problems and to help people. Gen Z wants something that’s meaningful and purposeful and they grew up in a technology age, so they’re used to having all this stuff and if they entered a job where it wasn’t available, they’d think, what are you guys doing? So I think that can really help attract talent and the industry really needs talent.

Judson Pierce:

What are some risks that you identified in the article or just in this conversation, what are some risks about relying on it too heavily if there are any?

Tycho Speekenbrink:

I mean, obviously if you think about the risk of relying too heavily on those, I mean we touched on it a little bit as well, but these systems are not perfect yet. They can make mistakes. So if you use it in the wrong way, in the wrong situation, there is a risk that something’s being communicated, which you don’t want to have or some wrong information is being given. So that comes back to the point of safeguards as well. And then also more from a technical perspective, you have a lot of security things that need to be looked at, so it’s definitely an effort to be able to mitigate some of the risks that are out there. So it is a learning process as well to be able to use these new technologies.

John Peters:

I suppose you could get lazy. It means you’re not staying on top of everything that comes through your system or your lens, so you just can’t rely on it as I can turn it off and put my feet up. So that’s always a risk.

Judson Pierce:

Well, why don’t our guests put their feet up for a little bit? We’re going to take a quick moment for another break from one of our sponsors and we’ll be right back to close out this session on AI and workers’ comp. We’ll be right back and we’re back. I wanted to bring up a hypothetical that might come up as attorneys with an injured worker who is perhaps out longer than it was expected medically, professionally, the job is interested in having her come back, they miss her, the job has been open all this time and she’s struggling to get back and the claims adjusters pulling her hair out like why isn’t this person seeing any real recovery yet? Maybe the surgery didn’t take and there might be another surgery down the pike. There could be social determinants that are prolonging the recovery. How can good AI programming and use of that technology help in this situation?

Tycho Speekenbrink:

I think in this specific situation, we would probably want to look at it a little bit earlier. So even in this example, in this hypothetical example, so suppose that someone gets injured, the first scenario would be to look at the witness statement and to try to determine what I mentioned before, who to assign as a specific adjuster. So assume that we have an adjuster assigned that’s very empathetic and is able to work well with the specific claimant. So that would be one thing where we can very concretely make a difference. Then the other point as well is that once this communication between the claim adjuster and the injured worker is going on, depending on specific type of statements that he or she makes, you’re able to proactively be able to help the injured worker to make sure that he or she feels hurt. If there’s something which for example limits them in terms of not having transportation, this is something that we can provide as well. So all these key points, which comes back to the social determinant’s health where we’re able to proactively actually help this specific injured worker as well. I think if we then go on being able to help them give information practically, making sure that they feel heard. What we see is because these social determinants of health are such a big part of how someone feels and how someone acts and how someone recovers, that’s our ways that we can really help and support them.

John Peters:

Yeah, I think is if this were a manufacturing plant, you’d want to do a root cause analysis. You’d want to find out, well, what are the root causes of this person’s being unable to return to work on the schedule that you would normally expect? And you would start with things like what’s their zip code? You can look up in public databases, the area deprivation index, which is indication of based on your zip code, where you live, what are the resources available to you? Are there things that are holding you back? Are you 10 miles from the nearest grocery store? Are you near a bus line where you can get public transportation? All these kinds of things that can affect your daily living can be part of the reason why the person is not able to come back to work. Examining what Tycho said, which is the conversations that have been had between the adjuster and the injured party can figure out is this person suffering from maybe a mental health issue, not necessarily a disease, but they have a problem, they have anxiety, they have depression, there’s something that’s going on there that’s mental that they need help with that can be assigned a behavioral resource.

Is it a financial problem that they’re struggling with? And those are all things that can be gleaned from the ai. Looking at the conversations and being able to summarize vast amounts of information over this two year period of time to say it looks like the arrow is pointing in this direction in terms of what might be the underlying issue for this individual. And then applying, we use something called ecological momentary assessment, which is checking in with the injured party at random times just to find out how they’re doing. It could be just a simple text that comes in the middle of the day that says, Hey, we’re thinking about you. How are you doing? It can give you additional information in the moment about what’s happening to that person’s life, which might be something that’s holding them back from following the traditional recovery pattern. Those are the kinds of things I think that can help reveal the situation and lead to a better outcome.

Judson Pierce:

When we had Dr. Musselman on an episode or so ago, she said something which sort of blew my mind, blew Alan Pierce’s mind. She said, well adjusters, they’ll write a get well soon card and they’ll send it out to the claimant and that does wonders or a text or something. And that is something that I don’t think I’ve seen. I mean as attorneys, we try to do that in our office to let our clients know that we’re thinking of them and we’re hoping that they feel better from surgery. But to get it from the insurance adjuster, the insurance industry showing that level of care and compassion could go and wonders for recovery and health

John Peters:

And the AI can construct the first version of that, show it to the adjuster, and the adjuster can add their own personal touch because they’ve spoken to this person numerous times so they can add something about maybe their family or whatever, what they know about the injured party. So it can really personalize it, but it’s automated in terms of it being constructed. And it kind of reminds you, oh, I can send this reminder, I can send this getwell card to Sally.

Judson Pierce:

There’s a significant problem that I think we haven’t addressed yet with ai. Doesn’t it as an industry need a whole lot more money to improve and who’s going to get behind that and actually fund those improvements? Or is this as good as it gets?

John Peters:

You’re talking about within the insurance industry or are you talking about AI in general?

Judson Pierce:

The AI in general, the source of the technology that the insurance industry would use, doesn’t it have to have some sort of financial impetus to gain, to gain traction and get even better?

Tycho Speekenbrink:

There’s a lot of money being invested currently in these AI models. So people are talking about trillions of money to invest in development.

Judson Pierce:

So it is being invested right now,

John Peters:

But they’re largely consumer platforms. These are consumer facing platforms, which is where the volume is and where the money is. And then I think once the technology is available, I think what the industry has to do, each individual industry is to figure out how to best tailor what’s capable to their own needs. And that takes internal investment. So nobody’s going to come and figure out how to make the workers’ comp version of AI the best it can be from outside the workers’ comp industry. I think that’s where the insurers within the industry have to put in their hat and say, we’ll invest in this too.

Judson Pierce:

Is it happening right now, our insurers using this technology in the way you guys write about in the article, or is this just a future hope?

Tycho Speekenbrink:

You’re starting to see the first use cases that use these type technologies. So the article that we wrote is specifically focused on this new version of GPT GPT-4 oh, which is able to analyze not just text, but also voice. So obviously that’s very recently, so you see that there’s a lag between once these models comes out and then having any type of industry, but especially insurance industry to actually adjust in being able to use it because a lot of additional work needs to be done around this as well. So you see the first inklings, but there’s definitely a lot that needs to be done to get to a scenario in the way that we describe it in the article,

Judson Pierce:

I found that in our office we’re still experimenting with just how to make this letter better or how to make this brief better. And once we put it in, it comes back and it just doesn’t sound like our voice anymore. So I’m wondering if that’s the same in the insurance field in terms of it not seeming authentically human. And that might be one of my fears is that it hasn’t grown to that or maybe we don’t need it to, right. Maybe it’s just an assist for us, but we still have to put our human markers on it.

Tycho Speekenbrink:

I think that’s for sure. You do need to put your human marker on it, and it’s a little bit what you talked about earlier, I don’t think that AI will fully replace humans. I also think that’s something to be said about learning how to use these types of systems. So that means trying things out, getting some feeling for it. It’s evolving as well. So I think that’s something that I would recommend everyone to do.

Judson Pierce:

Is there a good course you can recommend or a good YouTube or a good, other than just reading your article, which is fabulous, is there something that people, listeners can go to right away after they hear this episode and say, okay, I want to learn how to maybe use this in my practice. What would be a good first place for them to start?

Tycho Speekenbrink:

So I think the first place to start is using the chatbot tools that are out there. So whether it’s ChatGPT whether it’s Gemini from Google, whether it’s Cloud from Anthropo, being able to use these tools, upload some things, say I want to do this, and then getting some feedback. I think that’s a first start to actually get your feedback. If you’re thinking about, well, I actually want to see if I can implement something in a little bit more of a larger scale, and there’s a lot of information out there. I mean, obviously YouTube is good. I think Coursera, which has a lot of AI related content as well, for people that really want to delve into it, is definitely something that I would also recommend.

Judson Pierce:

Great. Well, anything you guys want to add just to finalize this show? I think we hit on a lot of important topics today.

John Peters:

Well, isn’t it true Tycho, if Judd were to write, I dunno, a million briefs that you could train their own version of AI to basically mimic the tone of those briefs. And so back to your point about, it just doesn’t sound like us. You can train an AI to sound like you if you give it enough information that’s based on things you’ve written. Not that I suggest you go and write a million briefs, but

Judson Pierce:

No, I’m far from that number. Maybe in another 25 years I’ll get closer, but it sure sounds like you guys have been doing what you’ve been doing for a long time. You’ve given us a lot of interesting things to think about, and I invite everyone to go read their article that was in Property Casualty 360. It was published on June 26th, and the title’s name again is AI Sparks New Era in Empathetic Workers’ Comp Claims Management. I’d like to thank our special guests from Japan, Tycho Speekenbrink, and from lovely Ohio, John Peters. They’re both at Gain Life, and if our audience wants to reach out to you and ask you any questions, is there a good way to reach you all?

John Peters:

Well, John, at gain life.com.

Judson Pierce:

That’s easy. JOH [email protected], folks? That’s

John Peters:

Right.

Tycho Speekenbrink:

Yeah. And Tycho at gain life.com. You’ll be able to reach me as well.

Judson Pierce:

Great. Well, from all of us here at Legal Talk Network and Workers Comp Matters, thank you for enjoying another episode. Go out and make it a day that matters. Take care.

Continue Reading