November 3, 2023 • 38 mins
Injury to Insight: The Future of AI In Workers Compensation with Michael Milstein
This week, Ryan Hathcock of Chartwell Law, www.chartwelllaw.com, and Steve Armstrong of Armstrong and Peake PLLC, www.armstrongpeake.com, are joined by Michael Milstein of Downey & Lenkov to discuss the growth of Artificial Intelligence in the handling of workers’ compensation claims. Michael takes a dive into the use of AI by insurers and claims professionals and how the data obtained through AI can assist attorneys in better predicting the outcome of claims including the claim value and potential fraud. Michael discusses the uses and the state of AI in workers’ compensation currently, with an eye toward the future of AI for insurers and attorneys. There are also serious risks that must be considered when personal information is involved in data collection and when attorneys choose to rely on AI without checking for its accuracy. You can find Michael on LinkedIn at https://www.linkedin.com/in/michael-milstein-741a9a58/ at his firm’s website www.dl-firm.com.
To learn more about DRI and the Workers' Compensation Committee vist www.DRI.org.
#DRILawyer
#DRICommunity
This week, Ryan Hathcock of Chartwell Law, www.chartwelllaw.com, and Steve Armstrong of Armstrong and Peake PLLC, www.armstrongpeake.com, are joined by Michael Milstein of Downey & Lenkov to discuss the growth of Artificial Intelligence in the handling of workers’ compensation claims. Michael takes a dive into the use of AI by insurers and claims professionals and how the data obtained through AI can assist attorneys in better predicting the outcome of claims including the claim value and potential fraud. Michael discusses the uses and the state of AI in workers’ compensation currently, with an eye toward the future of AI for insurers and attorneys. There are also serious risks that must be considered when personal information is involved in data collection and when attorneys choose to rely on AI without checking for its accuracy. You can find Michael on LinkedIn at https://www.linkedin.com/in/michael-milstein-741a9a58/ at his firm’s website www.dl-firm.com.
To learn more about DRI and the Workers' Compensation Committee vist www.DRI.org.
#DRILawyer
#DRICommunity
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:18):
Welcome back for another episode of thecomp Conversations podcast, a podcast developed through
dri where we explore the world ofworkers' compensation, including the personal aspects involved
in the realm of work injuries andits litigation, as well as future trends
and every aspect in between. Myname is Ryan Hathcock. I'm attorney lecensed
in the state of Georgia. Withme as always is my friend Steve Armstrong,
(00:39):
an attorney license in the state ofKentucky. In this episode, we
will discuss a topic many of ushave heard about in the past year,
but likely know very little about.Today, we all speaking with Michael Milstein
to discuss artificial intelligence, its usein the legal community, and the anticipated
impact it will have moving forward.Michael, who has practiced law for over
sixteen years this is a capital memberat Downey and Linkoff. He focuses practice
(01:03):
on defending Illinois workers compensation claims onbehalf of employers, insurance companies and TPAs.
Recently, Michael and his partner athis firm presented on the topic of
AI and its impact on workers' compensationclaims, and we are excited to finally
dive into this topic with him today. Michael, Welcome to the podcast.
How are you good, Ryan,Good morning, Steve, Thanks for having
(01:25):
me on guys. Good morning,Michael. Great. I'll admit it,
I'm pretty much a novice when itcomes to AI, besides playing around with
a little bit of Chat GBT atthe request of one of my friends who
suggested to me, I would nottake you for a big AI guy,
but maybe I'm wrong. What's yourexperience with AI and do you use it
(01:46):
in your practice. I have verylittle experience with AI, and I've been
on chat gt GPT all of twoor three times total, and so I'm
intrigued. I'd love to hear moreabout it. I'd love to be able
to use it in practical ways andlooking forward to hear what Michael has to
say on that, especially in termsof what we could do now versus what
(02:07):
we can do in the future.So with that, my thought is,
let's get started. Thanks for beinghere again today with Michael, and your
presentation is awesome. What can youtell us about what you do with AI
in your practice today? So Iguess I'll start by saying, Steve,
you answered that you said you've neverused AI as part of your practice and
(02:30):
I'm sure you have. You mayjust not necessarily realize it because AI has
been around in some form or anotherfor decades. We use AI even like
the most basic search function of youknow, a PDF is some sort of
AI that's helping us, you know, decipher information that we couldn't necessarily do
ourselves in a quicker fashion. Sothere are definitely those aspects that you use.
(02:53):
But the computing power has just increasedso much in the last six months,
twelve months, I've heard I've heardthe example being like what we used
before sort of a toddler, andnow we're we're up to a teenager as
far as the mental awareness of theAI, which in the short period of
time is a major jump. Butthe fact that we're only a teenager right
(03:15):
now shows that we have a longway to go still into uh, you
know, putting this in together ina way that we as litigators that we
can use it to really help ourpractice. I think that, you know,
do I. I don't use iton a day to day basis.
I'm not there yet. I don'tthink the programs are there yet. But
I do know that many insurance companiesare using it every single day, and
(03:37):
we can definitely dive into more howthey're doing that. You know, as
we talk, I think the veryinitial part is like what am I doing
to use it? I think themain thing that we're using is just chat
GPT and sort of understanding it's understandingof its understanding of how much it knows
about our relevant workcomp laws okay,and so chat GBT. I've again I
(04:03):
plugged in a few queries on thatand it did it. You know,
I don't know what it searched from. I don't know where it got whatever
information it got. Do you knowon that? What can you tell?
I think, you know, oneof the big pitfalls that we're going to
eventually learn about all AI is howit obtains its information and where it obtains
(04:25):
its information from. You know,currently there's a bunch of different one from
Google or Facebook or open ai,and you know, I think they pull
their information mainly just from the Internetin general, mainly you know, for
Google, be the Google Search,and it can it can identify the type
of information it's looking for, retainit and then contextualize it to give you
the right answers. But the keythere is where is it getting this information
(04:47):
from? Right? So if you'rea if it's getting information from my website,
you know you're going to get atilt towards how employers do cases.
There are do you is you cantype in a chat GPT is a you
know, certain type of case compensable? And if it's getting into information from
responding friendly websites, it's going totell you, well know, if it's
(05:08):
getting it from a petition of friendlywebsites, it's going to tell you.
Yes. Generally, what it tellsyou currently is it could be you should
you know? Thenically it says youshould retain an attorney to find out is
sort of often what it tells you. And so I'm very thankful that it
is smart enough to know consult anattorney. You know us three. It's
a little referral source for us tookay, I need I need to go
(05:29):
no more. But it can definitelybe helpful in a few ways of just
like you know, Steve, you'reyou're in Kentucky, Ryan, you're in
Georgia, right, I don't knowyour state's laws. And yes, I'm
I'm thankful to have you both asa resource if I have questions about your
jurisdictions. But if I don't havesomeone that's available for me in those jurisdictions.
Chat GPT is an excellent source tostart your research to find out where
(05:53):
it is. Now. It isnot the end all, and you should
not trust it just out of handbecause we can talk about it about it
an hour or later. But thereare stories of attorneys using these the chat
chet and it giving it incorrect information. So you know, you can ask
it, you know what is basicallywhat is the stantial limitations file workers conversation
(06:13):
Clay in Georgia. Right, wecan do that and it'll probably and if
you want in while we're talking,I can pull it up and we're gonna
answer these questions and we can seehow well, how accurate or not it
is. But end of the day, where we are right now, it
is not a tool that you shouldbe utilizing to answer all your questions without
secondary research of you know, humanreading it on lexus or whatever it is
(06:36):
that you need to do. Butit is a really good, uh starting
point. So from a starting point, you know, I guess my thought
is if we could somehow plug itinto a certain states laws or a certain
states reported recorded cases, reported casesthat would be helpful. Are we there
(07:00):
yet I think like for case law, I still think like lexus is where
you go, because again, whereis it getting its information from. I
don't think you know, the Internet. If you look up certain cases,
they're available and they're just available onyou know, fine Law that are just
free and they have all the casesand you can put that information a lot,
(07:24):
you know, into the into chatGBT, whoever you know owns it
and does it and they can learnit. And so part of the issue
that we've had in the past isthe thing there's something called structured data versus
unstructured data. Structured data is sortof what we have that anything that you
can put into a graph or intoa table, right you can you can
(07:44):
put all that stuff into an Excelspreadsheet, very easy to for a computer
to understand, and computers have beenunderstanding that information for some time. But
then there's unstructured data, things likephotographs, claim notes, other things like
that that are now being able tobe read and understood. And there's this
there's a lot of inherent difficulties indoing that. One of them is I'm
(08:07):
gonna sort of go off the topof my head, just understanding language is
a big portion of what chat cybtneeds to do. Like if you use
the sentence the head janitor, usethe head of a tool to reach the
head, you know, just repeatingthe same word over and over again.
But it has different, different contextualmeanings to us now as humans, we
(08:31):
clearly understand what those means immediately,but to understand it to a computer is
very different. Another one's like automobileversus car, versus vehicle versus Honda,
right, Like, all those areobvious to us, but it may not
be obvious to the machine as towhat it is. So teaching it that
what those mean is very difficult andmore importantly, very expensive. This using
(08:56):
chat cybt in preparing these softwares.It's not something for like a Mom and
pop that they can just go outand be like, I'm going to create
my new AI for my claim.Right, you can't do that because it
just requires so much money, becausethere's so much time and energy need to
be put into it, that wedo need to rely on these larger companies
that are willing to spend tends,not hundreds of millions of dollars to make
(09:18):
a good CHET, GPT or AI, whatever you want to call it.
So they it can be helpful forclaims, which is what we're you know
we're talking about right right, SoI get these ads almost every day from
either Thompson or Lexus or a casetext or somebody else about how it can
help you as a lawyer doing researchor doing this, that or the other.
(09:41):
My own wish list is it wouldhelp me do discovery or figure out
the claim on the front end.But what are programs? What can we
use it for right now in termsof working on a claim, or we
could go back to what you mentionedabout insurers or using it for these days
your objects and so on. Thatso, just so, just so,
(10:01):
I asked chatchpt what is the statueof limitations to file workers compensation claim in
Georgia and its response was to asmy last knowledge in September twenty one,
So that's sort of when it's cutoff. Is so if things have changed in
the last two years, it won'thave that information. It says to file
in Georgia typically one year from thedata work related injury or illness. That
means you have one year from thedata of injury to the day you became
(10:24):
aware of it to file a claim. Please note the laws change and so
it's essential to consult an attorney orcheck the most update workers compensation dot dot
dot right, so it is tellingyou it mean is that correct? I
mean, I mean that is correct. So it's still it's still correct.
Nothing's changed in the past couple ofyears, so nothing has changed. But
so it has the information, butit does as I said, it says
consultant attorney. But what I thinkwhere the cutting edge is right now on
(10:48):
using AI, and it is beingused at a very high level at some
larger insurance companies, is they're ableto take the data from claims they can
take so you can take us astructured data. Some of the structured data
age, gender, location, jurisdiction. These are things that you can easily
just filter in and out. Butthen even injury can be considered structured,
(11:13):
right, and you can plug allthat in and if you have if you're
an insurance company and you have thousandsof claims, you can now run your
search and say these are the typical. And now I have a motor vehicle
accident with someone who at are brokenhumorous, they're a forty year old male
from Illinois. All right, youcan put that into your thing and you
can then look up every other claimthat you've had like that and then find
(11:35):
out what were the results of thatclaim, how much money was spent on
that claim. That's a big one. You know, reserves, as you
know, as we all know,being you know, defense attorneys, reserves
is a very important aspect of ourclient's job. And sometimes I think there
are some claims examiners who take itvery seriously and there's some that aren't.
(11:56):
But you know that the people whoare higher up the chain, they take
reserves very seriously, and being ableto accurately reserve a claim at its onset
can have a lot of cost savingsto that company in the future. So
being able to do that. Butnow what's happening is not to just have
that structured data. They can nowread the medical records of all these prior
(12:16):
claims and they can see the typeof treatment that those individuals had. So
one of the things that I wastalking about recently with a large insurance company
is their data can let them knowwhether or not someone who is in this
with this structured data and unstructured dataput it together, are they likely to
return to work? Right? Theyof the thousand claims that we have that
(12:39):
are similar, only fifteen percent havereturned to work with that type of injury,
or you know other way, ninetyfive percent have returned to work.
Now, as an attorney, it'svery nice if I can get that data
the second I get the claim.Imagine if you've got a claim and it
says our data says that someone likethis is likely not to return to work,
that will have a direct impact onhow you handle that case. Because
(13:03):
sometimes as our goal is to getthese guys or girls to return to work
the injured worker, we want themthat MMI full duty return to work.
But if we know from the getgo that is a very unlikely scenario,
well, then it will impact howI'm going to attack that claim. I
may want to now try to sellthe claim much earlier, knowing that I'm
going to have bad results. Inthe same way, medically, I think
(13:24):
it helps because if I can findout that someone who has this type of
injury and you know it can breakit off. You can say, you
know, fifteen percent end up havingan orif five percent end up having CRPS
or whatever made up pain thing thatthe doctors want to put on them.
And this many improved with injections.Getting getting data on how many work compatients
(13:46):
get this type of treatment versus nonwork compatients and seeing the difference can be
very helpful as well. And thatdata is becoming available that the insurance companies
have from the get go. Theyneed to type in the information put in
them you know whatever information that whateverdata points UH, and they can find
out a lot of information. AndI asked, what what is the most
important interative? Think what you guysknow or what do you think? What
(14:09):
is the most important data point UHto determine the expected costs of a claim
type of injury? I would saylocation, Steve has it, the jurisdiction
is going to is the number onedata point to determining cost of claim,
which when you when you think aboutit, does make sense because Illinois,
(14:31):
for example, is a very expensivestate medically and for indemnity versus our neighbor
to Indiana is a very different,you know, very different ballgame in Illinois.
We don't get a control medical inIndiana. They do in Illinois.
Impairment rating is one factor to determineyou know, PPD in Indiana, that
is the PPD whatever the impairment ratingis, so these jurisdictions have wildly different
(14:54):
rules as we all know, whichis what you know, keeps us all
doing what we do separately and havesort of no clue what's happening in state
to state. But under just thatdata point alone, it's going to have
a massive impact on how uh theinsurance companies do cases and how obviously the
attorney's hand. Yeah, well,I know, I know, you know,
(15:15):
when we get contacted about a claim, you know, the first information
we ask is it's all that gymwal contact information about you know, uh,
the age, where you know,what's the type of an injury,
all that basic information that you knowtypically in your scenario, the insurance company
would already have and then kind ofput that into their claim projection software and
(15:35):
it kind of spits it out sothat we when we get running, we've
already got a ton of information.But have you actually already seen insurance companies
using this and are is it somethingthat they're beginning to do and transition to
their attorneys? So uh none,Sadly for me, none of my insurance,
none of my clients have a robustAI system set up yet. However,
(15:56):
I do know there are many outthere I was just at a seminar
last week about this. I wasin Philadelphia. I was asking someone are
about and it was with it doesn'tmatter who the insurance company is, but
it was a very large insurance companyand they are using and I And I
asked them like, are you relayingthis information to your attorney? Because I
was sort of maybe it was thecynic in me where an INTERNCE company could
(16:18):
use it as a test in away for their for their attorneys to be
like, we know what the likelyor likeliest outcome is, you know,
are we going to let our attorneyknow that in advance or are we going
to see what they do and howwell or not well they are able to
get to that end result that wethink is most likely. Now that's a
cynic in me thinking clients are youknow, sometimes trying to you know,
(16:40):
be secretive with us. But inreality she said no, we are absolutely
given that this to our attorneys,which to me would be it would be
such a gift if I got acase in I said, oh, and
also, here's our separate AI resultsthat we've run on this that gives you
all these potential scenarios. But evenI think I always thought medically. As
a work comp attorney, you becomevery dangerous in your knowledge of orthopedic medicine.
(17:03):
I am far from a doctor.I cannot read an MRI result,
but if you tell me someone hasa fractured blank or a torn blank,
I can likely tell you what willhappen next. All the way through MMI,
now there are things that happen alongthe way that sometimes surprise me.
And those are the things I wouldlove to know in advance, to say,
how often does this gunicular injection occur? Right? That's not something we
(17:26):
see too often, but it is, or this radio radio frequency ablation happen
right for this type of this typeof pernia disc in the lumbar spine.
Those type of things would be veryhelpful for us to know, because not
only for just making the case movealong efficiently, but as a red flag
monitor that will pop off and say, hey, this thing. You know,
(17:49):
we sort of know it inherently thatI don't see this very often.
If I don't see something very often, that is a red flag. I've
been doing this forever, and Isee pretty much any kind of injury and
treatment course recommended. But to knowthe not only whether or not. I
think it's a red flag. Butwhat are the results of the claims that
we've had in the past where thistreatment occurs. Right, If I know
(18:10):
it's absolutely worthless and no one getsbetter, I'm going to fight a lot,
very hard through every method I can, through IEME, through utilization reviews.
But then, as the client says, listen, we've actually don't see
this very often, but we seereally good results with it. So maybe
it isn't the best decisions to tryand fight it. That information would be
really good to know, because Idon't entirely I don't trust treaters. I
(18:30):
don't trust pain managment doctors because knowwhat a pain managment doctor gets paid to
do. Inject and to prescribe medication. That's what they do. That's how
they make money. So you don'tsee the pain management doctor who stays in
business by being like, no,you'll be fine with conservative care. That's
not what they do. So itis incumbent on us as responsibornity to have
that cynical mind or just very aggressivestance towards this type of treatments. But
(18:55):
the more data I have, themore information I have, the better I'm
going to be to represent our clients. Yeah, and I think you know,
if insurance companies are able to provideyou that information fairly early on and
you've got to claim that maybe isyou know, not in your best interests
in terms of the claim. Istreating with the provider of their own choice,
and you know they've got this imagingthat shows something and they're recommending some
(19:15):
certain procedure. But then you getthis documentation in or this you know,
this information that shows well and thisin these certain kind of cases, you
know this procedure is not typically warranted. I see how you could take that
information to maybe mediation or negotiation potentiallydrive down the value of the claim.
So having that you know, datais very useful in certain situations. Of
course, like you mentioned, ifit's not in your favor, it's good
(19:37):
to know that too, very veryearly on. So I can see where
all this information, having it atyour fingertips is very beneficial for attorneys,
not necessarily in terms of how youpractice, but in terms of managing each
claim as it comes along. Whatare the concerns that you see with AI
for practicing attorneys, Well, thebiggest concern I think what we mentioned,
(19:57):
sure, right, the outset isjust blind trust and that what is telling
you is accurate because it's not.And so there's a good there is a
good story for those of the worryto understand. Something that did happen in
New York State. An attorney youknow, new to AI. It was
a case involving an airline where someone'sI think he got someone got hit with
(20:19):
the metal cart that they push across. Right, that was the facts of
the accident, and they were emotionto semiss was filed and an attorney went
on chat GPT and said, hey, I need a case law for this,
and it gave them case law andthey used it in a brief.
The other his opposing party, youknow, as any good attorney does,
(20:40):
was like, hey, I needto research these cases, see what they
say, see how I can attackthem, and lo and behold. Could
not find those cases. So theysay to the court judge, you know,
they file a motion to I thinkquash the brief that they filed,
and they said, hey, noneof these cases there exist. So the
attorney said, oh, don't worryabout I'll go get you the cases.
So what do they do or theysay, hey, chat, are the
(21:02):
cases you sent me real, andchat ChiPT says, yeah, the real,
of course, the real. Sothe attorney goes back and says,
none, they're real, and Ithink he's able to print out some like
citation for them, and you know, still obviously not real. So it
was funny though, because when heasked chat GBT, are these legitimate cases?
The chat GPT said yes they were, and so I don't think legal
(21:25):
research And what was funny about what'shappening now? So they're trying to discipline
the attorney who presented those fake cases, and the attorney's response is, don't
you think I've been disciplined enough?Like I am a national news story and
an embarrassment for having Jesus they doyou think I need formal sanctions? Or
is my name being plent? Youknow put And if you look at attorney
(21:47):
chat GPT, anything negative, he'sthe first research result. So I would
say that he's he served his time, time to serve should be enough and
let him let him go. Butthat is a huge red flag that you
need to be aware of, isthat it's just not gospel and maybe one
day it will be. Maybe oneday it'll take the place of a Lexus
(22:08):
or a West law that we canuse for legal research. We're just not
there yet because the industry as awhole, and I mean the AI industry
is you know, it's it's stillbusiness and they're looking at where they can
make the most money and be helpfuland and there's lots of ways. It
just hasn't trickled down yet to beingfully enmeshed in the legal field. It's
(22:30):
getting there, As I said that, there are insurance companies using it,
and there are dedicated teams at theseinsurance companies who are being looked at to
create software and and create models.And but the thing is with these models
is every time it doesn't take aday to create a model, it takes
time, a lot of time.And by the minute they're done creating a
(22:51):
model, the the ability of AIhas so increased that that model is now
out of date and that the sortof go right back to the drawing,
which is can be a very imaginefrustrating thing. These individuals are also very
exciting because I feel like they werestagnated for a very long time in the
capabilities of AI to do anything,and now it's just increasing exponentially week by
week that it is a very excitingtime for it. But I don't think,
(23:17):
you know, for litigators, Idon't think it's anything to worry about
that like it's going to put usout of a job. Far from it.
I think that you know it,it's going to be a good tool
for us long term. And Idon't I don't know if for any three
of us it will be part ofour day to day. But I can't
imagine if like my son, myseven year old son wants to become an
attorney, Uh, it will besomething that he uses on a regular basis,
(23:40):
just like when I started law school, because I'm when I started law
school, we were still researching inbooks. By the time I finished law
school, nobody ever used a book. We were all on lectus snatches,
right, I just was write inthat time period, and who does legal
research on books? But you knowanymore, right, it's just nobody.
Everyone's library has been if people keepit for the photo is behind us,
(24:00):
Like, hey, we're lawyers,we have books behind us, right,
But to show it's going to continuouslyevolve and become a much greater part of
our practice. And I think inthe way that we talked about with with
not us having the resources, butother companies having resources where they put in
a lot of data that will helphelp us understand and it can even go
(24:22):
farther right, it can go youknow, in Illinois, the judges are
arbitrators, And what if I wasable to quickly go through an arbitrators decisions
like the one like the issues I'mhaving, or the decisions with regarding awards
like the one I have. Right, there's still the commission, there's appeals,
but I would love to know howdoes this judge do on a regular
basis over the course of the lasttwo hundred decisions on an accident defense.
(24:45):
Now I could even slim it downto my accident defense, but do they
ever find accident defense is good ornot? Like that would definitely color my
decision to our my client to belike, hey, our likelio to success
is blank? Right to offer?People are saying fifty fifty years. It's
one of the things I hate inlife. And when an attorney, when
I have a client call me andsays they say, like an attorney told
me it was a fifty to fiftychance, Like, what do you mean
(25:07):
a fifty fifty chance? There's nosuch thing as a fifty thfty chance you
have to give real advice. Butyou know, the more data we get
will make it easier to give ourclients more accurate and well informed advice.
It would be helpful if we knewwhich judge or which arbitrator was citing what
case is the most in certain instances, you know, if you have a
certain type of defense, it wouldbe great to know exactly what they rely
(25:30):
on and what they find credible.Right, and again that all this information
has to be public because they havethey have to find a way to be
able to get it. So likein Ellinois, all the Ellinois Work Commission
decisions which is the appeal level,all those are public. They print them
all out, but you'd have toliterally read through it, you know,
for me, I'd have to readthrough every single decision and do it all
myself. But if I can justgive it to a machine and say,
(25:52):
hey, you know, I cantell what I wanted to learn, you
know, would learn by judge,learn by these issues, and tell me
what happens, you can sort ofdo that. That those capabilities are available
now, but you just need tohave the right you know, engineers to
create it. And I think iflawyers, we don't have that. That's
not you know, we're lawyers.We're not here. Computer technological skills is
(26:17):
far from our where we shine.And it's funny when you talk to a
lot of lawyers, like the lackof technology knowledge that they have is sometimes
embarrassing, especially for some of theyou know what, I consider that the
generation before me, like people stillbuild by hand, right, they still
(26:37):
write it down. They won't enterit in the computer. There's a lot
of that, and just imagine tryingto get them to use an AI or
a chet GVT function. You knowit's going to be nearly impossible. So,
you know, some ethical codes insome states even have a requirement that
attorneys be aware of and learn newtechnologies. I don't know that that involves
AI. Yeah, but I don'tknow if it's done here. Another pitfall
(27:02):
potentially, and you know it's somebodyto think about, are just like hippo
laws. Are you giving people's medicalrecords to third parties? Right? You
know what what access are they allowedto have? Now? You could maybe
get rid of identifying materials like oneof the things that we actually have to
do in Illinois. If I'm submittingan exhibit to the arbitrator, I have
(27:23):
to remove all social Security number information, right that has to be blacked out.
They can't have that because of hippolaws, and you know, we all
know that hackers and that stuff becomesa real concern. And now with these
medical records for you know, millionsof people are now being handed over third
parties. That that's a that's amajor concern, not just for like,
(27:44):
not not for US litigators, butjust for the public because I don't think
that it would would be a goodthing for us to be doing, to
be handing over people's medical records withoutyou know, consent or just making sure
that it's protected in a way thatwill you know, make sure that they
don't have to deal with any unintendedconcerts, right right, So to go
back to the gentleman in New York, I know, in response to that,
(28:07):
I heard on another podcast a judgein Texas is issuing a standing order
that no attorney can use AI inany pleading unless they get specific permission to
do so. On the front end, I thought that was fascinating as a
way to just have a standing orderfor all attorneys in all cases. Yeah,
(28:27):
I mean, I guess it's notsurprising to some degree given how new
it is, and they want toavoid the issue that happened in New York.
Now. I think like a blankettime people that you can't use it,
but like you shouldn't cite, butlike you shouldn't cite to AI.
Right, like there is if you'reif you're writing a brief, it's case
law. Case law is readily availableto attorneys. It's not something that you
(28:48):
know, you need to really gocreate or ask the CHANGPT. You should
be able to find yourself. Butit's not surprising, and you know that
that is sort of the step peopleare doing as they see people making mistakes
that you know, that's let's letsome other jurisdiction figure it out first,
and then you know, we'll adoptthat. I will say, the legal
field is not necessarily always you know, the first one out there to try
(29:10):
something new. I'm happy with howthings happened with using technology as far as
remote court hearings, but I thinkthere was a lot of you know,
fight back, you know, fightingfrom different sides of the bar about even
just doing remote court hearings right thatthat that was a big deal, and
there are some concerns that do processconcerns and criminal cases that don't necessarily happen
in a pre trial hearing for workcount case that you know we deal with.
(29:32):
But so, yeah, so whatabout fraud detection? What do you
see h AI being usedful for interms of workers compensation fraud detection? I
think I think it's the same thing. So one of the other data points
that uh I was told is abig indication of fraud is date reported.
(29:53):
So if they know that you reportedthe case four days after the accident,
those simple things is a big redflag for them. And it's going to
allow and just on its own willwill be a good red flag for the
insurance companies and to find out whetheror not someone is committing fraud and in
work there's to me a lot ofdifferent type of fraud. You know.
There's there's the initial fraud of theobvious that they're claiming an accident happened that
(30:18):
never happened. And I I thinkwe've all had our fair share of those
type of cases that people are juststraight ling nothing happened to them, They
got hurt over the weekend, orthey're not hurt whatever. That's the that's
the obvious fraud, But I thinkwhere the fraud happens more in the every
day that we deal with is ofthe lingerers and the symptom exaggerators where they're
(30:38):
you know, the statistics are prettyclear about the results from a work compatient
compared to the general public, andhow much worse results a work compatient will
have than than the general public forpretty much any injury, and so AI
being able to spot those for whatever, you know, one reason being the
day reported, and you know,other reasons potentially being how much pain they're
(31:00):
complaining of. Because again, whatthe biggest difference now between what AI is
able to do now that it wasunable to do before is to understand context.
It can read a sentence and understandwhat it means, even if you're
using right the same word. Itcan understand the context. One of the
things I, you know, whatI've been doing my whole career is I've
been reading through physical therapy records,you know, combing through them, because
(31:22):
I think for me, physical therapyrecords is a great place to find inconsistent
statements and inconsistent exam results because forsome reason, people love their physical therapists
and they feel like it's a therapysession of a psychological therapy session versus a
physical therapy session. And they'll tellthem how, you know, over the
weekend, I, you know,fell down, I was in a car
(31:42):
accident or whatever it is. They'llbe very honest with them, or they'll
say I'm feeling great, and thenthey'll go to the doctor two days later
and be like, I'm feeling terrible. Ten to ten pain straight leg ray
is positive even though it was negativehere. So I mentioned you count have
AI do all that for you andcan be done within seconds and just point
out you petitioner said pain was threefour or five ten, and just list
(32:04):
it and you'd be like, oh, this is something something's wrong here,
just right off the bat, something'snot right. So I think for fraud
detection it is a it will bea very useful tool and trying to put
together, you know, a bunchof data points for you, whereas you're
only able to maybe focus on afew at a time because you're human versus
the machine. Can you know,look at sort of the whole thing is
(32:27):
as a combined and find out theseyou know, something's not right here,
red flag it for you and thatwill affect again how you handle the claim,
how your client wants you to handlethe claim, and what the expectations
are regarding the results of the claim. Yeah. Well, just thinking forward
in terms of the future, it'sreassuring, and you've kind of confirmed my
thoughts that I don't really see AItaking over the role of a lawyer.
(32:52):
It can certainly be very useful fordata collection and using the data points to
project a claim. As long asyou are using AI, you know,
with some relatively common sense not torely on it for terms of case law
and things like that, it canbe a very useful tool. Certainly.
We also discuss the concerns that wehave with you know, using other people's
(33:14):
information. I feel that insurance companiesyou know, can extract that data on
their own really, you know,in terms of the results of the case,
without really any type of approval fromfrom the patients, as long as
they aren't using the patients you know, identify identifying information. Are there any
other concerns or benefits that you seeother than what we've discussed as we're kind
(33:34):
of wrapping it up here today,I mean, as far as I think
we've hit the concerns pretty well,I think anyone who's just thinking about this
abstractly will come to the same realizationthat we are discussing here about what the
concerns are. I think there's there'sprivacy concerns, and there's reliability concerns.
Those are I think the major ones, and I think for what the pros
are, I think that the differenceis and it may be true also on
(33:59):
the con side is that we don'tknow yet. That's the you know,
this stuff is still growing at arate that is, you know, unlike
something we've seen in technology in avery long time. So we can try
and you know, look at thecrystal ball and to see what here's where
we think it can go. Butoftentimes when we look at technology, you
know, did anyone foresee when theiPod kind came out that the that the
(34:20):
iPhone was going to revolutionize the worldthe way it did like when we had
of course not, and that wasonly a couple of years away. So
I think that we have ideas asto things they can do, and I
think for attorneys it will be alargely a time saving tool for us and
and give us a lot of additionalinformation, which is also all great,
but I don't think we're going tobe displaced, you know, legal zoom
(34:42):
has been around for a long time. Our wills and will in the state's
attorneys out of a job, noright, our contract attorneys out of a
job. No, not maybe someare. But it's also sort of like
in a way like that stuff hasdemocratized law, which is a good thing.
I think it gives people the accessto hopefully good you know, good
work product without having to pace andyou know, so now everyone can afford
(35:04):
the right attorneys, and this willgive you know, the public knowledge which
sometimes as the attorneys, we don'tlove it when the public health knowledge is
the you know, and a foreignclient can be a dangerous one. They
think they know more than they do. But that's sort of for us and
our jobs, but and the wholeyou know, I'm still of the opinion.
And there's been a lot of doomand gloom about where AI is going,
the end of humanity. You hearabout you know that it's going to
(35:27):
put everyone out of a job.You know, I don't know if everyone
remembers when and this is like notAI but just technology. Remember when they
had the self served counters at thegrocery store, that all the workers at
the grocery store were going to beout of a job. Obviously, that
didn't come to fruition, right.So I think that when when these new
technologies come aboard, we're we sortof are fearful of them, and I
think part of that ends up,you know, damping them down, because
(35:51):
the public doesn't want something that's goingto take over everything. They want it
to just be a convenience for themost part. And so yeah, I
think the insurance company is going touse it for a good purpose and is
helping move claims long efficiently because again, costs, what are these insurance companies.
A lot of insurance compans care about. They care about costs, right,
keeping costs down. And I thinkthat's what AI is going to help
them do. And I think thatthat's gonna be good for everyone, right,
(36:15):
you know, trickle down, saywhat we want to trickle down economics
on a glow on a big level, But if a company that you work
for is able to save some money, then hopefully that will lead to reduced
premiums and and better results for everybody. That that would be the hope.
And that's an optimistic way of lookingat it. Yeah. Well, Michael,
I want to thank you again forgreeing to join us on the podcast.
(36:37):
Steve and I have been wanting todiscuss this topic for some time,
and we really appreciate your insight.Our listeners can't see us, but Michael's
dressed way nicer than us today,so he's he's out to do some real
lawyering today, so we'll let himget to it. But where can our
listeners find you if they want toget in touch with you or learn more
about AI and workers compensation claims?Well, thank you guys again. Thank
you both so much for having meon. It was great talking to you
(36:58):
guys. I do Illinois's compensation.My firm to sort of Illinois everything GL
Construction. You can find me atDL Hyphenfirm dot com. Is our address.
My name again, Michael Milstein.My phone number Casey went out reach
out three one two three two sevenzero zero four two and then my email
is Mmilstein at Dlhyphenfirm dot com.If you ever have any questions in Illinois
(37:21):
work com please do not hesitate toreach out to me, text, email,
pigeon whatever you need. I amalways available what I I always say.
What I try to differentiate myself onis you know, the customer service
that I provide is not the sameas what everyone else does. So you
can always reach me at any timeand I hope to hear from someone soon.
Thank you, Michael. I appreciateit very much. Thanks Michael.
(37:42):
Thank you every great day.
Welcome back for another episode of thecomp Conversations podcast, a podcast developed through
dri where we explore the world ofworkers' compensation, including the personal aspects involved
in the realm of work injuries andits litigation, as well as future trends
and every aspect in between. Myname is Ryan Hathcock. I'm attorney lecensed
in the state of Georgia. Withme as always is my friend Steve Armstrong,
(00:39):
an attorney license in the state ofKentucky. In this episode, we
will discuss a topic many of ushave heard about in the past year,
but likely know very little about.Today, we all speaking with Michael Milstein
to discuss artificial intelligence, its usein the legal community, and the anticipated
impact it will have moving forward.Michael, who has practiced law for over
sixteen years this is a capital memberat Downey and Linkoff. He focuses practice
(01:03):
on defending Illinois workers compensation claims onbehalf of employers, insurance companies and TPAs.
Recently, Michael and his partner athis firm presented on the topic of
AI and its impact on workers' compensationclaims, and we are excited to finally
dive into this topic with him today. Michael, Welcome to the podcast.
How are you good, Ryan,Good morning, Steve, Thanks for having
(01:25):
me on guys. Good morning,Michael. Great. I'll admit it,
I'm pretty much a novice when itcomes to AI, besides playing around with
a little bit of Chat GBT atthe request of one of my friends who
suggested to me, I would nottake you for a big AI guy,
but maybe I'm wrong. What's yourexperience with AI and do you use it
(01:46):
in your practice. I have verylittle experience with AI, and I've been
on chat gt GPT all of twoor three times total, and so I'm
intrigued. I'd love to hear moreabout it. I'd love to be able
to use it in practical ways andlooking forward to hear what Michael has to
say on that, especially in termsof what we could do now versus what
(02:07):
we can do in the future.So with that, my thought is,
let's get started. Thanks for beinghere again today with Michael, and your
presentation is awesome. What can youtell us about what you do with AI
in your practice today? So Iguess I'll start by saying, Steve,
you answered that you said you've neverused AI as part of your practice and
(02:30):
I'm sure you have. You mayjust not necessarily realize it because AI has
been around in some form or anotherfor decades. We use AI even like
the most basic search function of youknow, a PDF is some sort of
AI that's helping us, you know, decipher information that we couldn't necessarily do
ourselves in a quicker fashion. Sothere are definitely those aspects that you use.
(02:53):
But the computing power has just increasedso much in the last six months,
twelve months, I've heard I've heardthe example being like what we used
before sort of a toddler, andnow we're we're up to a teenager as
far as the mental awareness of theAI, which in the short period of
time is a major jump. Butthe fact that we're only a teenager right
(03:15):
now shows that we have a longway to go still into uh, you
know, putting this in together ina way that we as litigators that we
can use it to really help ourpractice. I think that, you know,
do I. I don't use iton a day to day basis.
I'm not there yet. I don'tthink the programs are there yet. But
I do know that many insurance companiesare using it every single day, and
(03:37):
we can definitely dive into more howthey're doing that. You know, as
we talk, I think the veryinitial part is like what am I doing
to use it? I think themain thing that we're using is just chat
GPT and sort of understanding it's understandingof its understanding of how much it knows
about our relevant workcomp laws okay,and so chat GBT. I've again I
(04:03):
plugged in a few queries on thatand it did it. You know,
I don't know what it searched from. I don't know where it got whatever
information it got. Do you knowon that? What can you tell?
I think, you know, oneof the big pitfalls that we're going to
eventually learn about all AI is howit obtains its information and where it obtains
(04:25):
its information from. You know,currently there's a bunch of different one from
Google or Facebook or open ai,and you know, I think they pull
their information mainly just from the Internetin general, mainly you know, for
Google, be the Google Search,and it can it can identify the type
of information it's looking for, retainit and then contextualize it to give you
the right answers. But the keythere is where is it getting this information
(04:47):
from? Right? So if you'rea if it's getting information from my website,
you know you're going to get atilt towards how employers do cases.
There are do you is you cantype in a chat GPT is a you
know, certain type of case compensable? And if it's getting into information from
responding friendly websites, it's going totell you, well know, if it's
(05:08):
getting it from a petition of friendlywebsites, it's going to tell you.
Yes. Generally, what it tellsyou currently is it could be you should
you know? Thenically it says youshould retain an attorney to find out is
sort of often what it tells you. And so I'm very thankful that it
is smart enough to know consult anattorney. You know us three. It's
a little referral source for us tookay, I need I need to go
(05:29):
no more. But it can definitelybe helpful in a few ways of just
like you know, Steve, you'reyou're in Kentucky, Ryan, you're in
Georgia, right, I don't knowyour state's laws. And yes, I'm
I'm thankful to have you both asa resource if I have questions about your
jurisdictions. But if I don't havesomeone that's available for me in those jurisdictions.
Chat GPT is an excellent source tostart your research to find out where
(05:53):
it is. Now. It isnot the end all, and you should
not trust it just out of handbecause we can talk about it about it
an hour or later. But thereare stories of attorneys using these the chat
chet and it giving it incorrect information. So you know, you can ask
it, you know what is basicallywhat is the stantial limitations file workers conversation
(06:13):
Clay in Georgia. Right, wecan do that and it'll probably and if
you want in while we're talking,I can pull it up and we're gonna
answer these questions and we can seehow well, how accurate or not it
is. But end of the day, where we are right now, it
is not a tool that you shouldbe utilizing to answer all your questions without
secondary research of you know, humanreading it on lexus or whatever it is
(06:36):
that you need to do. Butit is a really good, uh starting
point. So from a starting point, you know, I guess my thought
is if we could somehow plug itinto a certain states laws or a certain
states reported recorded cases, reported casesthat would be helpful. Are we there
(07:00):
yet I think like for case law, I still think like lexus is where
you go, because again, whereis it getting its information from. I
don't think you know, the Internet. If you look up certain cases,
they're available and they're just available onyou know, fine Law that are just
free and they have all the casesand you can put that information a lot,
(07:24):
you know, into the into chatGBT, whoever you know owns it
and does it and they can learnit. And so part of the issue
that we've had in the past isthe thing there's something called structured data versus
unstructured data. Structured data is sortof what we have that anything that you
can put into a graph or intoa table, right you can you can
(07:44):
put all that stuff into an Excelspreadsheet, very easy to for a computer
to understand, and computers have beenunderstanding that information for some time. But
then there's unstructured data, things likephotographs, claim notes, other things like
that that are now being able tobe read and understood. And there's this
there's a lot of inherent difficulties indoing that. One of them is I'm
(08:07):
gonna sort of go off the topof my head, just understanding language is
a big portion of what chat cybtneeds to do. Like if you use
the sentence the head janitor, usethe head of a tool to reach the
head, you know, just repeatingthe same word over and over again.
But it has different, different contextualmeanings to us now as humans, we
(08:31):
clearly understand what those means immediately,but to understand it to a computer is
very different. Another one's like automobileversus car, versus vehicle versus Honda,
right, Like, all those areobvious to us, but it may not
be obvious to the machine as towhat it is. So teaching it that
what those mean is very difficult andmore importantly, very expensive. This using
(08:56):
chat cybt in preparing these softwares.It's not something for like a Mom and
pop that they can just go outand be like, I'm going to create
my new AI for my claim.Right, you can't do that because it
just requires so much money, becausethere's so much time and energy need to
be put into it, that wedo need to rely on these larger companies
that are willing to spend tends,not hundreds of millions of dollars to make
(09:18):
a good CHET, GPT or AI, whatever you want to call it.
So they it can be helpful forclaims, which is what we're you know
we're talking about right right, SoI get these ads almost every day from
either Thompson or Lexus or a casetext or somebody else about how it can
help you as a lawyer doing researchor doing this, that or the other.
(09:41):
My own wish list is it wouldhelp me do discovery or figure out
the claim on the front end.But what are programs? What can we
use it for right now in termsof working on a claim, or we
could go back to what you mentionedabout insurers or using it for these days
your objects and so on. Thatso, just so, just so,
(10:01):
I asked chatchpt what is the statueof limitations to file workers compensation claim in
Georgia and its response was to asmy last knowledge in September twenty one,
So that's sort of when it's cutoff. Is so if things have changed in
the last two years, it won'thave that information. It says to file
in Georgia typically one year from thedata work related injury or illness. That
means you have one year from thedata of injury to the day you became
(10:24):
aware of it to file a claim. Please note the laws change and so
it's essential to consult an attorney orcheck the most update workers compensation dot dot
dot right, so it is tellingyou it mean is that correct? I
mean, I mean that is correct. So it's still it's still correct.
Nothing's changed in the past couple ofyears, so nothing has changed. But
so it has the information, butit does as I said, it says
consultant attorney. But what I thinkwhere the cutting edge is right now on
(10:48):
using AI, and it is beingused at a very high level at some
larger insurance companies, is they're ableto take the data from claims they can
take so you can take us astructured data. Some of the structured data
age, gender, location, jurisdiction. These are things that you can easily
just filter in and out. Butthen even injury can be considered structured,
(11:13):
right, and you can plug allthat in and if you have if you're
an insurance company and you have thousandsof claims, you can now run your
search and say these are the typical. And now I have a motor vehicle
accident with someone who at are brokenhumorous, they're a forty year old male
from Illinois. All right, youcan put that into your thing and you
can then look up every other claimthat you've had like that and then find
(11:35):
out what were the results of thatclaim, how much money was spent on
that claim. That's a big one. You know, reserves, as you
know, as we all know,being you know, defense attorneys, reserves
is a very important aspect of ourclient's job. And sometimes I think there
are some claims examiners who take itvery seriously and there's some that aren't.
(11:56):
But you know that the people whoare higher up the chain, they take
reserves very seriously, and being ableto accurately reserve a claim at its onset
can have a lot of cost savingsto that company in the future. So
being able to do that. Butnow what's happening is not to just have
that structured data. They can nowread the medical records of all these prior
(12:16):
claims and they can see the typeof treatment that those individuals had. So
one of the things that I wastalking about recently with a large insurance company
is their data can let them knowwhether or not someone who is in this
with this structured data and unstructured dataput it together, are they likely to
return to work? Right? Theyof the thousand claims that we have that
(12:39):
are similar, only fifteen percent havereturned to work with that type of injury,
or you know other way, ninetyfive percent have returned to work.
Now, as an attorney, it'svery nice if I can get that data
the second I get the claim.Imagine if you've got a claim and it
says our data says that someone likethis is likely not to return to work,
that will have a direct impact onhow you handle that case. Because
(13:03):
sometimes as our goal is to getthese guys or girls to return to work
the injured worker, we want themthat MMI full duty return to work.
But if we know from the getgo that is a very unlikely scenario,
well, then it will impact howI'm going to attack that claim. I
may want to now try to sellthe claim much earlier, knowing that I'm
going to have bad results. Inthe same way, medically, I think
(13:24):
it helps because if I can findout that someone who has this type of
injury and you know it can breakit off. You can say, you
know, fifteen percent end up havingan orif five percent end up having CRPS
or whatever made up pain thing thatthe doctors want to put on them.
And this many improved with injections.Getting getting data on how many work compatients
(13:46):
get this type of treatment versus nonwork compatients and seeing the difference can be
very helpful as well. And thatdata is becoming available that the insurance companies
have from the get go. Theyneed to type in the information put in
them you know whatever information that whateverdata points UH, and they can find
out a lot of information. AndI asked, what what is the most
important interative? Think what you guysknow or what do you think? What
(14:09):
is the most important data point UHto determine the expected costs of a claim
type of injury? I would saylocation, Steve has it, the jurisdiction
is going to is the number onedata point to determining cost of claim,
which when you when you think aboutit, does make sense because Illinois,
(14:31):
for example, is a very expensivestate medically and for indemnity versus our neighbor
to Indiana is a very different,you know, very different ballgame in Illinois.
We don't get a control medical inIndiana. They do in Illinois.
Impairment rating is one factor to determineyou know, PPD in Indiana, that
is the PPD whatever the impairment ratingis, so these jurisdictions have wildly different
(14:54):
rules as we all know, whichis what you know, keeps us all
doing what we do separately and havesort of no clue what's happening in state
to state. But under just thatdata point alone, it's going to have
a massive impact on how uh theinsurance companies do cases and how obviously the
attorney's hand. Yeah, well,I know, I know, you know,
(15:15):
when we get contacted about a claim, you know, the first information
we ask is it's all that gymwal contact information about you know, uh,
the age, where you know,what's the type of an injury,
all that basic information that you knowtypically in your scenario, the insurance company
would already have and then kind ofput that into their claim projection software and
(15:35):
it kind of spits it out sothat we when we get running, we've
already got a ton of information.But have you actually already seen insurance companies
using this and are is it somethingthat they're beginning to do and transition to
their attorneys? So uh none,Sadly for me, none of my insurance,
none of my clients have a robustAI system set up yet. However,
(15:56):
I do know there are many outthere I was just at a seminar
last week about this. I wasin Philadelphia. I was asking someone are
about and it was with it doesn'tmatter who the insurance company is, but
it was a very large insurance companyand they are using and I And I
asked them like, are you relayingthis information to your attorney? Because I
was sort of maybe it was thecynic in me where an INTERNCE company could
(16:18):
use it as a test in away for their for their attorneys to be
like, we know what the likelyor likeliest outcome is, you know,
are we going to let our attorneyknow that in advance or are we going
to see what they do and howwell or not well they are able to
get to that end result that wethink is most likely. Now that's a
cynic in me thinking clients are youknow, sometimes trying to you know,
(16:40):
be secretive with us. But inreality she said no, we are absolutely
given that this to our attorneys,which to me would be it would be
such a gift if I got acase in I said, oh, and
also, here's our separate AI resultsthat we've run on this that gives you
all these potential scenarios. But evenI think I always thought medically. As
a work comp attorney, you becomevery dangerous in your knowledge of orthopedic medicine.
(17:03):
I am far from a doctor.I cannot read an MRI result,
but if you tell me someone hasa fractured blank or a torn blank,
I can likely tell you what willhappen next. All the way through MMI,
now there are things that happen alongthe way that sometimes surprise me.
And those are the things I wouldlove to know in advance, to say,
how often does this gunicular injection occur? Right? That's not something we
(17:26):
see too often, but it is, or this radio radio frequency ablation happen
right for this type of this typeof pernia disc in the lumbar spine.
Those type of things would be veryhelpful for us to know, because not
only for just making the case movealong efficiently, but as a red flag
monitor that will pop off and say, hey, this thing. You know,
(17:49):
we sort of know it inherently thatI don't see this very often.
If I don't see something very often, that is a red flag. I've
been doing this forever, and Isee pretty much any kind of injury and
treatment course recommended. But to knowthe not only whether or not. I
think it's a red flag. Butwhat are the results of the claims that
we've had in the past where thistreatment occurs. Right, If I know
(18:10):
it's absolutely worthless and no one getsbetter, I'm going to fight a lot,
very hard through every method I can, through IEME, through utilization reviews.
But then, as the client says, listen, we've actually don't see
this very often, but we seereally good results with it. So maybe
it isn't the best decisions to tryand fight it. That information would be
really good to know, because Idon't entirely I don't trust treaters. I
(18:30):
don't trust pain managment doctors because knowwhat a pain managment doctor gets paid to
do. Inject and to prescribe medication. That's what they do. That's how
they make money. So you don'tsee the pain management doctor who stays in
business by being like, no,you'll be fine with conservative care. That's
not what they do. So itis incumbent on us as responsibornity to have
that cynical mind or just very aggressivestance towards this type of treatments. But
(18:55):
the more data I have, themore information I have, the better I'm
going to be to represent our clients. Yeah, and I think you know,
if insurance companies are able to provideyou that information fairly early on and
you've got to claim that maybe isyou know, not in your best interests
in terms of the claim. Istreating with the provider of their own choice,
and you know they've got this imagingthat shows something and they're recommending some
(19:15):
certain procedure. But then you getthis documentation in or this you know,
this information that shows well and thisin these certain kind of cases, you
know this procedure is not typically warranted. I see how you could take that
information to maybe mediation or negotiation potentiallydrive down the value of the claim.
So having that you know, datais very useful in certain situations. Of
course, like you mentioned, ifit's not in your favor, it's good
(19:37):
to know that too, very veryearly on. So I can see where
all this information, having it atyour fingertips is very beneficial for attorneys,
not necessarily in terms of how youpractice, but in terms of managing each
claim as it comes along. Whatare the concerns that you see with AI
for practicing attorneys, Well, thebiggest concern I think what we mentioned,
(19:57):
sure, right, the outset isjust blind trust and that what is telling
you is accurate because it's not.And so there's a good there is a
good story for those of the worryto understand. Something that did happen in
New York State. An attorney youknow, new to AI. It was
a case involving an airline where someone'sI think he got someone got hit with
(20:19):
the metal cart that they push across. Right, that was the facts of
the accident, and they were emotionto semiss was filed and an attorney went
on chat GPT and said, hey, I need a case law for this,
and it gave them case law andthey used it in a brief.
The other his opposing party, youknow, as any good attorney does,
(20:40):
was like, hey, I needto research these cases, see what they
say, see how I can attackthem, and lo and behold. Could
not find those cases. So theysay to the court judge, you know,
they file a motion to I thinkquash the brief that they filed,
and they said, hey, noneof these cases there exist. So the
attorney said, oh, don't worryabout I'll go get you the cases.
So what do they do or theysay, hey, chat, are the
(21:02):
cases you sent me real, andchat ChiPT says, yeah, the real,
of course, the real. Sothe attorney goes back and says,
none, they're real, and Ithink he's able to print out some like
citation for them, and you know, still obviously not real. So it
was funny though, because when heasked chat GBT, are these legitimate cases?
The chat GPT said yes they were, and so I don't think legal
(21:25):
research And what was funny about what'shappening now? So they're trying to discipline
the attorney who presented those fake cases, and the attorney's response is, don't
you think I've been disciplined enough?Like I am a national news story and
an embarrassment for having Jesus they doyou think I need formal sanctions? Or
is my name being plent? Youknow put And if you look at attorney
(21:47):
chat GPT, anything negative, he'sthe first research result. So I would
say that he's he served his time, time to serve should be enough and
let him let him go. Butthat is a huge red flag that you
need to be aware of, isthat it's just not gospel and maybe one
day it will be. Maybe oneday it'll take the place of a Lexus
(22:08):
or a West law that we canuse for legal research. We're just not
there yet because the industry as awhole, and I mean the AI industry
is you know, it's it's stillbusiness and they're looking at where they can
make the most money and be helpfuland and there's lots of ways. It
just hasn't trickled down yet to beingfully enmeshed in the legal field. It's
(22:30):
getting there, As I said that, there are insurance companies using it,
and there are dedicated teams at theseinsurance companies who are being looked at to
create software and and create models.And but the thing is with these models
is every time it doesn't take aday to create a model, it takes
time, a lot of time.And by the minute they're done creating a
(22:51):
model, the the ability of AIhas so increased that that model is now
out of date and that the sortof go right back to the drawing,
which is can be a very imaginefrustrating thing. These individuals are also very
exciting because I feel like they werestagnated for a very long time in the
capabilities of AI to do anything,and now it's just increasing exponentially week by
week that it is a very excitingtime for it. But I don't think,
(23:17):
you know, for litigators, Idon't think it's anything to worry about
that like it's going to put usout of a job. Far from it.
I think that you know it,it's going to be a good tool
for us long term. And Idon't I don't know if for any three
of us it will be part ofour day to day. But I can't
imagine if like my son, myseven year old son wants to become an
attorney, Uh, it will besomething that he uses on a regular basis,
(23:40):
just like when I started law school, because I'm when I started law
school, we were still researching inbooks. By the time I finished law
school, nobody ever used a book. We were all on lectus snatches,
right, I just was write inthat time period, and who does legal
research on books? But you knowanymore, right, it's just nobody.
Everyone's library has been if people keepit for the photo is behind us,
(24:00):
Like, hey, we're lawyers,we have books behind us, right,
But to show it's going to continuouslyevolve and become a much greater part of
our practice. And I think inthe way that we talked about with with
not us having the resources, butother companies having resources where they put in
a lot of data that will helphelp us understand and it can even go
(24:22):
farther right, it can go youknow, in Illinois, the judges are
arbitrators, And what if I wasable to quickly go through an arbitrators decisions
like the one like the issues I'mhaving, or the decisions with regarding awards
like the one I have. Right, there's still the commission, there's appeals,
but I would love to know howdoes this judge do on a regular
basis over the course of the lasttwo hundred decisions on an accident defense.
(24:45):
Now I could even slim it downto my accident defense, but do they
ever find accident defense is good ornot? Like that would definitely color my
decision to our my client to belike, hey, our likelio to success
is blank? Right to offer?People are saying fifty fifty years. It's
one of the things I hate inlife. And when an attorney, when
I have a client call me andsays they say, like an attorney told
me it was a fifty to fiftychance, Like, what do you mean
(25:07):
a fifty fifty chance? There's nosuch thing as a fifty thfty chance you
have to give real advice. Butyou know, the more data we get
will make it easier to give ourclients more accurate and well informed advice.
It would be helpful if we knewwhich judge or which arbitrator was citing what
case is the most in certain instances, you know, if you have a
certain type of defense, it wouldbe great to know exactly what they rely
(25:30):
on and what they find credible.Right, and again that all this information
has to be public because they havethey have to find a way to be
able to get it. So likein Ellinois, all the Ellinois Work Commission
decisions which is the appeal level,all those are public. They print them
all out, but you'd have toliterally read through it, you know,
for me, I'd have to readthrough every single decision and do it all
myself. But if I can justgive it to a machine and say,
(25:52):
hey, you know, I cantell what I wanted to learn, you
know, would learn by judge,learn by these issues, and tell me
what happens, you can sort ofdo that. That those capabilities are available
now, but you just need tohave the right you know, engineers to
create it. And I think iflawyers, we don't have that. That's
not you know, we're lawyers.We're not here. Computer technological skills is
(26:17):
far from our where we shine.And it's funny when you talk to a
lot of lawyers, like the lackof technology knowledge that they have is sometimes
embarrassing, especially for some of theyou know what, I consider that the
generation before me, like people stillbuild by hand, right, they still
(26:37):
write it down. They won't enterit in the computer. There's a lot
of that, and just imagine tryingto get them to use an AI or
a chet GVT function. You knowit's going to be nearly impossible. So,
you know, some ethical codes insome states even have a requirement that
attorneys be aware of and learn newtechnologies. I don't know that that involves
AI. Yeah, but I don'tknow if it's done here. Another pitfall
(27:02):
potentially, and you know it's somebodyto think about, are just like hippo
laws. Are you giving people's medicalrecords to third parties? Right? You
know what what access are they allowedto have? Now? You could maybe
get rid of identifying materials like oneof the things that we actually have to
do in Illinois. If I'm submittingan exhibit to the arbitrator, I have
(27:23):
to remove all social Security number information, right that has to be blacked out.
They can't have that because of hippolaws, and you know, we all
know that hackers and that stuff becomesa real concern. And now with these
medical records for you know, millionsof people are now being handed over third
parties. That that's a that's amajor concern, not just for like,
(27:44):
not not for US litigators, butjust for the public because I don't think
that it would would be a goodthing for us to be doing, to
be handing over people's medical records withoutyou know, consent or just making sure
that it's protected in a way thatwill you know, make sure that they
don't have to deal with any unintendedconcerts, right right, So to go
back to the gentleman in New York, I know, in response to that,
(28:07):
I heard on another podcast a judgein Texas is issuing a standing order
that no attorney can use AI inany pleading unless they get specific permission to
do so. On the front end, I thought that was fascinating as a
way to just have a standing orderfor all attorneys in all cases. Yeah,
(28:27):
I mean, I guess it's notsurprising to some degree given how new
it is, and they want toavoid the issue that happened in New York.
Now. I think like a blankettime people that you can't use it,
but like you shouldn't cite, butlike you shouldn't cite to AI.
Right, like there is if you'reif you're writing a brief, it's case
law. Case law is readily availableto attorneys. It's not something that you
(28:48):
know, you need to really gocreate or ask the CHANGPT. You should
be able to find yourself. Butit's not surprising, and you know that
that is sort of the step peopleare doing as they see people making mistakes
that you know, that's let's letsome other jurisdiction figure it out first,
and then you know, we'll adoptthat. I will say, the legal
field is not necessarily always you know, the first one out there to try
(29:10):
something new. I'm happy with howthings happened with using technology as far as
remote court hearings, but I thinkthere was a lot of you know,
fight back, you know, fightingfrom different sides of the bar about even
just doing remote court hearings right thatthat that was a big deal, and
there are some concerns that do processconcerns and criminal cases that don't necessarily happen
in a pre trial hearing for workcount case that you know we deal with.
(29:32):
But so, yeah, so whatabout fraud detection? What do you
see h AI being usedful for interms of workers compensation fraud detection? I
think I think it's the same thing. So one of the other data points
that uh I was told is abig indication of fraud is date reported.
(29:53):
So if they know that you reportedthe case four days after the accident,
those simple things is a big redflag for them. And it's going to
allow and just on its own willwill be a good red flag for the
insurance companies and to find out whetheror not someone is committing fraud and in
work there's to me a lot ofdifferent type of fraud. You know.
There's there's the initial fraud of theobvious that they're claiming an accident happened that
(30:18):
never happened. And I I thinkwe've all had our fair share of those
type of cases that people are juststraight ling nothing happened to them, They
got hurt over the weekend, orthey're not hurt whatever. That's the that's
the obvious fraud, But I thinkwhere the fraud happens more in the every
day that we deal with is ofthe lingerers and the symptom exaggerators where they're
(30:38):
you know, the statistics are prettyclear about the results from a work compatient
compared to the general public, andhow much worse results a work compatient will
have than than the general public forpretty much any injury, and so AI
being able to spot those for whatever, you know, one reason being the
day reported, and you know,other reasons potentially being how much pain they're
(31:00):
complaining of. Because again, whatthe biggest difference now between what AI is
able to do now that it wasunable to do before is to understand context.
It can read a sentence and understandwhat it means, even if you're
using right the same word. Itcan understand the context. One of the
things I, you know, whatI've been doing my whole career is I've
been reading through physical therapy records,you know, combing through them, because
(31:22):
I think for me, physical therapyrecords is a great place to find inconsistent
statements and inconsistent exam results because forsome reason, people love their physical therapists
and they feel like it's a therapysession of a psychological therapy session versus a
physical therapy session. And they'll tellthem how, you know, over the
weekend, I, you know,fell down, I was in a car
(31:42):
accident or whatever it is. They'llbe very honest with them, or they'll
say I'm feeling great, and thenthey'll go to the doctor two days later
and be like, I'm feeling terrible. Ten to ten pain straight leg ray
is positive even though it was negativehere. So I mentioned you count have
AI do all that for you andcan be done within seconds and just point
out you petitioner said pain was threefour or five ten, and just list
(32:04):
it and you'd be like, oh, this is something something's wrong here,
just right off the bat, something'snot right. So I think for fraud
detection it is a it will bea very useful tool and trying to put
together, you know, a bunchof data points for you, whereas you're
only able to maybe focus on afew at a time because you're human versus
the machine. Can you know,look at sort of the whole thing is
(32:27):
as a combined and find out theseyou know, something's not right here,
red flag it for you and thatwill affect again how you handle the claim,
how your client wants you to handlethe claim, and what the expectations
are regarding the results of the claim. Yeah. Well, just thinking forward
in terms of the future, it'sreassuring, and you've kind of confirmed my
thoughts that I don't really see AItaking over the role of a lawyer.
(32:52):
It can certainly be very useful fordata collection and using the data points to
project a claim. As long asyou are using AI, you know,
with some relatively common sense not torely on it for terms of case law
and things like that, it canbe a very useful tool. Certainly.
We also discuss the concerns that wehave with you know, using other people's
(33:14):
information. I feel that insurance companiesyou know, can extract that data on
their own really, you know,in terms of the results of the case,
without really any type of approval fromfrom the patients, as long as
they aren't using the patients you know, identify identifying information. Are there any
other concerns or benefits that you seeother than what we've discussed as we're kind
(33:34):
of wrapping it up here today,I mean, as far as I think
we've hit the concerns pretty well,I think anyone who's just thinking about this
abstractly will come to the same realizationthat we are discussing here about what the
concerns are. I think there's there'sprivacy concerns, and there's reliability concerns.
Those are I think the major ones, and I think for what the pros
are, I think that the differenceis and it may be true also on
(33:59):
the con side is that we don'tknow yet. That's the you know,
this stuff is still growing at arate that is, you know, unlike
something we've seen in technology in avery long time. So we can try
and you know, look at thecrystal ball and to see what here's where
we think it can go. Butoftentimes when we look at technology, you
know, did anyone foresee when theiPod kind came out that the that the
(34:20):
iPhone was going to revolutionize the worldthe way it did like when we had
of course not, and that wasonly a couple of years away. So
I think that we have ideas asto things they can do, and I
think for attorneys it will be alargely a time saving tool for us and
and give us a lot of additionalinformation, which is also all great,
but I don't think we're going tobe displaced, you know, legal zoom
(34:42):
has been around for a long time. Our wills and will in the state's
attorneys out of a job, noright, our contract attorneys out of a
job. No, not maybe someare. But it's also sort of like
in a way like that stuff hasdemocratized law, which is a good thing.
I think it gives people the accessto hopefully good you know, good
work product without having to pace andyou know, so now everyone can afford
(35:04):
the right attorneys, and this willgive you know, the public knowledge which
sometimes as the attorneys, we don'tlove it when the public health knowledge is
the you know, and a foreignclient can be a dangerous one. They
think they know more than they do. But that's sort of for us and
our jobs, but and the wholeyou know, I'm still of the opinion.
And there's been a lot of doomand gloom about where AI is going,
the end of humanity. You hearabout you know that it's going to
(35:27):
put everyone out of a job.You know, I don't know if everyone
remembers when and this is like notAI but just technology. Remember when they
had the self served counters at thegrocery store, that all the workers at
the grocery store were going to beout of a job. Obviously, that
didn't come to fruition, right.So I think that when when these new
technologies come aboard, we're we sortof are fearful of them, and I
think part of that ends up,you know, damping them down, because
(35:51):
the public doesn't want something that's goingto take over everything. They want it
to just be a convenience for themost part. And so yeah, I
think the insurance company is going touse it for a good purpose and is
helping move claims long efficiently because again, costs, what are these insurance companies.
A lot of insurance compans care about. They care about costs, right,
keeping costs down. And I thinkthat's what AI is going to help
them do. And I think thatthat's gonna be good for everyone, right,
(36:15):
you know, trickle down, saywhat we want to trickle down economics
on a glow on a big level, But if a company that you work
for is able to save some money, then hopefully that will lead to reduced
premiums and and better results for everybody. That that would be the hope.
And that's an optimistic way of lookingat it. Yeah. Well, Michael,
I want to thank you again forgreeing to join us on the podcast.
(36:37):
Steve and I have been wanting todiscuss this topic for some time,
and we really appreciate your insight.Our listeners can't see us, but Michael's
dressed way nicer than us today,so he's he's out to do some real
lawyering today, so we'll let himget to it. But where can our
listeners find you if they want toget in touch with you or learn more
about AI and workers compensation claims?Well, thank you guys again. Thank
you both so much for having meon. It was great talking to you
(36:58):
guys. I do Illinois's compensation.My firm to sort of Illinois everything GL
Construction. You can find me atDL Hyphenfirm dot com. Is our address.
My name again, Michael Milstein.My phone number Casey went out reach
out three one two three two sevenzero zero four two and then my email
is Mmilstein at Dlhyphenfirm dot com.If you ever have any questions in Illinois
(37:21):
work com please do not hesitate toreach out to me, text, email,
pigeon whatever you need. I amalways available what I I always say.
What I try to differentiate myself onis you know, the customer service
that I provide is not the sameas what everyone else does. So you
can always reach me at any timeand I hope to hear from someone soon.
Thank you, Michael. I appreciateit very much. Thanks Michael.
(37:42):
Thank you every great day.
Popular Podcasts
24/7 News: The Latest
The latest news in 4 minutes updated every hour, every day.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
The Clay Travis and Buck Sexton Show
The Clay Travis and Buck Sexton Show. Clay Travis and Buck Sexton tackle the biggest stories in news, politics and current events with intelligence and humor. From the border crisis, to the madness of cancel culture and far-left missteps, Clay and Buck guide listeners through the latest headlines and hot topics with fun and entertaining conversations and opinions.