Analyst Talk - Dr. Andreas Olligschlaeger - The Futurist Analyst - Leapodcasts

Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Welcome to Analyst Talk with Jason Elder.

(00:01):
It's like coffee with an analyst,or it could be whiskey with an
analyst reading a spreadsheet,linking crime events, identifying a
series, and getting the latest scoopon association news and training.
So please don't beat that analyst andjoin us as we define the law enforcement
analysis profession one episode at time.
Thank you for joining me.
I hope many aspects ofyour life are progressing.
My name is Jason Elder, and todayour guest has 35 years of law

(00:26):
enforcement analysis experience.
He is a data scienceteam lead with I three.
He is also the former president ofthe Society of Police Futurists.
He holds a PhD fromCarnegie Mellon University.
He grew up in Germany, but he foundhis home in Pittsburgh, Pennsylvania.

(00:46):
Please welcome Dr. Andreas Ole Schlager.
How we doing?
I'm doing just fine man.
How is Pittsburgh these days?
Cold.
Cold and miserable.
In other words, typicalPittsburgh fall weather.
I was just talking with Mike Winslow.
He was from Arizona I'min living in Florida.

(01:08):
He is in Arizona and we're talkingabout it just being slightly chilly
with like bright sunny weather and I waslike, everybody in the north hates us.
'cause I remember my days inPennsylvania where it just, you basically
didn't see the sun for four months.
Oh yeah.
That's about true.
Oh man.
So I got a lot to go over today.
Really looking forward to yourperspective getting your contributions,

(01:31):
going over your 35 years of experiencegetting all your accomplishments and
obviously talking about data 'causethat's so important for analysts.
So,
absolutely.
Alright, let's start fromthe beginning though.
How did you discover the lawenforcement analysis profession?
Absolute fluke.
So I was an economic geographer.

(01:54):
And one of my areas of expertise wasspatial modeling for econometrics.
And so from a very early,early period, I got into GIS.
And this was actually during myundergraduate years when I was at
Concordia University of Montreal.
And I learned GIS programming in fortran.

(02:15):
On a cyber CDC mainframe.
And yeah, I mean, those were the,those were the very, very early days.
And my claim to fame there was, Iactually caught a printer on fire
because I somehow put an endless loop.
I was doing a chlor plath map, and youhave your punch cards and you submit those
to the computing center, they run it andhalf hour later you pick up your map.

(02:39):
Right?
Well, I get a phone callsaying our printer's on fire.
You had an endless loop in yourcode and it kept overriding
and overriding and overriding.
So, oh my gosh.
Yeah, that was pretty funny.
So anyhow so I got into GIS andthen I went to University of British
Columbia, got my first master'sdegree, and then I came to CMU Carnegie

(03:01):
Mellon and worked on my PhD there,which was in public policy and also
oriented towards economic geography.
And then my advisor all of a suddenleft for Australia and I was kind
of stranded without an advisor.
So, will Gore became my advisor andhe's probably well known to a lot
of the older crime analysts 'causehe did a lot in crime mapping.

(03:22):
He asked me one day, Hey, we got thisgrant from the Justice Department National
Institute of Justice, and it's to explorethe use of GIS in drug market analysis.
And would you be interested in the job?
I said, okay, well what I need to do?
Well, he has, he said, it has one catch.
You gotta be embedded full-time inthe undercover narcotics division
in the city of Pittsburgh Pittsburghpd. And I said, sure, I'll do it.

(03:46):
And that's how it happened.
That's how I got into law enforcementanalysis, and I loved it so much.
I mean, that, that's what I'd been doing.
Ever since.
So when you were, when you were firststarting out with your education, w right.
, What was your dream or what was,what were you pursuing prior?
To this,
I was gonna be aprofessor that was mm-hmm.
My goal and mm-hmm.

(04:07):
I did wind up teaching at CMU graduate.
I taught grad school for about10, 11 years, something like that.
Mm-hmm.
Two graduate courses a semester.
And and it was fun.
I really, really enjoyed teaching.
What I did not enjoy wasthe university politics.
Oh.
And that kind of drove meinto not pursuing that career.

(04:28):
Plus, you know lo loving the thepractical side of law enforcement as
well as the analytical side and kindof merging that was my experience in
academia and actually being able to makea, a difference in the real world was.
To me more attractive at the timethan just a regular teaching job.
Even though teaching is probably isone of the most fun things I've ever

(04:52):
done, and I still keep in touch withstudents from like 30 some years ago.
Hmm.
All right, well, let's talk aboutyour your coming into the office
first time at Pittsburgh pd and you'regetting yourself established there.
Let's just take us back to thatexperience and what you remember and just.
Oh boy.
So here I was, this young college kid.

(05:14):
I was, this was in 89, so Iwas, I was about 28 years old.
And you know everybody waslike, WTF, who's this kid?
What the hell is he doing here?
And so it was, I got mercilesslyteased for the last first two,
three months that I was there.
And it was during the time whenPittsburgh pd, who, who at the time

(05:38):
and to a degree still today is, is not.
Known to be one of the mostprogressive police departments.
But so it was a little bit behind thetimes say compared to New York City or
other cities in the use of computers, theyhad just gotten their first electronic
records management system about a yearor so earlier, which didn't go over well.

(05:58):
There was actually a story of a,of a cop shooting the monitor.
He was so frustrated with it.
, The monitors had to be pretty big though back then.
Right?
Back.
Back then they were, no, they were,they were like a 12 inch monitors, 13
inch monitor or something like that.
Typical green screen kind of thing.
Yeah.
So yeah, and, and I had at thetime was probably one of the.

(06:20):
Best workstations you could get.
It was a a, a Sun Spark station and I wasone of the first, I think 100 users of ARC
Info which was an ESRI product back then.
And it used a database calledInfo and the programming language
was a scripting language.
It was called arc Macro Language.

(06:41):
And so I started working there.
I started incorporating data fromthe police records management
system from 9 1 1 Center as well as.
The county tax property tax database allinto one single system to try to use get
as much data as possible to let policemake informed decisions about whatever

(07:03):
it is that they were trying to do.
And that also then turned out to bethe basis for my PhD which was to
use artificial neural networks touse space time forecasting of where
open street drug markets moved.
And so that was prettyunique or new for its day.

(07:23):
In fact, my entire PhD advisorcommittee told me I was freaking nuts
trying to do it, but it worked out.
And I, it was probably one of thefirst examples of predictive policing.
Mm-hmm.
And definitely one of thefirst examples of using ai.
In law enforcement and in predictiveand in predictive policing.

(07:46):
So, so was, did codingcome natural to you?
It
kind of did.
I took a, a class in Fortranwhen I was an undergrad.
Mm-hmm.
And that made me as, back then,especially in geography departments,
one of the only one or two graduatestudents that knew how to code.
Which got me a lot of researchassistantships and a lot of

(08:08):
grant work things like that.
And then when I came to CMU.
I learned variety of other languages.
Most of 'em are dead now.
Pascal and c C is still used, obviously.
Then later I got into Python,Java more recently when I
say recently, last 25 years.
So yeah, it, it kind ofcame naturally to me.

(08:29):
I'm, I'm, when it comes to programming,I'm actually mostly self-taught.
Mm-hmm.
And I actually wound up teachinggraduate school for computer programming,
database design that kind of thing.
Yeah, so, they were just startingwith their records management system.
Right.
And so the data that you'reanalyzing at this time mm-hmm.

(08:51):
Is it, just describe that.
Is it, I mean, I can't imagine thatit's very clean and maybe you're getting
some computer data, sometimes you'regetting maybe some reports . I'm sure
you're juggling multiple data sources
Oh, absolutely.
In different formats.
Right.
So the, the 9 1 1 data wasthe easiest 'cause it was in
Oracle relational database.

(09:12):
The police records management systemwas in a proprietary, database.
So it was a real challengeto get data out of there.
Mm-hmm.
I, it was initially delivered onfloppy diss and eventually on tapes.
Back then there was nonetwork in the city.
Right.
You didn't have, you couldn't justlog into Oracle and, and transfer

(09:33):
something to your, to your workstation.
And.
The county data, I forget what it was in.
No, they had it in info, right?
They had it in our info, so that was easy.
But it's still, since we we hada very rudimentary network, so
all the GIS workstations in thecity were actually networked.
But the other ones weren't.

(09:53):
It was very challenging to, you knowmove data around, but we got it done.
And I would get like weekly updatesany new 9 1 1 calls for service.
New police reports, arrestreports any property tax changes.
That data was only updated onceevery six months or a year or so.
And then I took all thosedifferent data sources.

(10:15):
I designed one relational database, whichwas an info, and that was put in there.
And it was then I developed a GUIfor GIS where you could access
and analyze all of that data.
You talked about getting pickedon in the first couple months.
Mm-hmm.
Was there a point in time when I'm like,oh, this guy's actually gonna help us.
He's not.
There was
there was it was after about sixmonths, so I had fortunately some

(10:41):
of the investigators took me undertheir wing and I can't tell you
how many hundreds of hours I spent.
On the streets at night, you know workingalong with investigators, helping with
investigations, and they very quicklyfound that, you know I could do things
that they couldn't do by themselves usingcomputer science and data analytics.

(11:03):
And so one of the first things I didwas I took all the nine one one call for
service data, the arrest reports, overlaidthem and showed them that where they were
arresting people and where 9 1 1 callswere coming in for open air drug dealing.
Were not always in the same place.
, The seasoned one seasoned guy said,ah, he's full of it, you know?

(11:25):
And in fact, one set off a bunch offirecrackers under one of the maps I
had just made, burnt the crap out of it.
And who later becamegood friend of mine and.
So, but then I said,look, I'll challenge you.
Let's take some undercover cars.
Let's go out at like one or two o'clockin the morning and let's see if I'm right.
And we did, and I was right.

(11:45):
And they went like, holy crap wehad no idea this, this was here.
And so I tried to explain to them theidea of precursors so for example, it may
not be an established drug market yet,but you gotta interpret the 9 1 1 data.
So for example somebodyliving in an area that.
Doesn't know that whatdrug dealing looks like.
Right.

(12:05):
What they see is a bunch of kidshanging out on the corner, or
people hanging out in the corner.
They're drinking out in public,they're smoking, and there's cars
constantly pulling up and they'retalking to them through the window
and they see it more like a nuisance.
They don't see it for drug dealing ordidn't see it for drug dealing back then.
Mm-hmm.
And I quickly realized that'swhat's, that's what was going on.

(12:29):
And, and, and I cannot stress how valuableall that time spent on the street was
with respect to me learning all about thedrug trade, how it works, what the mos are
the kind of things science look out for.
And, and, yeah.
Then they quickly realized, yeah, I mean.
This guy's, this kid's helping us.

(12:49):
And from then on it was they actuallytook me along on raids because I could
do things that a regular police officercouldn't didn't have to read anybody
their rights or anything like that.
Just talk to 'em.
Say we would go into a nuisancebar, I'd just mingle, talk to
people, things like that, you know?
Mm-hmm.

(13:09):
Get some information that kind of thing.
Mm-hmm.
All, nothing major, nothing.
Nefarious just information collecting.
And I learned so much doing that as well.
Which was very, very important as well.
And really, I couldn't be where I am todayif I hadn't been for that experience.
It allowed me to learn how law enforcementthinks, how investigations work, the

(13:32):
kinds of things that they look for.
And it really informedmy research as well.
It showed me which direct Ishould go with respect to.
What could I do with a limited time thatI had to maximize the impact I could
have in practical law enforcement in,in this instance, on drug enforcement.
Yeah.
And, and here in this time, it's notlike you have a computer on you, right.

(13:54):
No, I mean, you're, getting morehuman intelligence and, knowledge.
Correct.
And I, I think, I think, I thinkin a way it's, it's unfortunate
because I think now many analystsmight not be given that opportunity.
'cause okay, if you, if you arecoming with you, us, you're gonna
have a computer and your yournose is gonna be in the computer.

(14:16):
It's not necess, it may not beout doing some of the, knowledge
transfer that you just described.
Exactly.
Exactly.
It's, it's kind of similar to goingto a doctor's office and the doctor
has his nose and a laptop and askyou maybe two questions, grunts
every second question, and thenyou're done in five minutes, right?
Mm-hmm.
No, this, this was hours and hours,weeks and months spent on the streets

(14:40):
not full-time obviously, but here andthere during the week and just taking
it all in and being totally immersed in.
Undercover narcotics and mm-hmm.
That really taught me that it isabsolutely vital for analysts, data
scientists, researchers from academiaand law enforcement to work together.

(15:02):
You cannot work in a vacuum.
Mm-hmm.
Because you put a computer scientist,an academic, and a cop into a room,
they all speak different languages.
They'll just talk past each other.
Mm-hmm.
They don't understand each other.
You have to actively geteveryone involved in a project.
Let's say you're, you're startinga new project in a city, some kind

(15:24):
of law enforcement related project.
You've got the stakeholders, whichis practical law enforcement.
You've got the people thatimplement it, which is , say a
contractor, computer scientists,software engineer, what have you.
You have analysts as well, andthen you've got often academia,
which has like a theoretical input.

(15:45):
And they all have to work together.
If an academic, I've, I've met so manybrilliant academics that have great
ideas, but they have no clue how toimplement them in the real world.
Mm-hmm.
And it's a shame.
So much good research has gonewasted because people didn't know
how to talk to police officer.
And by the same token policeofficers have historically.

(16:07):
Viewed academics as these pointahead people who have no idea
what they're talking about.
And then you have the computer scientiststhe contractors, whomever, they go,
oh crap, I'm caught in the middle.
You know?
Mm-hmm.
I've gotta please the cop, but I'vealso got to make sure that whatever
the academic says that I implementthat, whether it makes sense or not.

(16:31):
Yeah.
So, because that's whatthe grant said, you know?
Yeah.
And it doesn't always happen that way.
And it's not as bad as it used tobe, but I still see it happening.
Hmm.
So then you're.
Developing your knowledge,skills, and abilities.
Mm-hmm.
As as you're workingthrough this time period.

(16:52):
So how did you get from this point tohaving the idea for your PhD? It evolved,
right?
Mm-hmm.
So again, it it's, the idea was actuallyborn that something had to be done
and that something could be done.
I knew from my basic statisticscourses, predictive analytics
courses, things like that.

(17:13):
And then a big change in my life was abook called The Dancing Woo e Masters.
It's now very outdated, very old.
But it got me into theidea of neural nets.
And I thought, huh, see, the, theproblem at the time was that the kinds
of methodologies, algorithms, what haveyou, people were using were all linear.

(17:34):
Because if you look at spacetime, it's a non-linear problem.
Have you ever heard of the Game of Life?
Yes.
Yes.
So Game of Life is a prime example.
So the Game of Life, you have a randomlypopulated chess board, and the rules
are something along the lines of.
If you're alive, in other words,you have a checker and you're
surrounded by nobody, then you die.

(17:57):
So the next iteration of that board,that position is no longer occupied.
If you have exactly two neighbors,then a new checker is born.
And if you have three or moreneighbors, you get overcrowded and die.
Right?
So this.
This is a, I think John Conway who was amathematician presented this game first,
had the idea, but nobody had been able to,with linear methods to design a technique

(18:23):
that was accurately able to predict atinfinite every successive checkerboard.
Based on the rules.
So one of the first things Idid, I designed a neural network.
And at the time there wasn't anythingcalled open source or anything.
Everything had to beprogrammed from scratch, right?
So I developed a neural network Backthen, it was the latest, one of the

(18:44):
latest things was fee forward networkswith by propagation with a hidden layer.
I use spatial and temporal lags as,as the inputs for the input layer.
And the output layer was a predictorfor a chess piece or a square on the
checker board, whether it was gonnabe dead or alive, the next generation,
next iteration based on the previous.

(19:05):
And so I trained it, and it worked.
It was able to do that.
At Infinitem with ahundred percent accuracy.
So now that I had a proof of conceptI applied it to the data that I was
collecting in the Pittsburgh pd.And so you have leading indicators.
One of the examples I just showed youwas like 9 1 1 calls for service or

(19:28):
unruly crowd at a street corner, makingnoise, drinking in public kind of thing.
I also had indicators ofpolice activity, right?
So if an open air drug market hadbeen raided repeatedly, say for a
week they used to do that using whatthey used to call the, the impact
squad to try to eradicate that market.
Well then, geographic distillationkicks into effect, right?

(19:51):
So it's, think about like, you're steppingin a puddle, you're displacing a lot of
water but that water has to go somewhere.
So then the next question would be,well, where's that water gonna go?
And so then you start looking atattractors for open air drug markets.
So those might be areas thatare industrial have a lot

(20:13):
of abandoned properties.
And I had that data fromthe property tax data.
So I put all of this stuff together.
And we were able to predict withover 80% accuracy where a new street
level drug market was gonna pop up.
Hmm.
And
did you also do time as well in that?
Yes.
Mm-hmm.
Yep.

(20:33):
Alright.
And space time.
So just like the, think of a successivegeneration of the game of life.
Of the checkerboard.
That's time.
And I basically divided thecity of Pittsburgh into it.
Checkerboard had data points for eachposition on that checkerboard, and then
successive iterations of time and space.
Hmm.

(20:54):
I'm, I'm laughing a little bit because ifanybody's ever been to Pittsburgh, they
know that it's, it's definitely not grid.
So no, it's not.
Three rights will not putyou back on the same street.
Exactly.
And
especially back then everybody usedto steal street signs and it was
frustrating for me because I neverhad a freaking clue where I was
when I first came to Pittsburgh.

(21:16):
Yeah.
So was most of the time when you weretalking about the new drug market,
you talked about displacement.
Was it because of displacement or didyou, was there a difference between like
popping up just naturally and poppingup because of combination of things?
Right.
So when you, when you talk aboutdisplacement displacement of crime mm-hmm.

(21:38):
Back then you haddisplacement geographically.
Right.
So they would just move somewhere else.
You had displacement in time.
So okay, they keep popping usat two o'clock in the morning.
Let's try two in the afternoon.
Right.
And their, their customersknew where they were gonna be.
And they couldn't always movevery far because then they
would lose their customers.

(21:58):
Right.
And then when that didn't work, wefound a shift in mo so that what they
would do is either sell out of theirhouse or they had a concept of like
the equivalent of DoorDash today.
They would call in, they woulddeliver the drugs, they would have.
They would pay little kids on bikesto be lookouts with walkie talkies.

(22:20):
They would say 5 0 5 oh whenever theysaw either a cop car or one of the
undercover cars, which they would,every time we got a new car, it was
like burned within like two weeks,three weeks everybody knew the vehicle.
So, so, yeah.
And then today, of course, inthe world of the internet, you
also have virtual displacement.
So what was the drug of choice?

(22:41):
Did it, did it.
Did the drug impact the model?
No, no.
Back then open air drug marketswere primarily crack and heroin.
Mm-hmm.
Pot was, that was more of a social drug.
Crack didn't exist back then, or crystal Imean, crystal meth didn't exist back then.
It was mainly crack,heroin, and, and cocaine.

(23:03):
Those were the three main drugs.
Hmm.
And so what were, did you ever discover,you, you talked about the, the a was,
I guess, was there a scenarios whereit was like, oh, it's predicting
here, but there's, there was somethingthere that is like, oh, it's not most
likely not going to happen there.
Or I guess what were some.
Competing factors that youhad to, to, to account for?

(23:27):
None that I can think ofoff the top of my head.
You gotta remember bytoday's standards, right?
It was a very simple data set.
Yeah.
Basically 9 1 1 data city property taxdata and, and property data, period.
Like land use, that kind of stuff.
And records management's data.
So the data was not as nuancedas data sets that you have today.

(23:48):
Mm-hmm.
It was.
Pretty noisy.
But neural nets are great at handlingnoise compared to say regression or some
of your regular statistical techniques.
So, no there really weren'tany kind of measurable factors.
And, and the main reason is that unlike,say, regression or anything like that
with neural networks, especially at thetime there is, is no explanatory value.

(24:12):
Right.
That was one of the big critiquesI had in my dissertation.
It was that, well, what is the statisticalsignificance or con relative contribution
of this factor to the predictive value.
I wasn't able to answer thatquestion because with neural networks
back then you couldn't do it.
And my response was, look, for lawenforcement, the goal is accuracy.

(24:34):
They don't necessarily want to knowanything about statist significance or,
yeah, this, this variable has X percentor contributes x percent to the predictor.
They don't care about that.
All they care about is for somekind of a model, tell them that
something they don't know and thatit's accurate at the same time.

(24:55):
And that's, that's what Iwas trying to accomplish.
Hmm.
Because there was more of thatpractical aspect as you went.
Mm-hmm.
For your defense, was that somethingthat, a hurdle that you had to overcome?
It was
initially.
Mm-hmm.
But by that time , I hadspoken or presented at so
many different conferences.
I had been, geez, from, I wannasay like 91, 92 through like 94, I

(25:18):
was constantly being interviewed inlocal, national, international media.
My techniques were already verywell known and it really I mean,
I'm just guessing, but it, itprobably really helped CMU image.
Right.
When you're in the news fora positive reason, you know?
Yeah.
That's not a bad thing.
Yeah.
And I already beenpublished a couple of times.

(25:39):
Mm-hmm.
And they, it finallygot to the point where.
Look, Ollie, you've been here 11 years,why don't you staple all your papers
together and call it a dissertation andthat, and that's basically what it was.
I mean, you gotta remember back thenI was working full-time at Pittsburgh
PD teaching two graduate coursesand working on my dissertation.

(25:59):
I think I got maybe four hoursof sleep at night on average.
So it's it wasn't easy,but got through it.
Very good.
So then where do you gofrom Carnegie Mellon?
After I was done with Pittsburgh pdand which was around the time I got
my PhD, which was in 97, I startedworking for the Robotics Institute on a

(26:22):
something called project Gena, which was.
The precursor to the totalinformation awareness project.
I worked at the Robotics Institutefor a year, and then I got hired
by the School of Computer Scienceas a research associate, basically
working on the same project.
And so I did that fora total of three years.
And that was a fascinating project.

(26:44):
Not only because, or not least becausethe project manager, the program
director back then was none otherthan a John Poindexter, Admiral John
Poindexter, and who's a great guy.
We got along great becausewe're both sailors.
During breaks, we'd go out and talkabout bottom paint have a smoke.
It was fun.
Learned a lot doing that.
That's when I started learningabout natural language

(27:07):
processing, speech recognition.
Those kinds of things that Iwould some point later then go
on and use on other projects.
So after that was over in 2000 mycoworker and I at the, by that time
Will Gore, we worked on a project forthe A TF that lasted about four years.
We trained regional analystsand regional crime gun centers.

(27:30):
We did some geocoding work for them.
Some GIS work did a, something calleda universal geocoding repository
where no matter what the data sourcewas, there was a process that would
intercept an address and other metadata.
It would then geocode it andwithin seconds, in almost real
time you could map it out.
And that had not existed at the time.

(27:51):
This was just around the time that.
People were starting to be networked.
Yeah, that makes, thatmakes me cry a little bit.
'cause I, it was around that time, 2000ish, early, early aughts that I was
spending a lot of time geocoding manually.
Yeah.
So how was your transition froma local PD to federal agency?

(28:15):
It was pretty
seamless.
Mm-hmm.
I the thing, the thing was, and itwasn't just with the a TF but it was also
with other agencies that I worked with.
After that, the fact that I had workedwith Pittsburgh PD at that time for
about, see, oh, about seven, eight years.
It gave me street creds.

(28:36):
Yeah.
Because I wasn't just a pointyhead kid from academia or a
professor or something like that.
Coming in trying to tell seasonedagents and professionals, you know
what to do and how to do their work.
So the very important thing was,one of the first things I did,
you'd have to do is just listen.
Just sit there, listen, observetake in what they're doing.

(28:57):
And I find that's alwaysthe best approach.
Don't go in there guns blazing thinkingyou know everything 'cause you don't.
Mm-hmm it doesn't matter what job itis, whether it's a, even within the
same agency, every time you have a newproject, you have something to learn.
And do that first before youmake any recommendations.
Alright, well good.
And then it's around we're gettinginto the time where you get into

(29:20):
the Society of Police futurists.
Yeah.
That was around 2003.
So I became a member of theSociety of Police Futurists.
Around 2002, 2003.
And they had a very active list serve.
And I was a regular contributor.
And then one day I got asked tojoin a conference of what was a

(29:44):
collaboration between Society ofPolice, futs International and the
Behavioral Sciences Unit at the FBIcalled the police the Futures working
group, and I attended the conference.
I remember it was in Phillyand there it was fantastic.
I mean, this was right up my alley.
They were FU Futures research is about,in part, about looking at what technology

(30:06):
is coming down the pike five yearsdown the road, 10 years down the road.
And then trying to figure out howcan law enforcement use that to
technology to their advantage, as wellas trying to predict how criminals
are gonna use that technology.
A prime example is when theinternet first came out.
Mm-hmm.

(30:27):
Law enforcement got caughtwith its pants down.
Right.
OPH files that were even just alittle bit technologically savvy.
We're having a field day.
Law enforcement had no ideaabout computer forensics.
They had no tools.
Mm-hmm.
They didn't know how tonavigate the internet.
They didn't know anything about computers.
And so it took a good three years atleast, until law enforcement was even

(30:50):
in a position to conduct internetinvestigations, minus a couple of people.
And there always are those kinds of peoplethat were hobbyists, computer hobbyists,
and they knew exactly what was going onand how to conduct these investigations.
But they were few and far between.
Yeah.
'Cause and it go, itgoes a long way to there.

(31:11):
'cause not only do you have theinvestigation portion, you have
the, the prosecutors and thenthe juries and all all of that.
That has to be brought up to speed.
Absolutely.
Ab and that's true to this day.
Right.
I mean, yeah, if, if you have aninvestigation where the lead was based on
some kind of machine learning algorithm,you're gonna have to be able to explain

(31:33):
that in court during discovery you have tohave the code, you have to have you have
to be able to replicate every single step.
How you got from A to B. And youhave to have a human in the loop,
obviously multiple humans in the loop.
And that's what was true back thenand that's just as true today.
, Besides the internet, what were some,other things that you worked on while

(31:53):
you were in this futures working group?
Very good question.
Things like neighborhood driven policing.
One of our biggest peeves wasthat law enforcement in general
is a hierarchical organization,very similar to the military.
Mm-hmm.
And it has a top down approach.
It is probably one ofthe most inefficient.
Types of organizations there is.

(32:14):
So we were looking at thingslike, well, how does nature react?
How does nature work?
And it turns out that in natureeverything is net centric.
You've got people or animalsor bees, and they will tend to
concentrate around a, a problem.
A problem comes up and peoplewith the skill and the knowledge
to solve that problem tend toconcentrate on it and solve it.

(32:37):
And there's communicationacross the entire network.
In a hierarchy, information getsbasically filtered bottom up.
Mm-hmm.
Right?
And by the time it reaches the top,it's probably only 10% accurate
anymore compared to the original.
Whereas in a flattened organization,a network centric organization,
everything is fully connected.

(32:57):
Everybody has the same informationfrom the same source as anyone else.
And one of the comparisons we madewas with, international terrorist
organizations, they have a network centrichierarchy, you know form of organization.
They would all communicate, theyall knew where everyone was.
Anybody could talk to anybody.
You didn't have to go through threedifferent people to get to the top.

(33:20):
And that's what made them so efficient.
And so one of the things we saidat the time was that you can't
really fight an, a highly efficientorganizational structure with a very
inefficient organizational structure.
And that somehow gotlittle bit of traction.
I mean you started to see somedecentralization and some police

(33:44):
departments to some extent.
But law enforcement for the most partis still pretty much hierarchical today.
Mm-hmm.
That's interesting andI've heard that about.
Organized crime.
Mm-hmm.
Being, flattened.
It also helps them with adefense in, being dismantled.
Right.
It's, you don't have this we're gonna justcut the, cut a, the head off the dragon.

(34:04):
Exactly.
, You arrest like a half dozen people in a network of a hundred individuals.
Think about the cartels.
It's, it's the cost ofdoing business for them.
Yeah.
You've got a cartel leader and maybea couple of lieutenants or regional
leaders, but there's still a networkcentric organization and the loss
of a few people is acceptable tothem, and it does not cripple them.

(34:27):
Hmm.
, As you said, that didn't govery far in law enforcement.
That's a difficult one.
It is.
It is probably definitely difficult inthe United States given that there's
17,000 different police departments.
That is another yes.
Another thing that we talked about isif you look at the the difference in
training between Europe and the UnitedStates, for example, in Germany, it takes

(34:48):
four years to become a police officer.
Mm-hmm.
Here, depending on the state,anywhere between two and six months.
Wow.
And half, half of the, the time inwhen you're in school, say in Germany
I had several friends that became copsin Germany you're, you're basically
half a lawyer by the time you graduate.
And it's, it's a much morecomprehensive training.

(35:10):
And there's national standards.
There are no national standardshere in the United States.
At least.
At least no significant ones.
Every state has their ownstandard and sometimes counties
have their own standard.
Mm-hmm.
Right.
And so the quality of of police trainingvaries greatly in the United States.
And that's not to put anybody workingin law enforcement down at all.

(35:32):
It's just is It is what it is.
Right.
Yeah.
You gotta make do with what you have.
Yeah.
I think it's the, it'sthe nature of the beast.
And I, I think it is exactly.
And it's, and also you talked about thehierarchy and the delay in, communication.
Yes.
There's been plenty of, plenty ofexamples where criminals took advantage
of the fact that you, you can easilyhop jurisdictions and knowing that the

(35:55):
data's not being shared, so mm-hmm.
They can they can have this huntingarea that goes around multiple
jurisdictions and, and no one'sseeing the complete picture.
Exactly.
On the law, law enforcement side.
Exactly.
And, and combine that with the factthat a lot of agencies, even neighboring
agencies can't even communicate becausethey're different radio frequencies.
Mm-hmm.

(36:15):
Just simple things like that.
And, and that still it's gottenbetter, but it's still to a
large extent, persists today.
Yeah.
Hmm.
So what was the I guess you, you were,how long were you with this working group?
I'm sorry, I should have it, it'sright here in front of me about,
yeah.
Futures working group.
About, I wanna say 12years, 12, 13 years maybe.
And it was, it wasvolunteer work, you know.

(36:37):
Okay.
It was all volunteer work and I wasdoing that on the side top of the
other projects I was working on.
And the behavioral science unitgot disbanded, and I think they
got rolled up into the criticalincident response group in,
in dc.
. Do you have like your big accomplishmentwith this group or one that you're maybe

(36:58):
the most proud of during your time here?
Not really.
It was just a lot of fun.
Mm-hmm.
There were some absolutelybrilliant people.
In, in the futures working group, wehad there's a core group of maybe 12,
13 people consisting of police chiefstwo or three police chiefs, researchers,
practitioners also contractors, vendorvendors, occasionally and two or

(37:23):
three times a year, we used to meet.
Have brainstorming sessions.
We had quite a few publications wepublished in the FBI National Academy
Magazine about some of our work.
And so the work did get outand we did get some traction.
We sparked a lot of ideas and wereinvited to quite a number of law
enforcement oriented workshops workinggroups, conferences, things like that.

(37:48):
So yeah, it was very little costfor the FBI, 'cause most of us just
often paid our own way to travel.
Sometimes we got paid for thehotel room, sometimes airfare,
depending on what it was.
But it was, for the most part, it wasvery little cost to, the taxpayer.
But I think quite a bit cameout of it in terms of material.

(38:12):
Hi, this is Dr. Carlena Oroscofrom the Tempe Police Department,
Arizona State University, and mypublic service announcement is that
correlation does not equal causation.
If you find that certain things areoccurring that may be contributing
to a decrease or an increase.
In crime, for example, that givesan opportunity to investigate it a

(38:36):
little bit further to see if possiblythere are things contributing.
But it does not mean that one thingcaused the decline or the increase, it
just means that there's an opportunityto explore it a little bit further.
Hi, this is Eric Piza.
John Jay College.
It's been the last rough couple years, solet's just all be, be kind to each other.

(38:56):
I've been on social media a lot lately.
You're not being kind, so let'sall just be kind to each other.
So this is Sam and I wanna let you knowthat it's okay to talk to strangers,
obviously not if you are four or ifyou're walking alone at night or in
the woods, but in general, if you'rejust out in your day-to-day life or

(39:18):
you're traveling or whatever, talkto somebody, talk to strangers.
It makes you a more interestingperson because it gives you
more perspective on life.
Everyone is walking aroundwith an interesting story.
So many people willdefy your expectations.
When you you see someone and youmake certain assumptions about
them, whether they're consciousor unconscious, I love the moment

(39:41):
when you realize you were wrong.
It's a great feeling and I think itmakes your life richer in general.
You know, if you're too shy, thenmaybe just read Humans of New York.
That might help you to, to understandother people's experiences.
But I'm just here to saydon't not talk to strangers.

(40:04):
So then you make your way to TrueNorth data systems and mm-hmm.
More in a consultant role at this point.
And then, so what did you getinto while working, with them?
Well, we started a TF right,as I spoke about earlier.
And then the next project was,with an inmate phone company
and they kind of came to me.

(40:26):
I was doing some voluntary workfor the Pennsylvania Commission
on Crime and Delinquency.
Mm-hmm.
And it was a Pennsylvania company thatwas providing inmate phone services
and they were looking for somebody whoknew about speech recognition speech
translation and basically AI and machinelearning that could help 'em automate

(40:50):
the speech transcription of inmate phonecalls and make them searchable and.
Intelligently mind those calls forcertain topics, things like that.
The problem was that in your averagefacility, say with 5,000 inmates,
you generate about between two and3000 hours of phone calls a day.

(41:10):
It's humanly impossible tolisten to every phone call.
There's just not, not enough people.
Mm-hmm.
And so if, if there's a way that youcan transcribe the audio to text, play
it back with like a bouncing ball,highlight things that would be of
interest to investigators and allowthem to search basically those two,
three hours, two 3000 hours of phonecalls a day in a, in a large facility

(41:36):
as well as even smaller facilities.
Mm-hmm.
It would greatly helpcorrectional intelligence
analysts and, and crime analysts.
Oh yeah.
And then that's what I did.
So I designed a system that, and Ideveloped it from scratch using a
combination of C and Java using a speechtranscription software called BB n

(41:56):
Evoke, which originally came out of MIT.
And I made every one ofthose phone call searchable.
I used natural language processing todevelop topic detection techniques.
So for example, you could search forall phone calls that mentioned drugs
as well as weapons and where some kindof a threat was issued, and you would

(42:20):
have results in less than five seconds.
Wow.
And it would list all, here'sall the latest phone calls
that have hits on those.
You would click on that phonecall and it would take you right
to the point in a conversationwhere these things were mentioned.
You didn't have to evenlisten to the whole call.
And that was highly successful.
We our first.
Implementation was inMontgomery County Pennsylvania.

(42:43):
And from there it just mushroom.
By the time I left, which was around2011, I did it for six years, seven years.
No, I,
I was gonna ask you when this all started.
So this is like Right, 2005
ish?
2005 ish.
Yeah.
And I left in 2011, so aboutsix, six and a half years.
And by the time I left, itwas in about two dozen prisons

(43:05):
and, and local jails as well.
Highly successful.
Wow.
Then towards the end, I added the,I added the mapping components.
So when inmates were receiving phonecalls, we did a reverse lookup on the
phone numbers that were dialing in mm-hmm.
Where they, they were calling andgot the address, geocoded them,
and then combined that with a map.
So now you could zoom in on, say, aknown drug area in a, in a, in a city,

(43:31):
and query all phone calls that wereplaced to that area where the topic was.
Drug dealing.
Hmm.
And again, this, it's all about puttingdata together, using different techniques.
I call it multimodal interfacesall the way from playback to the
topic detection, to the mapping.
Just integrating everything.
And that gives you design, thinkof it like a very early, early

(43:55):
kind of Palantir kind of system.
Mm-hmm.
That's basically what it was.
Hmm.
And there's so much, I, I have lotsof guests that come on here and
tell me stories about the, the greatintel they get from Jill calls.
Oh, yeah.
And even though the folks are told that,yeah, this will be recorded, they still

(44:15):
seem to discuss, openly what's going on.
Well, and that brings us, brings meback to the point we were discussing
earlier about hierarchical networks.
The, the organized crime knows ittakes police a while to respond.
The inmates know full well.
That the odds of anybody listeningto that phone call, especially in a
large facility, are close to zero.

(44:36):
, Unless a, a correctional officer had somekind of a tip saying, Hey, this, this
inmate is threatening somebody, and thenthey go retrieve the phone calls for that
inmate, they wouldn't even know about it.
Yeah.
Yeah.
So, I mean, that's, I guessthat's a, you didn't say that.
I mean, I maybe you said that explicitly.
So this is obviously more than just maybestuff that's going outside the jails.

(44:57):
This is obviously detectingpotential violence inside the jail.
Oh yeah.
We, one, one of the earliest cases we hadthat we found using that system, it was
called Call iq, and the mapping componentwas called Map iq was an instance where a,
an inmate was working on the assassinationof a federal agent that arrested them.

(45:19):
Oh, wow.
Hmm.
That's,
Good.
And you, you said youbuilt that from scratch?
From scratch, yeah.
That is awesome.
Yeah, I was with basically, again,back then that around that time open
source was just starting to come up.
Right.
You had like open source IDslike net beans and eclipse.

(45:40):
I used Java for the most part some see,but mostly Java to design the system.
It was running on an Oracle databasebackend, and yeah, and it also had to
feed into the actual phone system itself.
Well, I, I could stay on that for alittle bit longer, but I'm gonna move on.
Sure.
So this and as we just, we talkedabout a little bit offline.

(46:02):
You're, got a lot going on.
You're volunteering, you're working,you, you got your hands on a bunch
of different cookie jars here.
Mm-hmm.
But you go on then to Marius Analytics.
Mm-hmm.
This is getting into 2016 nowas a senior research scientist.
So I guess just curious, why did youtransition from true North to, Marius?

(46:24):
Well, at the time federalfunding was, was running out.
Mm-hmm.
You know every time you hada change in administration
they had different priorities.
And around 2005 to 2000 and.
10 Really?
There was, there was very little fefederal funding for, for major research
projects or anything like that.
So that's why I went withthe inmate phone system.

(46:45):
Hmm.
And then, yeah, in 2016 I was justlooking see what jobs are out there.
So I was kind of giving upon the contracting mm-hmm.
And the things, and so first time inmy life I ever had to go job hunting
because every, every job I'd hadprior to that was word of mouth.
Oh man.
And yeah, every conconsulting gig or, you know.

(47:07):
Research gig was all word of mouth.
So I was contacted I put my resume outthere and I got a call from a company
called Marius Analytics, and it was astartup at Carnegie Mellon University.
They had there was a undergraduate studentand her undergraduate thesis was crawling

(47:28):
the web and analyzing the data from adultescort sites to detect human trafficking
specifically child sex trafficking,but also human trafficking in general.
And I was their very first employee.
They had a, a grant from the NationalScience Foundation it's called the Small
Business Innovation Research Grant, whichis basically the NSF has those grants.

(47:49):
DOD has those types of grants as well.
They're basically highrisk, high reward grants.
They're trying to findthe next Google the next.
Greatest thing since sliced bread.
So, and they're very smallgrants to begin with.
Like you're talking 150 somethousand dollars if that.
So I was their first and onlyemployee for about six months.

(48:10):
And so my job was to take what bunchof student developers had developed and
try to productize it right to the pointwhere we could actually sell the product.
They had a few customersalready nothing major.
And so I worked on that.
Then eventually we hired one moreperson after two years, I think we
had three or four full-time employeesand there were more people involved

(48:34):
with the company, but people thatwere focused on the technology itself.
There was about three orfour of us after two years.
And so what we did was, I. Wedeveloped these web scrapers and we
started out with Backpage with whichback then was the by far the largest
adult escort site in the world.
It was represented in almostevery country in the world.

(48:55):
So that was our main sourceof online escort ads.
We then started developing crawlers forother websites and learned a lot about
undercover crawling surreptitiously.
So the web websites couldn't detect,you don't want to go into the details
of that, but there, there are ways ofdoing that and getting around things
like Google capture that kind of stuff.

(49:17):
So Yeah.
Staying under the radar.
Yeah, exactly.
Staying under the radar.
So let's just say I could make it looklike I was surfing an adult escort
site their New York City portion from.
At T Mobile IP address in Queens, andso I make it look like all natural.

(49:38):
So yeah, we did that.
We added mapping again to it.
And again, this is not all my work.
I was working with a wholebunch of people, you know?
Mm-hmm.
Who were absolutely brilliant andwe basically took it from a startup
coming out of a university projectto within five years being a world
leader in the, in that space.

(50:00):
And we started working withpolice around the world in the
uk, in Canada, India Europe.
Then towards the end of mytime there, I was training law
enforcement around the world.
Giving talks at conferences, Austria,Germany was spoke at a conference in
in Berlin at the PKA and which was nice'cause also intimidating 'cause it was

(50:24):
the first time I'd ever have to give a.
Presentation in Germanat a conference and,
and you were speak speaking German, right?
Yeah, yeah.
I was speaking German.
Yeah.
Same thing in Austria.
And well, that's, that's through that is,well I met Sam Steiner before then, but
Sam helped me get into the Austria thing.

(50:44):
And so it was fun to see himin his natural habitat and
yeah.
But that was, but that was when Samgave you a hard time though, right?
Oh,
it was, yes.
At the airport.
Yeah.
He had a quote welcoming committeefor me, and everybody thought I was
gonna be arrested my fellow passengers
good old until I gave him a big old hug.
Yeah, that's, that's Sam.

(51:05):
Yeah, yeah, yeah,
yeah, yeah.
You gotta love him.
And so yeah, that, and thenone of the things we did
there, we were very successful.
We, I mean, this is just atotal estimate, but 'cause we
surveyed our, our, our customers.
All the time, or annually.
And one of the questions we ask is,how many people do they estimate

(51:27):
that our work our software helpedrescue outta sexual slavery.
And these were all estimates you'regonna take this with a truckload assault.
But I would say over the course ofthe six and a half years or so, I was
there, my colleagues and I helped rescueat least, I wanna say 5,000 people.
Oh wow.
Outta sexual slavery.

(51:48):
And that all made it worthwhile, right?
Yeah.
Then about.
Three years, two years before Ileft Marus and decided to move on,
we entered the IBM Watson GlobalAI for Good XPRIZE competition.
For those not familiar with the XPRIZEcompetition, the very first one was

(52:10):
it was founded by Peter Diamantes,who's a billionaire investor.
It was a prize that wasawarded to the first.
Private company who was able toput a human being in outer space.
And that award went to BlueOrigin Jeff Bezos outfit.
And then after that there were severalother X prizes and we started out with
about four or 5,000 companies worldwide.

(52:33):
All shapes and sizes, all differenttopics, the things that they did.
And they had severalrounds of elimination.
And it was, it was a lot of work.
It, it was every six months youhad to do a progress report.
And then after we made it through thelast round of, of 25 people which was
actually interesting in that in thelast round of 25, 8 of the companies

(52:55):
were from the United States, and sevenof those I believe were universities,
US universities that says a lot.
And three of the companieswere from Pittsburgh.
And that says a lot about Pittsburghand Yeah, it and Pittsburgh.
Nice.
And advanced research.
Right.
And then, so we actuallymade it down to the.
Final three.
And then COVID hit the, it wassupposed to have been awarded in 2020.

(53:17):
There was supposed to have been aaward ceremony somewhere that all
went out the window with COVID.
And then finally in 21 they announcedthe winner and we won third prize.
And the first prize went to an Israelicompany that did fantastic work.
They did aerial reconnaissance, remotesensing to predict where the next

(53:38):
outbreak of malaria was gonna be.
Oh, okay.
And the second company was the secondprize winner was from Montreal.
It was a company that used AIto look at brain scans to take
the first signs of dementia.
Oh, okay.
Hey, well wait.
And mental health issues.
It sounds like you lost to somelegitimate contenders right there.

(54:00):
Oh,
absolutely.
Absolutely.
Yeah, I mean, that was, really cool.
And
when and when you're in this finalround, you're the only US company left,
right?
We were the only US company left, yes.
In fact, after the, was it eightor 10 after the, the first round
of the, the round of 25, therewere eight US companies left.
Mm-hmm.
I think, and the round afterthat, there was like 10 companies.

(54:22):
Mm-hmm.
There were no US companies left, andthen it came down to the final three.
Wow.
Wow.
Now, so when you're.
Working with this program, it's AI based.
It's constantly learning.
And so it's, it, it just, it's kindof different from this idea of like,
if you think about coding in thebeginning when you're coding and

(54:43):
you're building a program and yougot this constant maintenance of it,
or you have to make upgrades to it.
But in the a in the AI world where youhave this AI and it's learning on its own,
it, it's, it's a little bit different,
not quite true.
Okay.
So we didn't actuallyuse a whole bunch of ai.
Mm-hmm.
With, with the human traffickingweb scraping is minimally AI based.

(55:07):
Mm-hmm.
Depending on what te technique you'reusing, the biggest thing we use for ai.
Were two things.
One, and this was developed at,at CMU prior to when Meredith
was formed, was image similarity.
So it used ai basically it usedconvolution, neural networks to

(55:30):
detect images, find images thatare similar based on either the
background or something in the image.
Mm-hmm.
Give an example.
We had one person that was beingtrafficked and at that time we had
hundreds of thousands, millions of images,tens of millions of different images
stored, and this was all in the cloud.
And so a an agent from a federalagency sent me an image saying that

(55:55):
we, we identified a trafficker.
This person was traffickingEuropean teenagers, and here's,
here's an image of this person.
That was being advertised.
And that image came from an adultescort site and we have like two
victims and we have the trafficker.
I then used that image similarity search.

(56:15):
I was able to find three more victims.
And the reason we were able to findit was because the trafficker had
staged photographs of the victimson his personal bed with the same
bed sheet, which was basically awhite background with hearts on it.
Mm-hmm.
And the algorithm was able to findall images with a person on a bed with
that, specific color bed sheet on it.

(56:37):
Mm. And so it was ableto find more victims.
And then another example that we used as,as we used a lot of facial recognition.
So we took all the faces of the peoplethat were being advertised and we
had huge facial recognition database.
And literally within seconds wecould submit an image, search our

(57:00):
database for any image similarities.
Right.
And yeah, I understand.
Or faces that match that person.
And if there's one thing I learned inthis whole process is that people have no
idea how many doppelgangers are out there.
I mean, so many times, especially ifyou have millions and millions, tens of
millions of images of different people,and you could swear up and down, this is

(57:22):
that person turns out often it wasn't.
Wow.
It was you have to literally have arelative or a friend or somebody who knows
that individual, confirm it sometimesit's obvious 'cause there's a tattoo.
Tattoo or something like that, right?
Mm-hmm.
Or a scar or something like that..But oftentimes it was just impossible.
And, and there's been alot of controversy, right?

(57:43):
About facial recognition.
Yes.
It's sometimes it'sbiased towards minorities.
That is the case.
But you have to look atwhat the goal is, right?
If the goal is to find people thatare being trafficked based on an
image, you just want accuracy.
You just wanna be able tofind the person, right?
Mm-hmm.
You're not accusing a person of anything.
You wanna find a person you wanna, sofor that reason, the national Center

(58:08):
for Missing Exploited Children usedour software to, every time they had a
report of a teenage runaway, they wouldrun their image through our system.
And it's my understanding, I don'tknow exact figures, that they were able
to actually find and recover a numberof runaways based on that system.
. That's great.
Yeah.
Alright, well then, let'smove on to where you are now.
As I mentioned in your intro you arewith I three as the team lead . So

(58:33):
what are you getting into now?
So now we are, I'm working for afederal agency and working still machine
learning, integrating data from avariety of different sources and try
to do things that has always been a petpeeve of mine or one pet peeve, right?
Mm-hmm.
And that is that analystsspend way too much time.

(58:55):
And, and the same thinggoes for investigators.
They spend way too muchtime sifting through data.
And way too little time doing actualinvestigations or actual crime analysis.
Mm-hmm.
So the, the kind of things we focuson are trying to find organized crime
groups and to try to find leads using avariety of machine learning techniques

(59:16):
and to give investigators leads thatare accurate that are derived from
billions of data points and that don'tsend people on a wild goose chase.
Hmm.
And we've been quite successful at that.
I would say we've.
Had developed systems thatare at least 90, 95% accurate.
, I'm fortunate that I'm working with someabsolutely brilliant freaking people.

(59:38):
Really, really good people,both on the government side as
well as all of the contractors.
It's been a wonderful experience.
And at my age, I keep learning newstuff every day, new techniques, I
get to develop new techniques still.
They say you can't teach an old dognew tricks, but you definitely can.
In my experience, I learnedsomething new every day.
Yeah.
So I'm curious if you think back, we'vegone through your journey now mm-hmm.

(01:00:02):
From the very beginning till now.
Is it easier or harder now?
Some things are easier.
Definitely for sure.
Because you, you build up the expertiseyou kind of go from being hired to be
a developer or something like that, ora trainer to, as you get older and as

(01:00:22):
you mature, you get hired more to V sme.
Mm-hmm.
And so that gets easier.
What has not gotten easier isthat technology is changing
at an accelerating ch rate.
So people that think, like in theold days, 30 years ago when I went
to college back then you got a degreein something and that degree would

(01:00:44):
last you for your entire career.
That is no longer the case.
It just ain't gonna happen anymore.
You are constantly gonnahave to be relearning.
AI is gonna have a profoundimpact on things in the future,
on, on some people's jobs.
Although I still think as of right now.
It's completely overblown.
Oh, okay.
Today what when people talk about AIin the media or anything, they, they're

(01:01:06):
really talking about large languagemodels, which are when you really
come to think about their giant hiddenmarkoff models, some, some of 'em are,
and it's really not that sophisticated.
'cause all they're doing is, isthrowing computing power at it.
And they're, the graphicsof are, are somewhat good.
Yeah.
You can create deep fakesand things like that.

(01:01:27):
Sure.
But that's not reallyproductive for anything.
Right?
Mm-hmm.
LLMs, they let people basicallywrite a dissertation form.
Is that a good thing?
You can say yes or no.
I mean, yeah.
It, it reduces the amount of time thatyou need to do research, but that research
doesn't become very rigorous becauseyou're not learning the actual techniques.

(01:01:48):
You're just regurgitatingwhat a model tells you, right?
Mm-hmm.
You're not actuallydoing the work yourself.
So it's definitely a drawback to that.
But that's neither here nor there.
But the thing is, in today's world,things are changing so fast that , you
are constantly relearning your job.
You're constantlyincorporating new techniques.

(01:02:09):
It used to be in the nineties, every five,six years, a new technique came out, or
new type of software came out and it wasthe greatest thing since sliced bread.
And now it's happeningevery three to six months.
Yeah.
Somebody somewhere developssomething, throws it on GitHub,
everybody goes nuts over it.
And then the key thing is alsois recognizing which new things

(01:02:32):
that are coming out are actuallyapplicable to law enforcement.
And the crime analysis field, or practicallaw enforcement field, because the
last thing you wanna do is implementtechnology for the sake of technology.
Right.
You only wanna do that if it makes sense.
So the challenge is, because there'sso much new technology, is sifting
through all of it, trying it out,testing it, seeing if it makes

(01:02:53):
sense in your area of application.
Yeah.
I I, and that's, that's a goodpoint that you make because I
feel for these police chiefs Yeah.
They're on limited budgets, they mighthave enough money, it seems like the,
each police department may have enoughmoney to spend time, energy, and money in
like one aspect . They got vendor aftervendor, after vendor coming at them.

(01:03:17):
Yes, yes.
And they're in this position notonly with all the other things
a police chief has to do mm-hmm.
Decide on what technology to invest in.
Absolutely.
And there in is lies a huge problem.
And that is that most police departments,less so with the federal agency, but

(01:03:40):
to some degree as well, they don't havethe in-house expertise to even begin
to analyze what's out there to takewhat a vendor says in a sales spiel.
Mm-hmm.
And to legitimately be able todecide, okay, what's the best.
Solution for us, right?
What's a bunch of smoke and mirrorsversus what is truly going to benefit

(01:04:03):
us and, and gonna be cost effective?
A lot of vendors know that, right?
And, and there are, not to name anyspecific vendors, but there are one or
two very famous, very often mentioned inthe media vendors that everybody said with
the greatest thing since sliced bread.
Oh my God.
Everybody has to get them,everybody scrambles to get them.
They're expensive as hell, but don'treally do anything that a sophisticated

(01:04:29):
in-house research staff or, orstaff of data scientists along with
crime analysts, software developers,engineers, and so on and so forth.
Cloud specialists couldn't have developedthemselves for a fraction of their money
using open source stuff, open sourcesoftware, a lot of these companies
historically is even today still.

(01:04:50):
They actually take opensource software, put their own
wrapper around it and sell it,
and
just little tweaks.
Yeah.
Essentially you're, you're they'lltake open source software and put their
own interface to it and sell it astheir own and make a bunch of money.
Then of course a big thing back in theday, way back in the way day was a lot of

(01:05:14):
the RMS and CAD vendors, they had theirown proprietary databases, so mm-hmm.
If you were using their systemnow, all your data was stuck in it.
So if, if you wanted to move to anew vendor, a new RMS or a new CAD
system, good luck getting your olddata out of it, your historical data.
They would charge you millions justto write a script to export data

(01:05:35):
into A CSV or what have you so youcould import it into, say, Oracle
millions of records.
Yes.
Yeah.
And, and that was yeah, that wasdefinitely one of the major pain
points in the, in the ninetiesand in the, in the two thousands.
It's, to some degree not the sameanymore, but there is still a lot

(01:05:56):
of beltway bandits that use thosekind of techniques whether it's
in law enforcement or elsewhere.
They'll have their own proprietarysystems, database systems that actually
a lot of 'em run on commercial.
Off the shelf databases, but theyput a wrapper on them that makes it
impossible to get your old data out.
Hmm.
All right, let's move onto the advice section.

(01:06:17):
I'm really excited about this sectionbecause one of the questions I
like to ask my guest is what I callthe return on investment question.
Right?
Meaning, what can analysts studynow because five or so years from
now, it's going to be important andI can't think of anybody else that
would be perfect to answer thisquestion other than you, Ollie.

(01:06:39):
Oh, okay.
So bunch of things at it.
Like, would, like you talked aboutwhat's the trying to , apply tomorrow's
technology just for the analysts.
What, what do they need to be ready for?
Well, the hottest thing right nowin law enforcement analytics is.
Are graph databases.
So graph databases are incredibleat connecting the dots.

(01:07:02):
If you want to discover organized crimegroups, people that are connected,
incidents that are connected,throw it into a graph database.
You would be amazed atwhat you can discover.
And here's the good part.
There are many really good high qualityopen source graph databases out there.
There's Neo four J, there's puppygraph just to name two of them,

(01:07:24):
and a lot of commercial ones too.
There's also commercial versionsof Neo four J, but the things.
We have been able to do over theyears, both at, you know when, when
I was working at Marius, we used thegraph database to identify probably
one of the largest national massageparlor organizations, Asian Massage

(01:07:47):
Parlor organizations ever investigated.
And now today I'm doing the same thing.
Now you're not just with thegraph database, you're not just
putting data into the database.
You can combine that with.
All kinds of things.
You can use LLMs on graph databases,you can use machine learning on
graph databases, community detection,class of build, classifiers.

(01:08:08):
It, it's endless, the possibilities.
And you can do that in what used totake a researcher even 10 years ago,
months to do in hours because theopen source software is just there.
Somebody else did it.
The biggest challenge is gonna be movingforward for analysts again, because
there is so much new stuff out there.

(01:08:29):
It's not trying to findwhat, what's out there.
It's trying to find the right thing that'sout there that is suitable for what it
is, what it is that you're trying to do.
And that is gonna require constantretraining, constant searching for,
for new stuff . Constantly educatingyourself, reeducating yourself,
and it's gonna be never ending.

(01:08:49):
It's not gonna be like you geta certification in something.
That certification is going tobe good for five to 10 years.
It might be for some things, but mm-hmm.
When it comes to being on the cutting edgeof law enforcement technology, things are
not by any stretch what they used to be.
And it's only gonna get more chaoticbecause technology is being developed

(01:09:11):
at an accelerating rate, and, andthat's the most important thing.
Or piece of advice I can give people.
If someone's listening to this mm-hmm.
And like, okay, I, Iwant to take your advice.
I want to try to learn as much as I can,keep on learning, be a career learner.
Mm-hmm.
This thing just Right., How would yousuggest that they maybe get started

(01:09:32):
and then obviously work their way upinto getting into more advanced stuff?
Okay.
There's all kinds of opensource places you can go, right?
Start with, let's just takeNeo four J for example, right?
And again, there's many others.
I'm just using it as an example.
There are all kinds of communities withNeo four J that have certain techniques.
I'm not aware of any law enforcementrelated communities for Neo four J yet.

(01:09:57):
That might actually be something ifyou can get a group of people together,
some to consider starting and, and.
Trying to help out other crime analystsor law enforcement analysts in general.
And there is just a plethoraof literature out there on
techniques and things like that.
One of my former coworkers TimRidge, actually wrote a book on it.
And introduction to Needle fourJ and machine learning and things

(01:10:21):
like that, that you can do with it.
It's like 20 bucks or something, $29.
And I'm not trying to promote it, butit's, it's really inexpensive and mm-hmm.
It's very well written, self-publishedtoo so but there's all kinds of GitHub
repos things like that that, that youcan go to and, and download sample data
sets that you can play around with.

(01:10:43):
And Neo four J will run on any laptop.
You don't need specialized hardwareor anything like that, but don't
look at just graph databases.
Look at different typesof machine learning.
There's the usual suspects like psychkit learn, and TensorFlow if you want
to look at computer vision for example,for identifying things in video or

(01:11:06):
doing things like facial recognition,there's also free free wear out there
and open source software out there.
It's everywhere.
I mean, all you gotta do is look reallythere isn't any one particular place
where everything is concentrated, whichagain, it would be nice if there was
one for law enforcement in particular.
If there was a website somewhere ora group or something that looked at

(01:11:28):
the latest technology coming out thatwould point people who are starting
out at looking at those kinds of thingsfor law enforcement, have a place to
start and know where to start and haveresources already collated for them.
Hmm.
All right.
Well, very good.
Let's finish up withpersonal interest then.
Yeah, so you mentioned sailing beforeand you're actually a gamer too, so yeah,

(01:11:52):
let's let's get into that a little bit.
'cause I mean, even as a sailorthere, you, lived in the boat
there for a while, right?
Yeah, yeah.
We, we have a live aboard yeah,two bedroom, two bathroom sailboat
and we sell to Great Lakes andbeen doing it for about 30 years.
Yeah.
And wasn't my idea of something Iever wanted to do, but when I met

(01:12:14):
my wife, the wife came with a boat.
So that's
funny.
'cause
I think it's something a lot ofcouples, it was the other way around.
Right,
exactly right.
So she taught me everything I knowabout sailing and since I do all the
maintenance on the boat, the engine andstuff like that, we do it all ourselves.
I've been wrenching oncars and bikes since I was.
Fall 13 years old, so I've knowmy way around an engine room.

(01:12:38):
Not a problem.
Yeah.
Or plumbing or wiring,electronics that kind of stuff.
And so yeah, we, we do that.
And I also have an RVfifth wheel that I pull.
We've only had that a couple of years,but again, it involves outdoors and
doing things going places and so, yeah.
And gaming, of course.
I'm
So what, you got a game you're playingnow or do you have a favorite game?

(01:13:01):
Yes.
Oh, a couple of favorite games.
I'm playing again, fallout four.
Okay.
One of my favorite games.
But there's a whole bunch out there.
There's the new version of.
Outer wills just came out.
I'm looking forward to playing thatover the winter, but it's just something
to do it's like keeps my mind going.
And you know that that 20 hours

(01:13:22):
a a day that you're
up?
Yes.
Yeah.
Well, not anymore.
Not, not that I'm older.
No.
I get my eight hours sleep everynight, so I'm not saying I'm not
nearly as ambitious as I was.
As you get older you justsay, screw it, you know?
That's for sure.
All right, ollie, this hasbeen great so much here.

(01:13:44):
I, I I could have spent so much moretime on each one of these topics.
Congratulations on it all.
Just your journey.
Thank you.
And everything.
Again, I know it's not all you, you talkedabout working on different teams and
doing things, but certainly you were apiece of a pie and it's so, so great to

(01:14:04):
hear your story and your contribution.
So I just wanna thank you not only forall this time today, but everything
that you've done for law enforcement.
Thank you.
Appreciate that.
Alright, our last segment tothe show is Words to the World.
This is where you can promoteany idea that you wish.
Ollie, what are your words to the world?
Something my dad taught me.
So I was a foreign service brat, soI've, I've lived in six countries along

(01:14:27):
the way, learned a couple of languages,and I always saw my dad, no, no matter
where we lived and my, my mom too, likehelping those that were less fortunate.
And I remember when the first timewhen we were living in England
my parents took in a refugee forabout three or four weeks and took
to find a permanent place to stay.

(01:14:47):
And then they did that also whenthey were living in Germany.
And so one of the best pieces ofadvice my dad ever gave me was
you don't have to be rich, youdon't have to have a lot of means.
Everybody can make sure that their littlecorner of the world is a little bit
better off than it was when you got here.
Mm-hmm.
And so whether that's volunteeringhelping the homeless, volunteering

(01:15:12):
in your local humane society.
My wife and I volunteered in soup kitchensand things like that, and part of it
also was it took a huge pay cut to, towork on this human trafficking project.
Just give back giveback to your community.
And especially for me as an immigrant Iwouldn't be who I am if it wasn't for the

(01:15:34):
opportunities I got in the United States.
I'm just giving back as, as a thanksand, and everybody should do it.
The world would be a lot better placeif more people were volunteered.
Awesome.
Use your, use your crime analysisskills to help humane society.
Build a database right of, of,of customers or donors or things
like that, or your local church.

(01:15:55):
Doesn't matter where everybodyhas skills that they can use
to contribute to society.
Awesome.
Very good.
I leave every guest with, you'vegiven me just enough to talk bad
about you later, but I do appreciateyou being on the show, Ollie.
Thank you so much.
And you be safe.
Thanks for having me.
Take care.
Thank you for making it tothe end of another episode of
Analyst Talk with Jason Elder.

(01:16:15):
You can show your support by sharingthis and other episodes found
on our website@www.podcasts.com.
If you have a topic you would likeus to cover or have a suggestion for
our next guest, please send us anemail at Elliot podcasts@gmail.com.
Till next time, analysts, keep talking.

All Episodes

Analyst Talk - Dr. Andreas Olligschlaeger - The Futurist Analyst

Episode Transcript

Popular Podcasts

Stuff You Should Know

Dateline NBC

The Burden

.css-15opob5{left:0;position:absolute;top:0.8rem;} All Episodes

.css-14f5ked{margin:0;word-break:break-word;display:-webkit-box;-webkit-box-orient:vertical;box-orient:vertical;-webkit-line-clamp:2;overflow:hidden;}Analyst Talk - Dr. Andreas Olligschlaeger - The Futurist Analyst