AI Explainer & 'Shark Week'

Ryan Gorman hosts an iHeartRadio nationwide special featuring Dr. Bart Kosko, best-selling author, world-renowned scientist, and expert on AI. Dr. Kosko discusses his 1993 international best-seller Fuzzy Thinking: How AI Thinks in Shades of Gray and breaks down some key questions on current issues involving the rapid expansion of AI. Dr. Greg Skomal, Shark Week Expert & Author of CHASING SHADOWS: My Life Tracking the Great White Shark, also joins the show. Dr. Skomal discusses his life working with sharks, correcting public perceptions about the species, and continuing global conservation efforts.

Episode Transcript

(00:00):
Welcome to iHeartRadio Communities, a publicaffairs special focusing on the biggest issues impacting
you this week. Here's Ryan Gorman. Thanks so much for joining us here
on iHeartRadio Communities. I'm Ryan Gorman, and we have a couple of important
conversations lined up for you. Injust a bit, we're going to talk
to shark expert Greg Skulmulus. Thispast week has been Shark Week on Discovery

(00:23):
Channel. He's author of Chasing Shadows, My Life Tracking the Great White Shark,
and he is a featured expert onso many Shark Week specials. Will
talk about his life observing sharks,also the conservation of sharks, and a
whole lot more. So stick aroundfor that. I think you'll enjoy that
conversation. Right now, to getthings started, I'm joined by doctor Barksco,

(00:44):
a best selling author, world renownedscientist and expert on artificial intelligence.
Is nineteen ninety three international bestseller FuzzyThinking How AI Thinks in Shades of Gray
is being rereleased in a digital version, and he's with us to help break
down some questions on current issues involvingAI. Doctor Costco, thank you so
much for coming on the show.And let's begin with your book. What

(01:07):
exactly is fuzzy thinking and what weresome of the things you were working on
related to AI thirty years ago.Fuzzy thinking is a way of referring to
fuzzy logic or thinking in shades ofgray rather than simple black white reasoning,
like a computer chip works on oroff. It's the way people think.
So, for example, we wouldsay that a pink rose is both read

(01:33):
and not read to some degree.Now that's simple enough, common sense,
but very different in terms of classicalWestern logic, which is either one or
the other, either or this caseit's both. And that kind of thing
softens decision making in the case ofAI applications. There's many applications of fuzzy
logic. In the case of AIapplications, they often have rules, and

(01:56):
the rules like programming an air conditioner, if the error is cool, set
the motor speed to slow. Well, what do you mean by cool air?
Its boundaries are fuzzy, they're great, they're not exact. It depends
exactly what temperature at, whether it'ssixty or eighty percent cool, and so
forth. And you and I meandifferent things by cool. So even words

(02:19):
like that which on the page saythe same thing, or appear to mean
the same thing. You and Imean different things by them. As an
example of an application, I havea Subaru car that uses funsyologic to control
the transmission to shift gears. Somany times per second sensor data comes in
and a bank of these rules makesa decision about whether the shift gears.

(02:40):
The net effect is it does sovery smoothly given changing conditions, and it
doesn't fall asleep and have accidents andthings like that. You probably have a
gadget that has fuzzyologic in it,whether a microwave, oven, a washing
machine, even a camera. Really, thousands of commercial electronic products, many
other systems are told by fuzzy logics. Famously, in Japan, in the

(03:01):
city of Sendai, you may recallthe city hit very hard by the tsunami
in twenty eleven. The city Ascendi, its main subway system, is completely
under control of fuzzy logic, andit's much smoother as a consequence. So
we had all that answer your questionline back in ninety three worked that out.
I had a textbook on it fromprinters Hall called Networks in Fuzzy Systems.

(03:24):
But what we didn't have is acomputing power we have today, and
that's a result of Moore's law,the doubling of computer chip density on off
circuits on a chip every two years, every two years, so in thirty
years we've had sifteen doubling. That'sa staggering exponential increase in computing power in
a fallen price. And you seethat showing up today taking very old algorithmles

(03:47):
with flight modifications in running them onmuch more powerful systems like applications to medical
diagnoses which are very successful with AIand you can you can go back and
validate them, but you also seethem when it comes to applying it to
language. And this is the systemsthat these are systems that are causing the

(04:10):
excitement and the fear now in particularwith the March twenty twenty three released by
open aye of chat GPT four,and that stimulated some competitors that I think
prematurely release other systems, and that'sa whole different game. Those systems,
some of which are fuzzy, butmost of them are using classical neural networks

(04:30):
but on a vast, staggering scaleand applying it to a lot of texts,
literally trillions of strings of texts,everything on Wikipedia, everything on your
social media, and everybody else's booksthat are copyrighted n copyrighted. It's already
led to some class action lawsuits,for example against Facebook in terms of copyrighted

(04:51):
books. But the effect of thatstarts off on completing a sentence for you.
So if I say the sky isand you say blue or the sky
is overcast or whatever, there maybe hundreds of thousands when you search worldwide
databases completions, and if I havethat sentence, I can then rerun this
to complete a complete paragraph and maybea sequence of paragraphs. And that's what's

(05:15):
really happening by fast, or hasbeen happening in training when you interrogate a
chat GPT system. So the numberof parameters on these things has exploded as
well. The new system called Lamafrom Facebook and something like sixty five billion
parameters where each parameter is a knobthat can be turned and it's similar to

(05:38):
a wire in your brain, asynapse, and so sixty five billion.
Wow, that was unheard of evena few years ago. Now Microsoft has
topped out, sorry, Google AIhas topped that. Microsoft has new system
coming up with something called Palm andpalm has five hundred and forty billion,
more than half a trillion parameters,and they grow, and so the ability

(06:01):
to recognize linguistic patterns magnificent, betterthan ever, but it cannot explain itself.
And these are unlike fuzzy systems,which have an auto trail and give
you a confidence measure. These areblack boxing eye systems. So they often
work, but they also make thingsup. They hallucinate, and if you've
ever dealt with what these systems,you'll, I think you'll see that they

(06:24):
might have been at least a bitprematurely. And if you saw the recent
sixty minutes interview with the folks atGoogle and their system, they asked the
system about inflation, got an interestinganswer it depended on five economics textbooks.
But it turned out when they checkedit there were no such textbooks. The

(06:44):
system literally live. It made itup connected dots that didn't exist, and
did not mention that to the sixtyminutes down and that's something you've got to
be very very careful with. We'rejoined now by AI expert doctor Bart Costco.
So in the thirty years that you'vebeen studying artificial intelligence, aside from
just the rock computing power and whatwe're able to do because of that.

(07:05):
What are some of the biggest advancementsthat really stand out to you. One
of them would be the ability withlanguage processing, mapping text to texts to
do real well on exams, inparticular the bar exam, the attorney stake,
the uniform bar exam. Recently,some of these systems have hit ninety
percentiles. To think about that,it's pretty impressive. They have looked at

(07:28):
these exams ninety percentiles and they'll justincrease from there. That's something that was
a result again of just computing powerin a few tweaks, but at some
point across a qualitative threshold or tippingpoint, and it did. So we've
seen that a lot of that inthe last two to three years. Same
way with chess or looking for waysyou can fold proteins, and you can

(07:54):
come up with systems if you runit long enough. It's not really reasoning.
It looks like it's reasoning. It'simitating that. It's just checking a
lot of cases, and it cancome up with cases the case of chess,
where strategies where you'd give up yourqueen, for example, we just
wouldn't say if you play chess,you just wouldn't think to do that.
But it can turn out to bea winning strategy. And the current systems
can beat any chess masters, andthat's only going to increase as well.

(08:18):
So in cases like that, orfinding new ways to put together proteins,
we've crossed some thresholds. We knewthat was coming. We knew it was
coming thirty years ago, and alot more is coming still in the next
thirty years. Because computing power thatwe won't always have the current version of
Moore's law, it's kind of hittingits limits. There'll be other things,
different forms of quantum computing computing,some nanotech and some other things, and

(08:41):
we really don't know exactly what.We're shrinking. The chips ryan so small
now that we're getting down the onoff circuits to a couple three atoms across,
it becomes unstable. But we'll dosomething. We'll find something, and
the old algorithms and some new agorithmswill again hit tipping points. Medical diagnostics

(09:03):
just wonderful. I think that willget better. As I said before,
you can validate that. But thisother scarier stuff, the generation of text,
generation of speech, which characterizes ourspecies. That's something we're going to
have to come to grips with.When we see different applications, different programs
say that they're powered by AI.What exactly does that mean? Well,

(09:28):
you got to be careful because itcan mean so many different things. To
first order, any algorithm today thatruns on a computer that uses data is
called AI. You've got to becareful. Just give you one example.
There's a split politically between computer scienceand electrical engineering. Now this is inside
baseball that it matters in a university, and we've been doing this in EE

(09:52):
electrical engineering for decades. The computerscience folks shunned neural network famously, starting
in the late sixties with a bookby my good friend and late colleague Marvin
Minsky, an m I t calledcalled Perceptrons. He simply found that earlier
kinds of neural systems in the sixtieswere limited what they could do, and

(10:13):
people, I think through the wrongconclusion. AI went off in a different
direction, really in an anti neuralnetworks way. I helped set up Professional
Network Conference, which is in nineteeneighty seven in San Diego, and the
slogan at the conference was quote AIis dead, long live neural networks.
Well today AI equalfnural networks real quick? What is a neural network? A

(10:33):
neural network is a system where youteach it by training it rather than programming,
which your face looks like. Wejust give a thousands and thousands of
examples, and it has these parametersof a tunny but if you look at
it, it kind of looks likea little piece of a brain. You
have the input where you put yourface, and then there's some wires going
forward if it were physical, butit's actually going to sompware, and there'd

(10:56):
be another layer of switches like neuronson or off, and they would feed
to another layer, another layer,they're called hidden layers, and at the
output in your case for your face, or just be one neuron and it
either like a lightbulb, there's eitheron or off. And so we could
train to find your face in imageswhere it's partially included, where the light's

(11:18):
bad, where it's been randomly mixedwith your next door neighbor. And the
more training we do, the betterit gets. And we could have multiple
lightbulbs. You could have a thousandlight bulbs, each one for a different
face. Now, what's happening ina chat system is there are tens of
thousands of light bulbs at the output, but each one stands for a word.
Or you could have a video systemand each one stands for a little

(11:43):
piece of a movie for a videosegment. In general, each light bulb
at the output stands for a pattern, and the longer you train it,
the better it gets to go forwardand backwards. For example, recently got
another pattern on how to speed upin a training process, but it's very
intensive. It can take months offline for big problems, and it will
learn patterns, but just like theway you and I learn a pattern,

(12:05):
we can't explain how we learned it, and we can recognize thayings without being
able to explain that. A goodexample would be the music to Mission Impossible
or the music to James Bond.Unless you're really well trained to music,
you probably cann't write that down,but almost everyone can recognize that the assistive
memory property and the neural networks havethat. So today that is the power

(12:26):
the neural networks, which was avery tiny subset of AI thirty forty fifty
years ago, now it equals AIin an ironic sense. So you've got
to be careful. Another example ofthat is to take other algorithms like the
one we use to get the firstment of the moon, called something called
the calm and filters. It todaywould be lumped in with that. Now,

(12:48):
what that does is it allows youto figure out if you're stumbling in
the dark where you are based onyour current guests of where you are and
any kind of measurement. So forexample, on Apollo eleven, when the
folks were in space, they mighthave been talking to the President Nixon in
in nineteen sixty nine, and theastronauts they had some direct measurements coming from

(13:15):
optical sections that they were using outthe window, but they also had some
estimates of where they were from theonboard computer. And it was just the
algorithm was just efficient enough to squeezeinto that that old Apollo computer. Well,
that algorithm today is the essence ofsmart cards and many many other things.
But it will be called AI.It cames a surprise to a lot

(13:39):
of us, including to the lategrade Rudy Coleman, who came up with
a nineteen sixty so powered by AI. You gotta be careful. It could
mean something like that. It couldmean one of literally hundreds of alternative algorithms.
But at ROUTE it's combining computers withstatistics. The artificial intelligence that we're
dealing with right now, an AIchat GPT. These are computer systems that

(14:03):
are just filtering through a lot ofdata. Basically, the limitations to this
is what we allow it access toright correct exactly, and that raises lots
of problems. So it raises privacyproblems, it raises inllectual property problems.
I mentioned the case of Lama,the system and meta at Facebook. This

(14:26):
month there was a class action lawsuitfiled against it. Where everyone's watching this
and the illegal feel very carefully onthe West Coast for violating copyright right there.
Several authors. Now the discovery processwill show us how many books are
actually used. The comedian and authorSarah Silverman has a book used in that

(14:48):
I think it was called The Bedwetteror something. So she's part of what
we call the class representative. Butit's going to be an interesting insight because
there there's a lot of opaqueness,say at the big tech companies and authors,
if you're within the copyright bounds,have a right to assert for copyrights,
and these companies are supposed to getpermission. So there's some evidence that

(15:13):
there is at least I think fourhundred bucks verified so far that have been
used in training. And that's interesting. Your other information that you're putting out
online, you're basically impliedly consenting tolet others have access to it. I
would just say, as a fatherand as a lawyer, I would encourage
people not to participate in that sortof thing. I would I would limit

(15:37):
my exposure online. Not just purchasesand browsing histories you can possibly do it,
but anything you post voluntarily because thatcan be stitched together used against you
in really unforeseen ways in the future, and used in things like deep fakes
that you can't foresee. And therecan be mass I don't want to say
surveillance, but mass algorithms applications combiningyou with other folks or and it can

(16:03):
be abused at the political level.So for example, in twenty twenty four,
I think we're going to end upcalling that election the first AI election.
There's going to be so much analytictechniques going on from the opposing parties,
from possibly from China, from othercountries, from commercial firms in a
way that goes well beyond just predictinghow people will vote, but manipulating them

(16:27):
and you might not even know manipulatingyou in a coercive way, but maybe
just slightly changing the probability of howyou would vote. And that's sort of
a thing. It'll be interesting tosee how it plays out, and I
think that will be the future ofelections too. We're joined now by bestselling
author, world renowned scientist an experton AI, doctor Bart Costco. Is

(16:48):
the big concern when it comes toAI that eventually, somehow it will expand
beyond just the information that we're givingit and begin to make its own decisions,
think independently of what we're trying totell it to do. You know.
That's that's the terminator kind of viewof it, and it's possible that

(17:11):
could happen. It's hard to see, I think, how that happens.
It's much more likely somebody's behind thescenes of pulling the stranger's programming it to
do something like that, because inthe end, in the end rhyme an
AI system is still a bunch ofmuliplications and additions. It is nothing magical,
and it doesn't have an intricate system. It doesn't have a hormone system,

(17:33):
it doesn't have a built in lussfor power. It's not human in
that sense, but it can beprogrammed that way, and not just so,
rather than a big system that wakesup like we've seen in some magical
way and it takes over. Couldit could backfire, It could have a
mutation, a DNA mutation type effectwhere it evolves into something that goes off

(17:56):
untethered again, probably as a resultof faulty programming. That's a real risk,
but I think just the risk ofbad actors in greedy folk. I
mean, one example, there wasa piece of software I don't even want
to name it online that some youngstersput out apparently that would enable you to
take a woman's image and then modifyit so you could see or naked,

(18:18):
and kids were using it in school. Of the mass violation of privacy.
The laws have not adjusted for that, and once that's out there can be
replicated at zero costs. You know, it's around the world. So it's
that sort of thing multiplied over theyears. As computing power increases, as
sensing increases, as databases increase,we're not prepared for we have to watch

(18:41):
out for it from a commercial pointof view, though, I think it
is a great opportunity for entrepreneurs tocome in and address the problems of hallucination
and privacy laws and just you know, some kind of detector out there to
let you know when not just beingspied on and surveyed, surveyed which you
are, but to extent to whichsome thing or somebody is building a profile

(19:02):
of you. One last question foryou, the disruption to our economy,
to jobs, to entire industries.Are we just at the beginning of that.
When it comes to AI, weare at the beginning of that.
And the effect is having on somewhite collar workers. Lawyers, My lawyer
friends are finding it very beneficial because, for example, you can take a

(19:22):
lot of discovery documents, maybe athousand pages worth, and boil it down
without hiring people to do that.But some lawyers just found out federal courts
that you don't want to rely onchet GPT to write your brief because the
hallucination faces. Yeah, you getfacers. So as a decision support system,
wonderful to help people, will displacea few and it'll make more jobs.

(19:45):
It is a little dice here,and we look deeper into the future.
Certainly, candidates for automation are jobsthat are simple and repetitive, and
we have to do more schooling.I'm also worried that it's it will undermine
our education process, especially Kater twelve, because too many kids I see already
are using this to generate essays.The hardest thing I think for soon is

(20:08):
the right text, of original text, of high quality, and it comes
with practice, not by relying onchat GPT. Doctor Bark Costco, a
bestselling author, world's renowned scientist andexpert on artificial intelligence, is nineteen ninety
three international bestseller Fuzzy Thinking, HowAI Thinks in Shades of Gray, is

(20:29):
now being rereleased in a digital version. Doctor Costco can't thank you enough for
all the time and the insight intothis complicated issue. We appreciate it.
My pleasure. All right. I'mRyan Gorman here on ihear Radio Communities,
and now I want to bring inour next guest. This week, Discovery
Channel has been celebrating Shark Week onceagain, and we have with us a
featured expert on so many Shark Weekspecials. Shark expert doctor Greg Schomel is

(20:56):
with us. He's author of thenew book, Chasing Shadows, my life
tracking the Great White Shark, DoctorSchoemel, I've watched you so many times
on the Shark Week specials. It'sa real privilege to interview you here.
And I want to start with therelease of the movie Jaws, going back
a bit and find out what kindof an impact that movie had on the
public's perception of sharks, obviously includinggreat white sharks and the population of sharks

(21:21):
around the world. Well, Ryan, I'll start by saying what it did
to me. You know, Isaw it in a movie theater in nineteen
seventy five and the character is playedby Richard Dreyfus, Matt Hooper, was
inspirational. So well, it pullspushed a lot of people out of the
water. It pulled me into thewater. But it characterized like so many
Hollywood films, do you know,the white shark as a as a as

(21:45):
a villain, and as a mindlessand you know manager, And obviously that's
not what these animals are. Butyou know, I don't fault the movie
Jaws for the demise of white sharksor shark populations. Frankly, I just
look at that as an entertaining film. You know a lot of people think,
oh God, without Jaws, we'dhave fuenty of white sharks. And

(22:07):
the truth of the matter is iswe had an explosion in commercial fishing and
commercial shark markets, you know thatoccurred in the eighties and nineties that led
to the demise of shark populations.It really wasn't the film. Now,
the first time you're face to facewith a great white shark, what was
that experience like? For me?It was, you know, an absolutely
pivotal experience. You know, itwas a seventeen foot, three thousand pound

(22:33):
white shark and I was literally rightnext to it, and I'm saying,
I can't believe how big this animal. You know, it was like it
was like seeing it was like studyingwanting to study big Foot and then finding
one. It was amazing. Youknow. I've watched so many specials over
the years on Shark Week about greatwhite sharks, and one thing I'm always
struck by, despite all the attentionthey've gotten, there's still such a mysterious

(22:56):
creature. One of the things Italked about in the book is the fact
that I thought I could not becomea shark biologist because we knew everything there
was to know about these animals,and that's absolutely was not true and it
remains untrue. You know, someof the basic biological questions people have,
particularly in terms of reproduction. Forexample, you know, the white shark

(23:18):
here on the east coast of theUS and in the Gulf of Mexico,
we really don't understand or know whenand where it reproduces. Where to males
mate with females, How long dothe developing young inside the female? How
many developing young young wills she have? And where did she give birth to
those young? You know, that'sa really those are simple questions, but

(23:38):
they're very difficult to answer. Whyare they so difficult? How come we
don't have a better sense of themating habits of great white sharks? Well,
pretty much the same reason we don'tknow a lot about the mating habits
of most shark species. It's becausethey're highly migratory. They live in the
ocean. You've got to be inthe right place at the right time to

(24:00):
actually observe this kind of behavior,and that just doesn't happen enough, you
know, So we try to useindirect methods. For example, if we
examined a lot of pregnant females overthe course of the year, we would
get a sense of the timing oftheir when birthing occurs in the development of
these young truth is, we've neverseen a pregnant female white shark in the

(24:23):
Atlantic Ocean. We're joined by sharkexpert doctor Greg Schomel, author of Chasing
Shadows, My Life Tracking the Greatwhite Shark. Another thing I found really
interesting about great white sharks how theyhunt and depending on their location, the
different types of tactics they use.I believe it's off the coast of South
Africa. That's where they come upfrom below the seals and they do those

(24:48):
breeches out of the water, attackingtheir prey that way. But then I
was just watching a show that youwere on up in the Cape Cod area,
and they have a completely different huntingstyle. Yeah, Ryan, I
love this question because it allows meto, you know, to talk about
how plastic these animals are in termsof adaptability. You know, they're they

(25:11):
go in South Africa. The environment'svery different from Cape Cod, which is
very different from Mexico, right,but it's the same species, and so
they adapt their their hunting behavior tothat environment. So you mentioned you know
the missile launch that they go asthey breach, pinning seals against the surface
as they strike from deep water offSouth Africa. In Cape Cod, we've

(25:33):
got really shallow water, it's murky, it's there's shifting sandbars, there's heavy
serve most days. It's a reallychallenging and very different environment for white sharks.
And they're ambush predators, so theyhave to adapt their strategy, you
know, when they hunt off CapeCod in order to ambush their prey without
their prey seeing them, and it'sdifficult. You know, they have to

(25:55):
strike from the side. You don'tsee them launch themselves like you do in
South Africa. But it's really coolbecause the species adapt to its environment.
As I'm sure you're well aware,there's been a lot of focus and a
lot of talk about our oceans recentlyfollowing the tragedy involving the titan submersible and
how far down our oceans go withTitanic wreckage twelve thousand, five hundred feet

(26:18):
below the surface, how deep dogreat white sharks go? Do we know
the depths that they swim to?Yeah, I think the new technologies that
we're using over the last couple ofdecades have really opened our eyes to white
shark behavior, a lot of sharks, a lot of species behavior. We're
now able to put satellite technology andsensors that will detect temperature in depth.

(26:41):
And what we found out we publishedthis a few years ago, it's obviously
going to be is in the book, is that white sharks, when they
move off our shoreline and go outinto the open Atlantic, will dive to
depths as great as three thousand,thirty five hundred feet deep every day.
Are really really broad temperature range.Now they're not getting as deep as the

(27:03):
Titanic, but we always thought ofthese sharts as being coastal animals. If
the new technologies tell us no,they actually go to parts of the deep
ocean. We're joined by shark expertdoctor Greg Schomel, author of Chasing Shadows,
My life tracking the Great White Shark. Final question for you. We
think of the great white shark asthe apex predator in the ocean, but

(27:27):
I've watched a number of specials that'sshown where that's not always the case,
especially when it comes to orcas killerwhales. Can you tell us what we
know about that? Well, there'stwo you know, we always think of
the white shark as being the toppredator. But yeah, that's that's we're

(27:48):
learning that's not the case. Firstof all, you know, it has
a couple of predators. We thoughtit was primarily manned, right, white
sharks are largely killed by people moreso than anything else. But recently,
over the last decade, we've figuredout also that the killer whale, the
orca, is a predator of whitecharts and it literally scares them out of

(28:08):
the area when they are when whenorcas are around, and they're capable of
killing white charts and targeting their liver, which is absolutely fascinating. So you
know, every time you think there'sa top predator out there, there's going
to be something, you know,a little bit bigger, a little smarter
that's going to take it out.And finally, where can people find out
more about your conservation efforts because yourfront and center on all of this,

(28:32):
and I want to make sure everyonehas that information too. Thank you.
I appreciate that the I work reallyclosely with a nonprofit called the Atlantic White
Shark Conservancy, And you know thebest way to follow us in what we're
doing is to go to their socialmedia page. Is whether it be you
know, Facebook or Twitter. Youknow, we show a lot of our
films of what we're doing out there. You know, obviously you've seen me

(28:53):
on Stark Week Sharks Fast. Youknow, I love those as vectors to
get out what we're doing to thepublic. But you know, Conservacy is
a great website. Doctor Greg Skomela featured Shark Week expert and author of
the new book Chasing Shadows, Mylife tracking the Great White Shark. Doctor
Skomel, really appreciate time and insightand all the work you're doing for sharks

(29:15):
all around the world. Thanks somuch for coming on the show. Thanks
Ryan, great to be here.All right, and that's going to do
it for this edition of iHeartRadio Communities. As we wrap things up, I
want to offer big thanks to allof our guests and of course to all
of you for listening. If youwant to hear previous episodes of this show,
we're on your iHeartRadio app. Justsearch for iHeartRadio Communities. I'm your

(29:37):
host, Ryan Gorman. We'll talkto you again real soon.