Video summary
The video introduces the concept of "belief landscapes," a metaphorical model used to understand how populations update their beliefs over time. Instead of viewing misinformation as an isolated problem, the speaker argues that people's existing latent beliefs act like gravity on a topographic map, pulling individuals toward stable valleys of thought while others remain in turbulent areas or move freely. By analyzing vast amounts of text data from news and social media using natural language processing, researchers can locate where groups of people sit on this landscape and predict their future belief trajectories with surprising accuracy. This approach allows scientists to identify converging regions, whirlpools of polarization, and areas susceptible to misinformation, offering a dynamic way to study how collective intelligence evolves rather than just looking at static opinions.
Beyond mapping current beliefs, the discussion extends to the challenges of aligning artificial intelligence with diverse human values, a field known as value pluralism. Since AI models are often trained on data reflecting specific cultural biases, such as Silicon Valley's liberal worldview, they may inadvertently exclude or misalign with users from different backgrounds. Researchers are currently exploring how to adapt these models to recognize the validity of various cultural perspectives, particularly regarding complex topics like gender and family dynamics. The goal is not to create a culturally neutral AI, which is impossible given that all advice stems from some context, but rather to ensure that technology serves diverse populations by understanding that what constitutes "right" or "wrong" depends heavily on one's cultural environment.
The High School at Syracuse University serves as an interdisciplinary hub where technical computer science meets social science, fostering a unique environment for students and faculty to tackle real-world problems. Through initiatives like the C4 Lab and the Center for Computational and Data Sciences (CCDS), the institution connects students with industry leaders and local communities to work on projects ranging from predictive analytics in agriculture to understanding belief dynamics. Students engage in hands-on research without needing funding, collaborating on diverse projects that build technical skills while addressing human issues. This entrepreneurial spirit encourages cross-pollination of ideas between those focused on code and those focused on society, creating a rich ecosystem where students learn to manage AI tools wisely rather than relying on them blindly.
Ultimately, the conversation concludes with advice to remain critical of media hype and to engage with new technologies like AI as tools that amplify human creativity rather than replace it. The speaker emphasizes that while the world can seem scary due to rapid technological change, a closer look reveals nuance and great potential. Just as previous innovations transformed how we travel or play music, AI offers new ways to solve problems if approached with an open mind and critical thinking. The core message is to avoid hiding from these changes but instead to understand their structure, use them as targeted solutions within broader frameworks, and appreciate the unique opportunity to bridge the gap between technology and humanity.
Read the full video transcript
Hello, I'm Jeff Hemsley and this is
another episode of Infoiversity from the
High School at Syracuse University.
Today I'm joined by my colleague Josh
Intron. He's an associate professor here
at the high school at Syracuse and he's
the director of the C4 lab and CCBS.
Josh studies how social technologies
shape collective intelligence and belief
formation, including large-scale work
like belief landscapes, which is a
computational social science project
that he's going to tell us more about in
a bit. In fact, let's just start right
there. When you talk about belief
landscapes, what does that mean and how
are you studying that?
>> Yeah. Well, so belief landscape is a
metaphor, right? of course um for trying
to understand how
populations of people update their
beliefs over time. So um let me give you
a little bit of history on that and then
maybe I'll I'll walk you gently into
belief landscapes. So
um we all know that you know or or many
of us here in the high school have uh
focused on misinformation
and uh misinformation is really about
the proliferation of information that is
that is not correct right people are
putting it out there and u but I've
always been wondering I've always been
interested in in in trying to figure out
why people believe in misinformation
and it struck me that Maybe the problem
isn't the information itself, but it's
really the uh latent beliefs that people
hold that lead them to buy into that
misinformation. So, uh I think about 10
years or so ago, I started um looking
carefully at beliefs, trying to model
beliefs. And um the
metaphor or model that I came up with is
that uh beliefs could be organized as if
they were on a landscape and the
population is moving over that landscape
as they change their beliefs. Okay? And
you can kind of another another metaphor
is to think about um beliefs as being a
current in a river, right?
>> Wait, can you give us a real world
example in the like in today's
landscape?
>> Yeah. So, um you can kind of think about
uh conservatives and liberals as being
different positions on that landscape.
Of course, there's not just conservative
and liberal, right? There are lots and
lots of little clusters of beliefs here
and there. And so if you imagine instead
of beliefs moving between people, you
have people moving between beliefs, you
can begin to think about how populations
move across that landscape of beliefs,
right? And so, um, I went into this with
that model in the back of my head and,
uh, through a lot of natural language
processing and some really fancy data
analytic techniques, I was able to show
that this model works pretty well. And
in fact, it works so well that once you
can locate people on the belief
landscape, you can predict with a fair
degree of accuracy where their beliefs
are likely to go in the future. So you
can't necessarily do this with 100%
accuracy for an individual, but when you
start to look at populations, you can
say, "Oh, this group of people who hold
these sorts of beliefs will go here in
the next half a year, in the next year."
And it's surprising how accurate that
is. So the landscape metaphor which is
you know
maybe helps to think about is you can
imagine a rough topographic landscape of
beliefs and we'll imagine there are
stable positions on this landscape and
those are the valleys of the landscape
and the population of people moving
across that landscape are governed to
some degree by gravity
>> and so they flow into these low regions
and they tend to stabilize there. They
tend not to move back out of those. Now,
of course, people are, you know, unique.
People are all over the place. And it
turns out that there are some people
that are pretty stable and there are
some people that are just wild and
they're moving all over the place. Um,
most people are fairly stable. Um, it's
really a a small periphery of people
that move all over the place.
>> So, in my mind, what I what I kind of
envision is like a map with different
cities and each city represents a belief
space except some cities or some belief
spaces have more gravity or are more
attractive to people.
>> That's right.
>> Is that kind of
>> That's exactly
>> in the ballpark. That's right. So, now
to mix metaphors a little bit.
>> Yeah. Um, if you take that back to say
the uh stream metaphor or the current
metaphor, there is actually a lot of
movement in this landscape. People are
in motion all the time. But in the same
way, you don't really see those streams
directly until you drop something in the
water and you begin to see how those
particles move, like leaves on a stream.
That's kind of how I do my work, right?
Is I model where people are and I look
at their collective movements. And so
you can begin to identify things like
converging regions, little whirlpools,
areas where there's a lot of turbulence.
And then you can look at how that
changes over time. You can look at the
emergence of turbulent areas or new
strong whirlpools emerging. So this
becomes a really powerful way to look at
things like polarization
and belief dynamics.
>> Yeah. So as you talk I think about some
sort of per perturbation which probably
equates to events that happen in the
real world that can shift things.
>> That's right. And so part of what I'm
doing in my current project is trying to
measure the um
fragility of an existing belief
landscape such that when something
happens will we see a massive movement
or a shift in the beliefs of the
population. So you might not be able to
predict COVID, right? But you might be
able to say that, oh, we're poised to do
some pretty interesting things when a
pandemic uh comes along.
>> So, the project that I'm looking at is
looking for the uh you know, can we can
we predict
um where social unrest is likely to you
know, arise in the wake of certain kinds
of events.
>> Okay. So in layman's terms,
>> how do you model this? How do you study
this? Like what kind of data might you
use? What what does that look like?
Right? So the data that I use is any
text data. So I look at anywhere where
people might be voicing some kinds of
beliefs. So this could be news media,
social media of course. Um,
and but there are flavors of social
media, right? I'm not going to be
looking at YouTube videos typically
because it's hard to lift beliefs out of
that that. But any text data works for
me.
>> Um,
>> could you use Tik Tok data if it was
transcribed?
>> If it was transcribed. Yeah. Yeah. Um,
one of the things that's interesting is,
you know, if you look at things like
local news organizations, you can do a
pretty good job trying to understand the
beliefs of local populations by looking
at just those news organizations.
And so you can begin to understand how
beliefs uh are distributed
geographically as well.
>> Okay, that's pretty cool. So now I know
another space that you're looking at is
AI and culture and you have some PhD
students working in that space.
>> Um can you tell us about that?
>> Yeah, sure. Um, so I think uh
I don't know if Google coined this
phrase, but um there's this idea of
value pluralism
in AI right now, which is that people
have lots of different values. What's
right, what's wrong is really dependent
on context and culture. um there is no
you know one-sizefits-all set of values
but most of the AI models that are
generated are generated from you know
handful of tech companies out there and
they plug into those tech companies a
set of values which might not fit
everyone right and so here in you know
the US we've got uh you know models that
are generated in Silicon Valley and
they're channeling usually liberal
western
progressive values and those don't even
work that well in the US right there are
lots and lots of people with different
ideas and so you have to ask the
question what is you know is this are
these kinds of models excluding large
portions of the population because
they're coming into things with these
bakedin values that don't really match
the end users so what we're trying to do
or or my my two uh PhD students are both
looking at aspects of this. One of them
is looking at um values around not know
values is a little narrow actually.
She's looking at conceptions of gender
across different cultures trying to
understand how different AI models uh
interpret and promote different views of
gender. Um and then my other student is
uh this Jinfen is is looking at how you
can actually adapt AI models to u do
something that recognizes the validity
of the values of the enduser.
Right? And so the whole point here now
there's a the larger question uh you
might have heard of the the alignment
problem in AI which is how do you make
sure that the AI is aligned with human
values right but the question then that
begs the question what values are we
talking about here there is no onesize
you know fits all set of values so we're
trying to approach that problem by
saying well you
Since we can't say there's one
sizefitsall, one set of values, how do
we align AI with different populations
of users that are coming from different
cultures, different backgrounds?
>> Okay. So, from what you're saying, it
kind of sounds like you couldn't make AI
to be culturally neutral, or could you?
It seems really hard to do that as soon
as you talk about anything that has, you
know, if you want to ask AI for advice.
At that point, it's really hard to be
neutral, right? From a uh collectivist
culture, you know, in the east, you
might want to ask, should I move out of
my house? Right? And from a western
perspective,
move out of your house, get a job, do
something. From an eastern perspective,
spending time taking care of your
parents, your aging parents, is really a
laudable thing to do. And so these
values are going to inform how the AI
responds,
>> right?
>> So let's briefly go back to belief
landscapes.
>> Yeah. So how is that an high school
topic and how do you engage high school
students in that kind of work?
Yeah, it it intersects with the high
school in a couple of interesting ways.
Um, from a method standpoint,
uh, I am doing things that high
schoolers do. We're processing language.
We're looking at social media. We're
doing computational social science and
we're trying to understand the
implications there. It's not necessarily
that we're looking at how social media
influences beliefs yet.
>> However, you know, I started with
misinformation. The question then
becomes, can we identify places that are
especially susceptible to misinformation
and what kinds of misinformation are
they susceptible to? So, you know, maybe
we're never going to stem the tide of
misinformation. It's been around
forever. It's becoming more and more
powerful. I don't see how we're ever
going to get rid of deep fakes, right?
That that that uh you know train has
left the station. But maybe we can do
something to identify susceptible
regions of a belief space. And based on
this analysis, maybe we could say, "Oh,
this is a kind of belief space that is
highly susceptible to misinformation.
This might be uh you know in another
situation another another set of beliefs
might be a highly resilient set of
beliefs. It's not like we're trying to
change anyone's mind about anything but
maybe there are ways we can introduce
new sorts of dynamic processes which
lead people to be more resilient in the
face of bad information. So that's one
place where this um you know this sort
of intersects really strongly with the
high school and with information
sciences in general. another my my goal
here to some degree um we do do things
like try to understand polarization
right maybe that's political science
maybe it's information science maybe
it's communications but it's somewhere
in all of those places I think there are
a lot of different ways to characterize
the dynamics of belief that go on in a
population that are not just
polarization Right? Polarization is
pretty simple. It's based on the idea of
a two-pole axis of beliefs. But that's
not how beliefs actually work. There's
fracturing, right? There are beliefs
that are converging on some dimensions
and moving apart on others. There are
beliefs that are frozen. There are
beliefs that are mixing. So using the
belief landscape analysis, I develop a
much richer vocabulary to describe these
dynamics. And from there we can do a lot
of other things to think about our uh
information ecosystem and how it
intersects.
>> Yeah, I like how you talk about how your
work is actually intersecting with a lot
of other fields
>> and that's actually kind of a key thing
about high schools, right? We kind of
sit in between a lot of other spaces,
>> right?
>> So students that are interested in
multiple kinds of things can find work
like this interesting from a lot of
different perspectives.
>> That's right. It's highly
interdisciplinary and we're always
moving into this theoretical literature
or that theoretical literature and
pulling down new new techniques.
So tell us about CCDS. What is CCDS? Um
what's your role there and again how
does that fit into the high school and
how do students get involved? Yeah. So
this year I am the director of the
center for computational and data
sciences.
Uh CCDS has been around for quite a
while. Um but I think one of the things
that I'm hoping to do with CCDS
um is actually turn it into something
more of a uh a connector, right? that
connects uh students in the high school
and faculty in the high school and maybe
people across the campus with
others in uh in the area like in central
New York, right? Industry leaders
um who all share an interest in doing
some sort of data analytics, right? So
we have this uh really dense
concentration of people with expertise
in data analytics and AI and in the high
school we're really interested in realw
world problems right
>> and so there's a beautiful marriage
there between what's going on in the
high school what's going in on in
central New York at large what's going
on across the campus with people
processing data
And so I envision CCDS to be a nexus of
sorts where all of those people could
come together. So through that we can
provide students with real world
opportunities to get involved in data
analytics projects, maybe help them
extend their networks, right?
uh and then we can also centralize data
repositories and perform services for uh
others you know out out in the world or
on campus. Um and hopefully one day we
could establish new sorts of research
relationships, engagements between
faculty here in the high school and
people trying to solve real world
problems with AI, automation, data
analytics, forecasting.
>> So now you also are in the process of
trying to build some industry
connections.
>> Yeah.
>> Are those things you can talk about?
>> Yeah, sure. Um, so we have a uh contest
coming up soon. This was a uh a
connection we made through made through
an high school student, a current high
school student. Um, there's a uh company
in the U
Midwest called Professional Agricultural
Marketing.
um and they're running a contest here in
the high school for high school students
to participate in
um addressing a couple of the challenges
that they're facing with the deployment
of AI and forecasting predictive
analytics. So, uh they reached out to us
and we said we'd love to do that. And so
we have I think um a total of 16 student
teams starting right now. Um we have a
kickoff meeting this Friday and uh
they're competing for prizes but also
they're gaining a lot of experience with
real world problems. Um, and then I've
been reaching out to MACE, which is the
Manufacturers Association of Central New
York. Um, and I think we are going to be
running a webinar in early April and I'm
working to try to build bridges into uh
into industry more generally. Um,
there's a center at SU called the Case
Center. um and they've been helpful in
trying to provide me this contact. So,
it's work in progress. I feel like we're
a little startup here in the high school
right now.
>> Yeah. Well, and that's kind of exciting.
And in fact, the high school has a long
history of being entrepreneurial and
starting up things and creating
connections. So, it kind of fits right
in.
>> Yeah.
>> So, you also have a lab called the C4
Lab.
>> Tell us about that.
Well, the C4 lab uh I started when I
first uh I'm a I'm a you I have a PhD in
computer science and I I did all my work
in a lab environment and I like having
lots of students working together. I
think there's a lot of synergies there
and so when I got to the high school I
wanted to replicate that environment. So
I started the C4 lab. T for stands and I
I I don't even remember the ordering but
it's uh communication computation
cognition complexity that's C4 and uh
those are all elements of you know of my
interests in my work um and so I have
you know at any given day
anywhere between 15 and 30 students
working on different projects in the C4
lab. A lot of students come to me and
they say, you know, professor in drone,
I'd love to work on a research project.
I say, I don't have any money. And they
say, that's okay. I really want some
experience here. I really like the idea
of learning outside of a traditional
course. And so I find spaces and
projects. Students come in and they
drift out again. you know students are
busy but you know quite a few of them
get some good experience and they stick
around for a while. So within the belief
landscape project I probably have five
or six students that are working on
different aspects of the project. I have
students working on a another project, a
narrative project. I have students
working on another project which is
building an app. Um, and so it's a
really rich environment and I get a lot
of a lot of good collaboration and
synergy there.
>> Okay, I'm going to switch to teaching.
>> I want you to tell me um what classes
you're currently teaching
>> and you know in education right now
there's tension around students using AI
to learn things and using it in class
and using it. what's your philosophy
there
>> and and how do you see students applying
this in useful ways in your context?
>> Yeah, so right now I'm te teaching a
graduate course in uh applied machine
learning. Um I'm only teaching one
course this semester because my u
project is pretty demanding
um the funded belief landscape work. Uh
but uh I've been teaching this class for
a few years now. It's evolved a lot
especially over the last few years as AI
as you might imagine uh has has become
really really prevalent. People are
using AI all the time.
Um, as a programmer, I have about four
decades of experience programming and I
use AI all the time to do my
programming, right? I'm a a good coder,
but it's just so much faster, right? And
I think that moving forward, we're going
to see students using AI more and more
and more. And there's really
no way to say don't don't don't do that.
I think it's possible to use AI and use
it wisely to do really powerful things,
but that's the trick, right? Is using it
wisely because you can say to an AI, do
X for me. And if X is too large or
you're not paying attention,
you have no idea what you're getting,
right? And so what I'm doing with my
applied machine learning class is
focusing really on core theory to try to
make sure students are understanding
what's going on. But then within the
coding part of the class, I try to
scaffold interactions with AI so that
people students become very effective
technical project managers. Right?
That's one of the things that I think uh
requires a lot of sophistication and is
not well addressed in computational
classes. It's kind of software
engineering, but it's also a lot of what
you call computational thinking.
thinking about how you break code down
into modules, how you make sure that
those modules are performing correctly
and how you adapt the code as you move
forward. That's the kind of thing I'm
focusing on right now in class. So
rather than sort of put my head in the
sand,
um I think uh grappling with AI directly
and trying to teach students the new
skills that they need in order to make
sure they're using AI wisely in the
course of doing technical work is um
that's that's my philosophy.
>> So what if there's one thing you want to
make sure your students leave with with
respect to AI, what is it?
H well
I think some some you know condensation
of what I just said. Um
even though uh you can use AI to solve a
big problem right now, it's important to
keep looking at your problems critically
and understanding how they're structured
and using the AI as a targeted solution
within that structure. So my hope is to
make sure that AI doesn't lead students
to disengage critically with the
technical material that they're
grappling with. So I hope to preserve
that critical thought that goes into
these sorts of uh technical disciplines
even though they have AI available to
them if that makes sense.
>> Yeah. So another thing that I think a
lot of students don't realize is there's
many times we learn as much from them as
they are from us.
>> Talk about that. What what are you
learning right now from your students?
And I know you work with undergraduates
and master students and PhD students. H
so
it's it's hard to put it in a in a
nutshell because I learn different
things from different students.
>> Sure. Right. Um,
a lot of,
uh, what I learn is,
um, how to articulate,
uh, things that I have just sort of
absorbed over time, but I never really
turned into uh, some sort of lesson. I
have a lot of intuitive approaches for
dealing with things. And I'll say to my
students, you know, when I'm first
starting out with a CLA class, I'll say,
"Do this. It's obvious, right?"
And the students tell me, "No, it's not
obvious." And so that helps me
articulate exactly what it is that you
need to do in order to do X, right?
Whatever.
>> Sounds like good prompting.
>> Yeah, it's excellent prompting. And so I
have to sort of step in it first and you
know figure out oh I just said something
is completely uninterpretable
and then you know I I once I get the
feedback I'm able to fix that. But then
you know the other on the other side
when I'm working with uh you know PhD
and master students outside of the
classroom what I get out of that is just
the benefit of diversity right you're
only one person and and as one person
you're going to think some subset of
things but as soon as you got a bunch of
other people working with you they're
going to show you angles to problems
that you haven't even thought of. Yeah.
And that's sort of the magic of working
with large groups of students in the C4
lab.
>> All right. So, why should a student come
to the high school? What is it that we
do that's just really different that's
unique?
>> Well, um
I have a particular
I have a particular answer. My answer
might not be anyone else's answer. Uh
I'm a I'm a computer scientist, right? I
come at things with a fairly technical
approach. Uh that computational
thinking, right? But uh I really
uh the things that drive me are the
human problems.
>> Yeah.
>> Right. So I really bring my more
technical approach to human problems.
And that doesn't mean I reject other
modes of thinking. In fact, I I love to
be exposed to those other modes of
thinking because that's the grist that
helps me to learn over time. So, you
know, many of the things many of the the
the new technologies we're grappling
with as um as as a society
um as a country are, you know, what
makes them hard is not the technology,
it's how the technology meets people.
Yeah.
>> Right. What all these systems, all these
human systems we have are going to
change dramatically. Right. You can talk
about the future of work. You can talk
about AI and medicine. You could talk
about education. Right? These are all
the kinds of problems that the high
school deals with. And really nowhere
else do you find that, you know,
academics that are solely focused on
that that that interstitial layer
between technology and humans. And we
come to it, many of us come to it with a
somewhat more technical background, but
also an awareness of social sciences and
theories. And so that blend of things is
really unique and I think maybe one of
the most important places people can be
working right now. Right? You can throw
up your hands and say, "Uhoh, what's
going to happen with, you know, the job
market, with AI? What's going to happen
with crypto?" Right? Well, yeah, that's
exactly what we're studying. That's
exactly what we do in the high school.
>> Yeah. And to add to that, this is the
place where a computer scientist like
you can work with somebody who does
completely social focused work. And so
you're going to get those other
perspectives that you were talking about
earlier, right?
>> That you know, the way you think might
not be the way I think or might not be
the way Ingred thinks or might not be
the way Steve thinks. And having all of
us together in the same room sometimes
generates surprising results.
>> Right. Right. Exactly.
>> So Josh, do you have any last words of
wisdom or semi-wisdom for us?
>> I don't know. I'm not the wisest guy in
the world. Um
I don't know. Uh
I think uh
you know
the world is scary right now for lots
and lots of reasons.
But um
very often what you see in the media is
sort of dismal, right? It's like, "Oh
gosh, this is just how is this ever
going to work?" Um
uh and you got to be you got to steal
yourself a little bit to walk into it
because you see some pretty some pretty
scary things. But I think when you get
close to what's actually going on, when
you move through the
media narratives that are designed to
alarm so as to engage,
>> y
>> you find it's actually a little bit more
nuanced than that. There's a lot of
interesting stuff, a lot of great
potential, a lot of scary stuff. Um, but
I think uh if there if there's a word of
wisdom there, it's don't believe the
hype.
Look carefully. Try to understand what's
going on. Try to engage with the new
technology.
Don't get beaten into submission and
hide.
>> Yeah. I guess so. One of the things that
I think about based on what you were
just saying or that that comes to mind
for me is AI is a new technology but we
have weathered many new technologies and
a lot of times it's not so much that
things are better or worse they're
different and we've got to be prepared
for that differentness.
>> Yeah, I think that's right. I you know I
just as as a cod to that I was part of a
little email thread my my father and uh
one of the one of his his friends said
well here's how Hal is here now but but
that's that's not right it's not it's
not that Hal is here right now and it's
not that AI is going to take every
single job there's so much more beyond
that horizon I feel sometimes like with
AI I've gone from you know playing uh
you know the the cello to being a
conductor of an orchestra.
>> Yeah.
>> Or instead of taking a trip on foot I
now have a pocket lejet and I can travel
vast different distances. So it
amplifies my creativity and my ability
to do things in ways that are completely
unique. Um, and so there's that flip
side and I think that I' I'd love for
people to approach it with some of that.
>> Great.
>> Josh, thanks for coming and talking to
us today.
>> I think it's been great.
>> Yep. Thanks. Thanks.