00:00:00
welcome welcome everyone to the latest edition of the analytics at Wharton AI at Wharton podcast series we're doing a
00:00:09
series here on artificial intelligence and today's episode is looking to be extremely exciting I'm happy to be
00:00:16
joined today by two of my colleagues uh the first is Zab Elizabeth Johnson who's
00:00:20
the executive director and Senior fellow of the wart Neuroscience initiative and
00:00:25
my colleague Michael Platt who this will take a minute listeners who's the both
00:00:29
the faculty director of the Wharton Neuroscience initiative he's the James S
00:00:32
ree Prof pen integrates knowledge Professor he's in my home Department of marketing he's also in the department of
00:00:38
Neuroscience in the Perman school of medicine and he's also in the psychology
00:00:41
department in the School of Art Arts and Sciences so Michael and Zab Welcome to our
00:00:46
podcast thanks for having us thanks Eric it's great to be here well why don't we
00:00:50
start with the basics I know Zab I'll start with you um for our listeners that
00:00:54
don't know of course they can go to your website and see all about it but what is
00:00:58
the wart Neuroscience initiative and then we'll get into what it has to do with our episode today Ai and
00:01:03
Neuroscience great uh so the wart Neuroscience initiative is a research center under analytics at Wharton here
00:01:11
at the Wharton School um and really what we're doing is to help uh businesses
00:01:17
individuals uh and Society RIT large understand why Neuroscience might impact their lives um so so we're working um at
00:01:26
at in on all different levels uh both education of course here we are at the University of Pennsylvania um trying to
00:01:33
lead the charge in in in encouraging curiosity um and engagement in the neurosciences RIT large we're not trying
00:01:41
to change business students into neuroscientists but we want them to be aware um of of what the power of of
00:01:47
looking under the hood is um at brain activity itself and uh and then we have research
00:01:53
active res an active research portfolio a lot of times we do that in in conjunction with companies um to answer
00:02:01
and uh and think about questions that haven't yet had an academic and practice
00:02:07
partnership um to do bigger and and better things out in the wild um and of course engagement is like our third
00:02:14
really our third pillar it's um it's encouraging people to think very broadly
00:02:18
about this three-pound organ that they have in their inside their heads um and to think about how that might um impact
00:02:24
their their lives their own individual work and um and how they live their lives so Michael maybe a question for
00:02:31
you I've always said that you know people ask me what I do for a living and I say I'm a professor I'm a statistician
00:02:35
but I really say I'm a professional data Chaser I chase interesting forms of data
00:02:40
could you tell us since you and I are the same age we've been in Academia the
00:02:43
same basic amount of time how has technology changed the field of Neuroscience because you know when I was
00:02:48
a young researcher you had to put people over in an fmri machine if those even existed you'll even tell me if those
00:02:54
existed 30 years ago we had to people in an fmri to get brain activity how is technology just changed even that aspect
00:03:01
what you do it's a great question and I think that what what's interesting from
00:03:05
the perspective of Neuroscience is the kinds of data that we get right so it's
00:03:09
it's not directly observable data it's not data that people can um verbally
00:03:14
Express typically because it's like we don't have good access to what's going
00:03:17
on in our heads and and just the question just by being asked that changes our appreciation of that uh
00:03:24
technology and Neuroscience is bewildering now so it is just every year every day just exploding the the Myriad
00:03:32
ways in which we can measure and actually manipulate um brain structure and brain function the vast majority of
00:03:39
those Technologies are not really readily applied to humans because they would involve you know putting something
00:03:45
in your head now that said there's an active race in the private sector even
00:03:49
right now you know neuralink being you know one amongst many companies that is creating and building implants with a
00:03:56
vision that someday you know maybe all of us will be uh perfectly happy to have uh some sort of Technology within our
00:04:03
heads that could allow us to communicate with um machines computer com communicate excuse me with computers
00:04:11
even communicate with each other but uh fmri still exists uh it's been around
00:04:15
for some 40 years but um it is still a really great tool for peering deep into the brain uh but you know it's expensive
00:04:25
it's cumbersome it's not very scalable you can't put it on a consumer's head
00:04:29
while they're walking around you know shopping at Walmart or on your employees
00:04:33
head while they're you know while they're at work so uh so we reserve it
00:04:37
for specific kinds of of studies like to test hypothesis about you know why did somebody make a risk adverse decision
00:04:44
versus not it provides kind of a foundation that we try to do uh from there is to use that as a springboard
00:04:51
for applying other more scalable Technologies um that can be done you know with many more people in a lighter
00:04:57
weight cheaper kind of fashion that's more uh more useful for business right
00:05:03
and I think that's where we're in a really exciting place in the last decade
00:05:07
was especially the last couple of years which is the development of very high uh
00:05:12
High signal quality wearable neurotechnology and so there's a whole variety of of gizmos that are on the
00:05:20
market or coming to Market and um and I think and we'll talk about this some
00:05:24
more obviously but but yoking that combining that with advances in AI and machine learning puts us in position
00:05:32
to um really capitalize on the ability to you know to under to measure brain activity in the real world you know at
00:05:40
scale thousands maybe millions of people uh in a much wider array of activities than is possible in the laboratory so
00:05:50
that's going to give us incredibly new insights I believe so one of the things
00:05:54
as both of you know that we asked you to do before this episode was to write a set of questions that I could ask you
00:05:59
and this is episode is no different matter of fact Zab and I joked before you got here Michael I she joked with me
00:06:04
like who do you think wrote these questions us or chat GPT which is a perfect segue to my first question so um
00:06:10
either one of you can answer this I don't matter of fact this is one of those times I'm asking a question where
00:06:14
I actually really really don't know the answer so what is organic intelligence
00:06:19
and how is this different from artificial intelligence so Zab I'll start with you what is what is organic
00:06:24
artificial what does that mean to you as someone that's trained in Neuroscience
00:06:28
so I think it starts with just thinking about organic matter right so so in general organic intelligence is used in
00:06:35
the domain when it's carbon based um so that's what I have yeah so that's what
00:06:39
you have the wet the wet gooey stuff that's inside right but also all of the
00:06:44
animal kingdom has that um so so you can argue about intelligence right and and different metrics of intelligence and we
00:06:51
can talk about that later but but in general it's it's this idea that it you
00:06:56
know that you have a nervous system that's carbon based um that that you know has a certain kind
00:07:02
of behavior um and integrates signals in a certain kind of you know chemical and
00:07:07
electrical um system that's carbon based and I think in opposition to what we
00:07:14
classically think of as ai ai is done in silica right um I mean that's the way it
00:07:19
has been so far we'll see um if uh if we start to grow to to grow artificially um
00:07:26
in the lab using wet stuff um that might that might change it might get in it you
00:07:31
know and I think some people would argue that there is already still carbon in involved but I think the the sort of
00:07:38
root of what people think of as this definition is is based on actually having a nervous system rather than not
00:07:44
having a nervous system and doing it artificially and so another interesting question that you guys put down is what
00:07:50
are some common traits shared between AI systems in the human brain so or let me
00:07:56
ask another question should people that are building artificial IAL intelligence
00:08:00
systems today do they have a team of neuroscientists working with them because in some ways if one of the goals
00:08:07
of artificial intelligence is to mimic human intelligence shouldn't we actually
00:08:11
understand how the human brain works before we try to build systems that are trying to mimic some aspects of that
00:08:17
well I think there's several questions here which is what are the goals of AI
00:08:21
like an AI researcher so one might be to mimic the properties of human nervous systems but maybe we can do it in a
00:08:27
different way and I think that's what we're seeing now maybe we can even go
00:08:30
beyond the capabilities of human uh intelligence so I think that there are there are a number of different
00:08:36
um commonalities between the two so and it's kind of interesting when you think
00:08:41
historically about where some of the basic algorithms in like machine learning like reinforcement learning
00:08:48
came from they actually had an origin in Psychology and in neuroscience and we'
00:08:51
it so it's kind of been you know this really interesting feedback between I never thought about
00:08:57
that is would like what Pavlov did with his experiments would those be considered reinforcement learning that's
00:09:02
the origins of of reinforcement learning and actually the you know the the the basic um reinforcement learning model
00:09:09
was really written out here by um by a professor in Psychology Bob R scor back in the 19 uh 60s and early 70s so
00:09:16
there's a you know pen has a really I think important part in the history of
00:09:21
the development of of AI but um it's for a long time people thought that there I
00:09:27
think that um we look at human brain or animal brains in general that yes reinforcement learning is important for
00:09:34
learning to navigate and learning what's good and what's bad what to approach
00:09:38
what to avoid but maybe it didn't go much deeper than that and then there were there sort of circuits and this is
00:09:43
certainly true circuits that are have prescribed functions that are sort of built in if you will kind of hardwired
00:09:49
and that AI with its V you know and the other thing is is that there's constraints on neural function okay
00:09:55
we've got a three pound device in our heads 86 billion neurons 100 trillion connections but it's actually pretty
00:10:01
limited honestly and it's limited by energetic constraints which I think we'll talk about in a bit um whereas
00:10:08
AI you know what are the limits well how big is the data warehouse that you you know how many servers can you put
00:10:14
underneath uh the hood of chat GPT and so so I think the thought was that like oh that Brute Force kind of approach you
00:10:22
know in in Ai and machine learning couldn't deliver the kinds of intelligent creative um kind of thinking
00:10:30
that human beings do and in fact that is what we're seeing and now when we look
00:10:35
back at human brains we I think we're starting to rethink that conceptualization which is like uh well
00:10:42
actually our brains have a ton of experience under the hood right so Evolution and then from the time you're
00:10:49
an infant right and you're being bombarded with all this information and that's a lot of time uh for that system
00:10:57
to kind of learn using SAR principles like gradient descent which is really at the root of AI and now when we go back
00:11:04
and we look at what you might call neurons in a you know in an artificial neural network and
00:11:10
real neurons in a neural network that often they have very similar properties which they seem to have arrived at
00:11:16
through processes like gradient descent uh this is Eric bradow professor of marketing and statistics and vice dean
00:11:22
of analytics here at the Wharton School and we're here today in our Ai and Neuroscience podcast uh Edition and I'm
00:11:27
here talking to uh plat who's the James S ree professor of marketing Psychology
00:11:33
and Neuroscience and zap Johnson who's the executive director of The W Neuroscience initiative and also a
00:11:37
senior fellow so Zab let me ask you um how is artificial intelligence being used in your field of Neuroscience today
00:11:47
whether it's from the I always say there's at least two sides to AI one is
00:11:51
the more traditional one which is as in you know images can now be ingested by an engine and data can be now output
00:11:59
from a like for example you could take a voxelized picture of voxelized blood flow from the brain jam it into an AI
00:12:05
engine and out could shoot a big long Vector of stuff that's kind of the more
00:12:09
traditional the other could be more in the generative AI way so any thoughts to our listeners about how AI is being used
00:12:16
in Neuroscience yeah it's being used in so many different ways and uh so I'm a
00:12:21
visual neuroscientist and and I think that some of the beginnings actually of the way that um that cognitive science
00:12:27
vision science visual Neuroscience were coming together with AI and Engineering happened really early so so actually it
00:12:35
was oftentimes through the neuroscientists that were thinking about Vision um how we see or how we can even
00:12:42
train machines to see um that some of the very beginning of these algorithms emerged um and actually like you know I
00:12:49
think deep neural Nets um and and you know CNN's for example like were like they were an outcropping actually from
00:12:58
people in my dis IPL um and so like I I remember the ear one talking is that a cat right you know so you have some
00:13:04
picture an image with a bunch of pixels and then it's putting it into some neuronet some compression engine and
00:13:09
then it's got a you know encoder and a decoder I I agree with you I think Vision was probably one of the earliest
00:13:15
ones that got people excited yeah and I think one of the really interesting things in that dialogue was was that you
00:13:22
could get to the same you could get to the same answer you could actually make machine c um but it turned out to be
00:13:28
fundamentally different from the way that that that uh that the human does right and so I think that some of the
00:13:34
discoveries that are happening now are how can it inform actually how we understand neur neural processing um and
00:13:40
one of the powers I think of AI is that it's a you know it can seek patterns um
00:13:45
and even multiple answers to a single question that seems to you know push on the limit at least of single
00:13:51
investigators or you know even you know even teams um and so I think we're about
00:13:55
to learn much more um I think some of the most innovative work right now is happening where you can see communities
00:14:03
and of of both AI researchers and neuroscientists back in dialogue with one another um some some recurrency um
00:14:11
in that in that conversation and you know I think that I think we're in the very beginnings of of seeing what the
00:14:18
power will be but I think um you know I think to give you some concrete examples
00:14:23
um uh a couple of researchers uh uh yukiyasu kamatani who's at Kyoto University and Frank Tong who's at
00:14:31
Vanderbilt University about 20 years ago a little over 20 years ago did some of the the foundational um work um to look
00:14:39
and see whether you could de go decode um using algorithms um the information that people were seeing um that was that
00:14:47
was sort of the beginning um and that was long before we had generative AI we were just thinking about algorithms and
00:14:52
the power of pattern detection um then you know quickly after that um Jack gallon and Alex Huff um who Alex is now
00:15:02
at the University of Texas Austin um and Jack who is who work you know whose lab
00:15:07
he was in at the time at UC Berkeley were thinking about semantic meaning um and thinking about the kinds of brain
00:15:13
activity that came up with semantic meaning but also um visuals um uh and movie decoding um and like they were
00:15:22
starting to to use algorithms to look at the patterns of brain activity to help decode you know from a researcher
00:15:29
standpoint what people had actually seen um and uh comani then did this really interesting thing where he actually
00:15:35
decoded dreams so like one of the I think one of the interesting aspects of of that work is like you know telling
00:15:43
you something about that maybe people actually have a really hard time reporting right um or it's impossible to
00:15:50
to to to report but another kind of imagination or visualization so Michael let me ask you
00:15:57
a follow-up question to what Zab said so I tend to be and maybe this is why I'm
00:16:01
not a basic scientist well I'm a scientist I'm a basic scientist sort of
00:16:06
um I tend to believe things that are of low-dimensional representation but maybe
00:16:10
the world is really complex so uh Zab mentioned something about pattern recognition how much of like the future
00:16:17
of what we're going to learn in Neuroscience is because we can take this very high dimensional or let's call it
00:16:24
three-dimensional time series data put it into some AI engine and we're going to notice some 86
00:16:32
dimensional interaction that no human could possibly find is the world built that way with 86 dimensional
00:16:40
interactions or is it like no if you understood these I'll make it up I'm
00:16:45
literally making it up and you'll correct all of my vocabulary if I if these neurons are voxelized areas if I
00:16:52
put them into the right bracket then it really is just a three-dimensional four-dimensional thing what are we going
00:16:57
to learn from AI wow that's a deep big question we have six to eight hours here on this
00:17:04
episode you you know I can talk for a long time I know you can so I think I have sort of two answers to that I mean
00:17:10
I think that in on the one hand in the applied sense maybe it doesn't even matter but what it does allow us to do
00:17:16
and there were three striking examples of this kind of following what Zab talked about this year three major um
00:17:23
discoveries Publications whatever you want to talk about was you're taking this very high dimensional data and
00:17:29
you're reducing it and you're turning it into something useful so there was one
00:17:33
study um you know was the culmination of Decades of work but by a group at lazan
00:17:38
that basically took a guy who this gentleman who had a spinal cord uh injury he'd been paralyzed for you know
00:17:44
a couple of decades uh you take the data out of the brain you feed it through an
00:17:48
algorithm machine learning algorithm and now rather than trying to actuate a robotic exoskeleton or something like
00:17:54
that uh you actually pipe it back into the spinal cord beneath the S side of the injury and now the guy can
00:18:00
walk I mean you know breathtaking who knows how it's working what it's really
00:18:04
happening does it tell you how the brain works not necessarily but it's incredibly useful similarly uh for work
00:18:12
um out of Eddie Chang's group at at uh UCSF allowing a woman who's been uh
00:18:17
unable to speak for you know again a long time more than a decade uh due to a stroke to actually have a conversation
00:18:24
real time with her uh with her husband okay so you know speech is generally taken to be the most important aspect of
00:18:31
being a human being those are the thing the parts of the brain you want to avoid
00:18:34
um injuring in any kind of uh surgery and now she can actually have a conversation that is Meaningful and then
00:18:42
another decoding uh one that um kind of building on what Zab talked about uh an fmri study and fmri let's appreciate
00:18:50
it's not a great technology it's slow it's sluggish it relies on blood flow so
00:18:54
not very precise right but um but in that study uh the scientists were able to uh decode what a person was reading
00:19:03
not word for word but the gist which I think is really interesting not from language areas but kind of from all over
00:19:09
the brain that was my question and uh and they could then decode what they were thinking okay now as idiosyncratic to
00:19:16
each individual I couldn't take like my brain decoder and put it on you probably
00:19:21
wouldn't work although we sh share a lot of similarities but um I think those so
00:19:26
those are like ways in which the sort of the data reduction dimensionality what dealing with all that complexity is just
00:19:34
helpful and it's just useful right so but I think that um in other ways some
00:19:40
of the work that we've done and and we have a paper uh in review right now that
00:19:45
records which we recorded data from thousands of neurons in monkeys these monkeys are rather than being engaged in
00:19:53
a task they're actually just doing Monkey stuff with each other so they're
00:19:57
engaged in totally natural behavior they engage in like 27 different behaviors usually in any kind of experiment it's
00:20:03
like one or two different things and the data on the face of it looked kind of um
00:20:10
you know if you look neuron by neuron which is what you would typically do it looked kind of boring and not that like
00:20:15
didn't tell us that much you take all that data together the pattern of data
00:20:19
across the you know and that's thousand some dimensional data and you do something we use map which is one way of
00:20:27
sort of dimensional reduction and uh you pack that into three dimensions and suddenly all of the that population data
00:20:36
clusters into distinct all those 27 different behaviors and not just that but like who you're doing it with who's
00:20:43
next to you what's going on number you can and so and it's you know it's kind of shocking but also
00:20:50
pretty amazing and so you know I think that maybe what that tells us is that even when dealing with the complexity of
00:20:57
the world which is is is Rich and complex that the brain finds very efficient answers right I mean it's had
00:21:04
hundreds of millions of years of opportunity you know to do that and I think that that's what we're seeing so
00:21:11
maybe in the last couple minutes we have could you tell me about the application
00:21:14
since Michael I know you teach a course in brain science for business that may not be the exact title but it's probably
00:21:18
pretty close pretty close and Zab I know you teach a course in visual marketing that one I know is exactly right um
00:21:25
could you guys give me a sense of the big application areas of Neuroscience in business today that you
00:21:34
are seeing like what are the ones that excite you the most well I mean I I this so my course is really a sort of Gateway
00:21:42
course think of it that way it's sort of like Soup To Nuts everything you need to
00:21:45
know but also like what and somewhat idiosyncratic what are all the different application areas where I think you know
00:21:51
Neuroscience either is already having an impact or will have an impact where it's
00:21:54
already having an impact is in marketing brand strategy I mean that is is a sort
00:21:58
of at this point a no-brainer if you are not collecting neuro dat of some sort you're leaving high quality data on the
00:22:06
floor that could you would make better ads you would develop better Brands you'd position them better we
00:22:11
demonstrate that just you know over and over and over so you can turn that crank
00:22:15
and just do a much better job throw away less money um the areas where I think things start to get really interesting
00:22:22
and exciting are places like HR and management where a more precise scientific objective understanding of
00:22:32
people their individual talents traits and motivators right and what it takes to be really good at for example
00:22:38
whatever job that they are aspiring to can help to make that match and identify the training development Etc that can be
00:22:45
done to kind of get you from here to there uh so we know companies waste tons of money tons of time on this right
00:22:52
churn is is huge uh that's frustrating for employees you know they're unhappy
00:22:56
so I think it's an opportunity for a win-win same thing for teams right so what you can do for individuals it's
00:23:03
more complicated than teams but we can absolutely do that too so Zab I know you guys have an upcoming conference I don't
00:23:09
know if my guess is this podcast will not necessarily go up before it but people could obviously there'll be
00:23:15
results from it there'll be video from it there'll probably be summaries of it
00:23:19
all on the Wharton Neuroscience initiative website um what's happening at the upcoming conference it's
00:23:24
happening this exact Friday it is um it's Friday November 3rd um and uh this
00:23:31
year's uh theme is on brain Capital thinking about um all of the cognitive um skills and we think about that as as
00:23:40
you know emotion and creativity um and you know what I think was classically thought of as as
00:23:47
cognition like what's necessary actually to equip people to be productive members
00:23:52
across the entire lifespan um and to think about the ways that we can make Maybe devise uh strategies economic
00:24:01
strategies to to to to Really leverage that sort of like a like a lunar Mission but now uh thinking about cognition um
00:24:09
and and there's actually one segment of the day's programming that's really um
00:24:14
taking a deep dive into this idea that you know that Ai and human interaction is coming fast um and that and that this
00:24:22
is really a a a moment to seize and think about both you know an ethical but also an optimization of what those new
00:24:30
teaming structures are going to look like um how how can we really equip um you know the human agent to think about
00:24:38
ways to to to do and live better um given given this new role of AI uh that's coming uh but we're also you know
00:24:47
we're also diving into other other aspects of of brain development in children and in aging right thinking
00:24:54
about the really the brain Capital across the lifespan and how like even early childhood ensures better cognitive
00:25:03
end points um more productivity across the entire lifespan which will help economic um and productivity and
00:25:11
businesses Thrive and individuals Thrive um and and probably um you know build better protections for for mental
00:25:19
illness and mental health deficiencies well Michael in the last like 30 seconds or so we have if we were all sitting
00:25:25
here 10 years from now what we looking back and saying whether it's the intersection of AI and Neuroscience or
00:25:33
Ai and humans what kind of problems do you think gets solved in the next 10 years that just we were not capable
00:25:39
whether it's as you said because of more data better servers what are the big
00:25:43
Frontiers in your world over the next 10 years yeah well I think that we are going to see hopefully a lot of today's
00:25:51
um you know issues where brains go arai right so like whether that's like depression whether that's uh you know
00:25:58
neurodegenerative disorders um whether that's the sort of Despair that we see
00:26:04
across the population that we're going to make significant advances in that um
00:26:08
a lot of that's going to be techn technology driven uh AI is going to be a huge um uh Force for good I think in
00:26:15
this in terms of helping us to uh come up with more creative ideas right and help us select amongst those ideas and
00:26:21
then really important for translating them into real solutions too well I'm getting older by the minute so I'm
00:26:27
counting on you I'm counting on you both so uh this has been the AI and Neuroscience uh edition of the AI
00:26:33
podcast series here at the Wharton School again I'm Eric bradow professor of marketing and statistics and vice
00:26:38
teen of analytics and I'd like to thank my guests Michael Platt and Zab Johnson
00:26:42
for our episode today thank you thanks Eric