00:00:00
AI has been looking to develop and
00:00:02
enhance a wide range of areas, even how
00:00:04
we deal with certain human emotions like
00:00:07
loneliness. It is something that people
00:00:09
have had to deal with since really time
00:00:11
began. But can chat bots alleviate or
00:00:15
eliminate some of the issues around
00:00:17
loneliness and what might that mean for
00:00:20
things like other medical issues
00:00:23
potentially something like mental
00:00:24
health. Pleasure to be joined here by
00:00:26
Stephano Panton who's a marketing
00:00:28
professor at the Wharton School and
00:00:30
co-director of the Wharton Human AI
00:00:33
research project. Stephano, great to
00:00:35
talk to you again. How are you, sir?
00:00:37
>> I'm good and great to be here. Thanks,
00:00:38
Dan.
00:00:39
>> You know, the thing is we've talked in
00:00:41
so much about AI in the business sense
00:00:43
or the professional sense. We haven't
00:00:45
talked about it a lot from the human
00:00:47
sense and how consumers are thinking
00:00:49
about chat bots. So, what are you seeing
00:00:51
play out right now?
00:00:52
>> Yeah. So we uh in the business
00:00:55
conversations around AI. Lots of the use
00:00:57
case that we talked about are you know
00:00:59
uh professional or um uh for
00:01:02
productivity and work and things like
00:01:03
that. But lots and lots of people all of
00:01:05
us really are using chatbots in our
00:01:07
lives as consumers and we use them for a
00:01:08
variety of different things. We look for
00:01:11
information. We are trying to learn
00:01:13
skills. we uh just play with it in some
00:01:17
way and uh we get a lot of you know
00:01:21
useful things done with it from recipes
00:01:24
to best time stories and lots of other
00:01:27
stuff. But uh one use case that has been
00:01:30
uh uh become quite prominent and that
00:01:33
maybe you know people wouldn't maybe
00:01:35
expect uh to that extent until recently
00:01:38
is the use chatbot for companionship.
00:01:41
The idea that you can have what you can
00:01:43
call maybe a synthetic friend a or
00:01:46
partner a chatbot that uh plays out the
00:01:50
role of being uh a relationship to you
00:01:54
so that you can engage in socio
00:01:56
emotional processes whether those are
00:01:59
you know um friendship based or or um
00:02:03
coaching or mentoring based or or maybe
00:02:06
even romantic and sexual uh in nature. I
00:02:10
would think though for something like
00:02:11
that, consumers would want to and in
00:02:13
many cases need to see it actually
00:02:15
occur, you know, actually be in action
00:02:18
in order to understand kind of the level
00:02:21
the capabilities that that this might
00:02:24
have. It sounds a bit of a crazy idea
00:02:26
that you would basically um you know go
00:02:29
on a chatbot to find a friend, but uh
00:02:32
you just have to spend a bit of time
00:02:33
using these technologies just to
00:02:35
understand how humanlike they really
00:02:37
are. Especially today when the AI
00:02:40
companies are adding a voice layer to
00:02:42
the system so that you can interact not
00:02:44
through text but through voice. is
00:02:45
incredibly humanlike. And in our own
00:02:48
research, we find that people don't
00:02:50
expect uh to um see much benefit for you
00:02:55
know things like loneliness but uh uh
00:02:57
they do in fact find that when they try
00:03:00
it. So to some extent for many consumers
00:03:02
might become more um appealing a use
00:03:05
case as they get used to the technology.
00:03:08
>> So what's going on then in this market?
00:03:10
I mean how is this developing and and
00:03:12
who are some of the main players in this
00:03:14
process right now?
00:03:15
>> Yeah, so this is a burgeoning area. It's
00:03:17
been uh really interesting to see how
00:03:19
this is uh developing. On one hand, you
00:03:21
could say the big AI companies are being
00:03:24
very careful to in a way stay away from
00:03:27
this compishing angle. So if you take
00:03:29
for example uh OpenAI and Chach, they
00:03:32
are making I think an admirable and very
00:03:34
good job at trying to avoid humanizing
00:03:36
the technology. there's no name, you
00:03:39
know, humanlike avatar. Everything is
00:03:42
very tech and explains to you that this
00:03:45
is a chatbot at every turn. And I think
00:03:47
that uh that is besides it to try to
00:03:49
avoid some of the uh liability or risks
00:03:52
that might come with uh this kind of
00:03:54
market. Maybe uh we might see that
00:03:57
changing a little bit uh uh with time.
00:03:59
Certainly we've seen uh say Grock being
00:04:02
much more aggressive in this space than
00:04:04
some of the other big AI companies and
00:04:06
otherwise has been less to smaller
00:04:07
startups companies like Character.ai or
00:04:10
Replica that have millions of uh active
00:04:13
users and um where a lot of people are
00:04:16
you know playing with but also uh
00:04:19
forging relationships with these spots.
00:04:22
>> You were also part of research that
00:04:24
looked into how AI chatbots might impact
00:04:26
something like mental health. Yeah,
00:04:29
we've done a few research projects. This
00:04:31
uh this line of research is led by
00:04:33
Julian Desertus from Harvard Business
00:04:35
School and what we are doing is
00:04:38
basically to uh try to explore the
00:04:40
consequences of chatbots for these
00:04:44
social emotional processes. We have one
00:04:46
paper where we explore the impact on
00:04:48
feelings of loneliness like you hinted
00:04:50
at earlier where basically we are have
00:04:53
participants engage in a 15 minutes
00:04:55
conversation with a bot and we measure
00:04:58
feelings on loneliness before and after
00:04:59
the interaction and we find that they're
00:05:01
quite effective at reducing loneliness
00:05:03
and we find also that those people for
00:05:06
whom the effect is largest are those
00:05:08
people that report higher levels of
00:05:10
baseline loneliness. So perhaps uh you
00:05:13
know that that could be some kind of a
00:05:15
tool for the alleviation of um
00:05:18
loneliness's societal level. We also
00:05:20
have in another project um we have an
00:05:23
audit of AI systems and then we are
00:05:25
looking at actual transcripts of uh
00:05:27
consumer interaction with bots and we
00:05:29
find evidence that a small percentage of
00:05:32
individuals uh interacting with these
00:05:34
bots is likely to be experiencing quite
00:05:36
severe mental health issues which
00:05:38
obviously raise some consumer safety
00:05:41
concerns. This technology is not meant
00:05:44
to be a let's say mental health tool or
00:05:47
an FDA regulated. Well, that element of
00:05:50
of consumer safety becomes an important
00:05:52
component of this conversation, doesn't
00:05:54
it? And for the companies that are
00:05:56
developing these technologies, they have
00:05:58
to be wary of, you know, potential
00:06:01
reputational risk that might hurt their
00:06:04
uh business longer term.
00:06:06
>> Yeah. Not only reputational risk, even
00:06:08
uh uh liability legal uh in a legal
00:06:11
sense. And I think when I was saying
00:06:13
earlier that some of the big eye
00:06:15
companies have been cautious in the
00:06:16
companionship domain is probably coming
00:06:19
partly from this. And so right now there
00:06:22
is around a dozen lawsuits making their
00:06:25
way in the courts. Some of them have
00:06:26
been high-profile cases at uh newspapers
00:06:30
like the New York Times have been
00:06:31
featuring multiple times and they're
00:06:33
directed mostly towards AI companion
00:06:37
companies especially Character.ai AI as
00:06:40
a few but also recently there have been
00:06:43
lawsuits to go towards open AI making it
00:06:45
the point that even though chach is not
00:06:48
meant to be an AI companion people can
00:06:50
craft prompts that create a companion
00:06:54
out of chachet and that the company
00:06:56
hasn't been monitoring that activity
00:06:59
enough. I'm just uh had recently I went
00:07:02
to a conference when I saw a
00:07:03
presentation by one of the key lawyers
00:07:05
bringing uh some of these cases to court
00:07:08
and the list of uh claims being made is
00:07:12
quite uh quite something. I can read it
00:07:14
to you. Product liability, religions,
00:07:17
unlicensed practice of psychotherapy,
00:07:20
harmful death and slaughter and
00:07:22
encouragement of suicide. We'll see
00:07:25
what's happening with these lawsuits. uh
00:07:27
but certainly this is also an element of
00:07:29
pressure on AI companies and
00:07:31
>> but for for the conversation we're
00:07:33
having here though this like a lot of
00:07:36
things around AI still feels like it's
00:07:38
very much in the development process
00:07:40
right now correct
00:07:41
>> yeah I mean these models are improving
00:07:43
very rapidly companies are learning uh
00:07:46
how to deploy these models for uh many
00:07:50
um uh use cases also learning how to
00:07:53
build safety guard rails
00:07:56
And this space is very develop still
00:07:58
very developing very fast. So I think
00:08:00
it's a bit early to tell um what uh you
00:08:04
know this will look like a year or two
00:08:08
from now. But it's quite clear that a um
00:08:11
chatbots can be very effective
00:08:13
companions at least give this sense of
00:08:15
companionship this illusion if you wish
00:08:17
depending a little bit on your point of
00:08:19
view and b that there is a lot of demand
00:08:22
from uh from consumers for companionship
00:08:25
of that sort. So um certainly there's
00:08:27
going to be a lot of market pressures
00:08:29
towards uh growing this comparatorship
00:08:32
area while there are also obvious
00:08:34
liabilities and concerns like we
00:08:36
discussed.
00:08:37
>> Stephano always great to talk with you
00:08:39
and get your insights. Thanks very much.
00:08:41
All the best.
00:08:42
>> Thank you.
00:08:42
>> You got it. Stephano Penton, marketing
00:08:44
professor here at the Wharton School and
00:08:46
also co-director of the human AI
00:08:48
research project.
