I am I'm Nancy greenly associate dean of
the for research of the College of Design

there I want to welcome you to the fifth
distinguished lecture on robots and

jobs Kelly distinguished sure that
is funded by a generous donor

the lecture series is a collaborative
effort of Iran and the College of Design

it would not be been possible without the
support of Dr Clinton Kelly who's with us

today and who has a strong interest in
how our increasingly robot Assayas world

will be affecting the world of work and
already is Dr Clinton Kelly

has have a just highly distinguished
career of leadership and management and

research and advanced technology projects
in both industry and government and

was at the forefront of the U.S. strategic
computing initiatives I'll just mention

a few highlights for you he served as
director of the U.S. strategic computing

program the national fifth generation High
Performance Computing Research Initiative

at the Defense Advanced
Research Projects Agency or

as we call it DARPA he also directed
DARPA engineering applications office

overseeing research programs that included
robotics an autonomous systems and

especially one feature he's very proud
of our accomplishment there was starting

the autonomous ground because program
in one thousand nine hundred four

before that he directed the U.S.

Department of Defense to study on
Japanese manufacturing technology.

He.

And he also served on the committee for
the national carrier science or

some unmanned ground period goals.

He serves and he has served on the board
of no less advisory board no less than

seven universities so he has had a bear in
this is just highlights of his career but

it's a very long and distinguished one so

please join me in thanking Dr Kelly for
making.

Thank you.

It looks all right.

With.

My.

Group that I go way back many
years ago she was behind

me I was fortunate
enough to have her as my

partner said she's actually personally
shaped a lot of mathematics for a year and

I think that in addition to her
academic or she hoped founder and

chief experience officer to get us back
to the Jeep which is what should become.

The TIME magazine recently selected.

Seventeen driving out.

Of her work so simply access right here.

Actually she was doing a try well before I
was to do it before anybody knew what was

going to be able to write
today to research the design

technical innovation investigation of
the impact of social robots and human

behavior attitudes in our past and what
that translates to a lot of research but

let me get out from research into the wild
wanted to know and really looking at how.

Well being and

help in various human populations is
really trying to change the world.

So because it recognizes.

For all of this work.

The United Nations is the speaker
on education for everyone so

it is very fortunate for
us to have her here are like thank.

YOU THANK YOU Well it's a it's
a pleasure to be here it's.

De.

I wanted to talk both about
the research side of my work but

also the commercialization and I have
to say it's been somewhat I have to say

a life's journey of seeing Star Wars
as a kid in these droids and

it wasn't just about how cool they
were in the things they could do but

that the experience and the relationships
that people had with them was part of

why the world fell in love with these
that idea of a robot back then so

much of my work is about trying
to make that actually happen so

I wanted to touch both a lot of
the science and the research of what

I think is actually differentiated
about the social robot technology and

why I think it actually does have the
potential to shape the world around us but

then also the path to getting it out in
the world that's a whole other adventure

so you're going to see both sides of that
in this talk I actually I did bring.

I don't know if I have time
to demo during the talk but

during the reception I'm going to set
him up with a couple of the fun toys and

that way you guys can play around and
interact with him.

So is in the House he's
just not out of the box.

So you know these days this is really a
big question that I think about every day

both on the research scientific side as
well as on the commercialization side and

just to acknowledge that we all
know the AI is pervading I mean so

many aspects of our society but
there's been a real shift you know I you

know just recent years of who interacts
with AI and what context right so

with devices like Google home young
children seniors I mean you name

it people can talk to these intelligent
devices now when I get talks they

say today's children are not just digital
natives they're AI natives they're growing

up in a world where they've always been
able to talk to intelligent machines and

we actually know very little about that
and now of course I got to show this kind

of promo video of the world is
continuing to change and evolve.

With new technology is coming fast.

So spectrum is getting faster.

There's going to be billions of
connected devices in the next few years

somebody is going to power
the stuff that's why spectrum

is accelerating internet speeds
across their fiber powered network.

To keep everything.

That means everything connected are you
ready for what's next spectrum so

this is in our lab just to say that you
know the social robots now are also coming

so as we think about digital systems that
we can talk to to desktop social robots

to robots that but contentiously roll
around your house and manipulation.

Robotics has been a field where we
keep feeling like the time has come or

it's going to become ubiquitous technology
I actually think now is the time we're

actually getting a little with that
could that could really happen so

these technologies are coming
into the marketplace but

if we look at the research side of
the House actually very little is known

about how these technologies are going to
impact us as part of our daily lives right

it's one thing to bring people into the
lab and do a randomized controlled trial

it's another thing to put them in
a specific work context where professional

has a stake in using a tool to make better
decisions or to do a specific task it's

completely the other thing when it's just
part of the fabric of our daily life and

our families and so this is just a slide
this was a review article written on

the number of longtitude in all human
robot interaction studies Don And

the punch line is very few have
been done is because this is very

hard research to do it's hard to collect
the data it's hard to run these kinds

of longitudinal studies part of that
because there aren't robust capable social

robot platforms out on the market which
is a big reason why I wanted to do as

a company but a lot of it also involves
a very little study of long term

interaction with innovative algorithms I
mean a lot of these studies have been just

taking the product as it is like a room
bunch of studying in the home maybe of a.

Period of months or even a year but it's
just the robot as it's not pushing the on

below of the capabilities of these
technologies so we're at this point now

we're able to live with these technologies
are coming into our homes and

we really need to push the field.

To these directions right
we need to build for

increasing levels of autonomy we need to
deploy them in real world environments

where that's the home or hospitals or
schools we really need to study them

in real life context with
real life stakeholders and

we need to look at over longer term
studies and impact and evaluations this is

really where the field needs to go in at
the very very very beginnings of this

part of this is going to be enabling
the community with the right tools to

do this kind of work again it's extremely
difficult to do this work right now and

then there's a whole other theme I'm going
to talk about in this presentation which

is again there are populations now
interacting with these technologies that

could have important impacts on how they
develop how they you know with ageing

population how the Asian place all these
kinds of impacts on the fabric of society

I think we're all getting a sense of the
plus side in the downside of social media

you know they're been launching shouldn't
all studies now of empathy measured in

college graduate students over
thirty year showing a decline

empathy since the year two
thousand technology media a lot

of things are being
attributed to this decline so

we know that how we interact of these
technologies can impact our behaviors and

how we interact with one another this is
a particular demographic that we know

very little about and you'll see a lot of
my research is really trying to understand

this because I think that the profound
societal need just to highlight that

you know there are companies who
are trying to create technologies for

a wide range of demographics you know
the less mainstream early adopter but

to everyone and
I don't know how many of you followed this

story a must tell who wanted to
create an Alexa like device targeting

you know parents of very young children so
skills related to ordering diapers and

things like that but also being able
to have children talk to the device and

the punch line is they went very very far
down the pipe in getting ready to law.

Launched this technology and

kind of in the eleventh hour they
decided to pull it from the market and

not release it at all and this was based
on you know parent Abbott ski groups

concerns about lawmakers you know a lot
of concern about we don't understand

the implications of this technology
especially our young children yet and

I think that the punch line for me here is
really do we really know enough to make

an informed decision and
as I talk to Noora scientists and

development psychologists The bottom line
is these policies are being made with

a dearth of science we're not making these
decisions and policies from the vantage

point of being well informed so
this might have been the right decision or

not the problem is we just don't know
enough to actually make these these these

these calls right so that's just the state
of the world we live in today and

this is again this is shaping a lot of
the research that I'm doing right now

the other side of this is really just
appreciating a has become obviously

this pervasive technology it's pervading
our work practices are institutions and

our home life so much of it is around
the fish and sea and productivity and

decision making which is fantastic of
course from the commercial kind of

business side of this but I also worry and
think a lot about this question for

me I don't think it will truly achieve its
full potential if it can't be designed to

also help us Flora we as human beings
are experience who we aspire to be can

I help us do these things as well so
the task isn't

building something where something
externally rolled in some sense the object

to be impacted is
the person themselves and

a personal goal they're trying to
achieve to live a better life either for

themselves or for their family and
as we look at you know frameworks for

Flora we see that wellbeing is
certainly at the Intersect of not just

cognitive aspects but deeply social
emotional and relational So if you want to

create technologies that can really help
support us in these critical areas as well

we have to design technology in
a eyes that address a much more.

Human experience and so much of the work
I'm going to talk about today is

exactly that kind of digging deep not only
into these cognitive aspects that kind of

dominates how we think about AI today but

when you talk about AI's that need
to dovetail with human behavior and

in our how we think and how we learn how
we experience life with each other in

these technologies you really need to come
up with a much more holistic framing and

investigation of the impact and
dovetailing of all of these aspects and

it has a profound impact on the nature
of algorithms you design right so

it's not just about the human factors the
human factors deeply inform the design of

the algorithms themselves in the behaviors
of the these technologies themselves

which again impact human behavior so

it's this coupled system that we really
need to understand so I'm going to

highlight kind of a few themes from
the research side I'm not going

to dive deep into the algorithmic science
what I really want to highlight for

you is the impact on people because I feel
this is a side of work that people don't

learn about enough or don't hear
about enough and it's again profound

when you think about the future of our
life with these kinds of technologies so

when I first mentioned
as we first started this

field of human robot interaction it's
Holmes twenty years ago believe it or not.

We started to try to publish work on
social robots robots actually interacting

with people and we would try to apply to
traditional human computer interaction

conferences and the papers would get
rejected outright because it was done on

a robot they wouldn't even pay attention
kind of like the science that was done

it was about why would you ever do this
on a robot robots are more expensive

you know they take longer to get
things out in the war all of you

know they're brutal like we just put it on
a digital age and a graphical agent just

put it out on the Internet like that's
all you ever need to do right so the few.

You know we believe there's actually
something interesting about robots and

robots can do things that
digital agents can't and so

we had to basically start
justifying our existence

by doing systematic randomized controlled
trials comparing the impact of a.

Simple co-president social robot
to a graphical agent to a just

somebody talking agent you name across
all these different conditions.

And so one of the things that we've
discovered is there's a different quality

of the emotional lift I called
emotional lift from this

physical social co-presidents and
I want to highlight a particular use case.

That we've been working with
Boston Children's Hospital

this is a pediatric context that we
just finished a clinical trial and

we've been working with child life
specialists so these are the professionals

in a hospital whose job it is to do
with kind of the emotional wellbeing and

coping of children when there's a
stressful situation of being in a hospital

one important thing that
you'll see again and

again in this work is that right now
the recommended ratio of a child life

specialist to patients is about one child
left to roughly like fifteen pediatric

patients the reality reality is it's like
one child like professional to forty

pediatric patients so what does that mean
there are many many children whose needs

are not being served to the extent that
everyone would like to see that done so

this opportunity to create a technology
that can engage children and

their kind of emotional needs becomes very
interesting to specialise as a way of

having continuous monitoring of
the child's emotional experience

Dr Peter Weinstock talks about emotion
is the fourth vital sign because simply

you know emotion is the one thing we
can't continuously measure we have all

these issue must measure all these
other you know biophysical heart rate

breathing rate you name it but emotion is
the thing that pervades everything about

the hospital what it does from the quality
of care and education of the patient to

the compliance to the recovery rates to
the ability of like the longer it takes

a child to give you their arm in order to
draw blood that's like that's time that's

dollars right even the efficiency of
the hospital to customer satisfaction

of parents wanted to bring the child back
to the hospital like it's pervasive and

yet we have no way of
continuously monitoring and

child like Specialist have no way of
using that information to prioritize.

And when to engage patients so having the
social robot that's kind of part of their

team of extending the quality of
their intervention is very very very

interesting them so I'm going to show
a video clip just to put this in context.

Which I love because Dr Dude or Logan
is actually talking about the case for

social robots you're hearing it from
the standpoint of a practitioner and

the opportunities that they
see a life it's got a chronic.

Can you hear that.

Can you hear that it's.

Really.

One offered It's like one.

To see OK where they are in what's
otherwise a really stressful experience

but I think there's a way
of connecting with tense

that's different from
what grownups can offer.

Credible imagine and
they can really suspend disbelief and

there can be a relationship that
develops the teen huggable and addition.

To.

Eat.

Was.

I.

Was.

Trying.

To learn.

That right even.

When it's like us.

OK So you see the case of
a child specialist was actually

to Child Left specialists in this study
often when you start a new project or

use this was the Wizard of Oz technology
right this technique where we

allow one child life specialist to be the
huggable certainly for a clinical trial

the hospitals are very concerned about the
behavior of this intervention and then you

have a child life specialist in the room
so in the study the child life specialists

used one of these three technologies
to the best of her ability to deliver

her service to the child a second child
specialist in the case of the robot or

a virtual agent on a tablet basically
was the huggable bear we capture all of

the data all the interaction data and this
becomes an important datasets for us to

develop autonomy around these capabilities
so this is again a early exploration

the first pilot we offer compared it to a
physical clash which to you know the trial

of specialist is also animating right so
each of these cases the intervention

is a socially interactive experience
whether it's one of these three and so

the question is if you're a child life
specialist you have the patient in

the room and you have even family members
in the room what is the impact of

the difference of embodiment in all three
given that they're being brought to life

in very similar ways so we looked at
a number of measures So again we recorded

everything about the interaction recorder
what the robot said or the social agent

with the child specialist and with the
child that even with the family said we

wanted to facial expressions body language
physical touch all of these different

modalities to try to understand
is there something different or

is an opportunity here that distinguishes
this physicality of intervention and

we looked at children from the age of four
to ten across a wide range of conditions

from oncology to surgical to you name it
because again we're trying to understand

from this the initial vantage point where
is the sweet spot of opportunity for

intervention the punch line here is
often when you talk about engagement

of the child patient the social
emotional engagement is really important

you want them to have positive
emotional experiences in the hospital

you want them to be so.

Gauge with the child life specialists you
want because that can lead to greater

parents you want them to be able to
engage in shared attention behaviors you

can you want them to feel very engaged
with a child special when their room so

the question is which of these
interventions provides the most impact so

that one of the plots here I can't I
guess I don't want to but this one

looks at valence of the utterance of the
child's utterances over time you see that

overall with the interactive agent being
the virtual agent or the robot it's higher

overall than the plush but in the case of
the physical robot you see the valence

the positive valence going up over time
the child's utterances where you actually

see it start to decrease with the virtual
agent when you look at shared attention so

social interpersonal engagement you don't
want children just staring at the robot

right you want them engaging in shared
attention with the child specialists and

the agent right you see way more shared
attention behaviors with the physical

robot than the virtual agent than
the plus when you look at just the number

of utterances into every question you see
that children just tend to be more verbal

when the physical robot's in
them they just say more and

you can control for the number of
other instances of the huggable

as well as of the child life specialists
children are just more verbal shows and

look at a level of engagement when
you look at the cooperative so

we looked at used Watson sentiment
to characterize the cooperative

of the utterances again more cooperate or
says for the robot and

for the digital agent from the plush but
if you do regression analysis

you still see the tend to be more
cooperative and it's with the robot and

then when you look at touch and you're
going to see in these videos especially

with people of all ages honestly social
touch becomes a kind of an important

aspect of interaction with these physical
technology is just done in this very if

fairly it of almost sometimes affection
like companion animal way we see

in the case of the robot versus the Avatar
versus all aspects of touch all the touch

with the robot as a social relational
touch versus a topping the tablet or

even just throwing the plush
across the room right so

very very interesting so
this is just an example

as a highlight if you look at
between the year of two thousand.

One to twenty sixteen and I mentioned
all of these randomized control trials

comparing physical robots to a virtual
agent of the same robot like you saw in

the study or a physical robot to a video
of a robot that you know was real but

just doesn't happen to be in the physical
room with you to disembark voices exciter

etc There's been many many many studies
done over these last many many years

interesting of what you're finding is
that the physical presence of the robot

actually matters in the depth of
engagement so when you look at

the studies of the significance of this
is sickle significance of the results for

the robot higher trust higher affection
enjoyment higher empathy hiring gauge

meant higher persuasiveness preference
performance you name it so again there's

something about the physicality of CO
presence of these relational technologies

that shifting Gage kids people of all ages
honestly differently in an impactful way

so that's a funding that's actually quite
robust around the field and it's important

for us to keep in mind because these
are not these are both performance outcome

measures as well as interpersonal
working alliance measures when we

talk about creating an age when that's
going to collaborate with someone else and

other not just be used as a tool these
social perception engagement models

are really important for

the quality of nature of teamwork that you
can try to create with such a technology.

The second thing I want to talk about is
this notion of Allied engagement with this

that the neighbor by this interpersonal
user interface it's not just what

you say but the non-verbal communication
that's happening between people so

we know in just human
communication people will say so

up to sixty to eighty percent of what we
communicate we communicate nonverbally and

it's not just the semantics of what we
say it's communicating things like am

I an ally of yours can you trust me do
we feel a feeling did you know it's

all of these social judgments again
all of these are being communicated

through our bodies through a dynamic
regulated interaction and a lot of this is

done subconsciously for people so when
you design these technologies what we're

finding is that people are reading
these non-verbal cues and.

The agent in these ways whether you intend
them to or not is like this is the way our

brains work right as a social species
those social neural specialist structures

of our brains are pulling out that
information right so it's really important

to think about how you design to support
that and that's a really important to

think about in designing agents that are
much more seen as a partner a collaborator

versus a tool so let's talk about a lot
of work that we've been focusing on

particularly in the demographic of early
childhood Now why are we looking at early

childhood there's a lot of reasons why
this is really interesting the first

issue is that this is a huge issue in
the United States right the quality

preschool education we know it's really
critical to start children off in

kindergarten ready to learn there's been
very very many new studies showing that if

children don't even start kindergarten
ready to learn they don't catch up because

their peers who start really learn to stay
ahead we see that these children are more

risk of dropping out of school
of teenage pregnancy of drugs of

all of these things so this is like it
gets down to a social justice question

right how can you create
a technology that can support

young learners in this case often in the
home because a lot of kids are going to

preschool they're going to daycares or
they're staying at home and things like

that to really help all children ready to
learn there's also been a lot of economic

argument saying that of all ages that
gives the biggest bang for your buck in

intervention it's zero to five so
huge societal need the other thing is

children from zero to
five learn differently

then kids when they're in high school or
college right they're not going to sit

behind a computer they're not going to
read a book they learn through interaction

they learn through play they learn from
play in the physical world where these

social emotional physical qualities really
matter there's been a lot of science

showing that young children tend to
learn from friendly support of others so

how do you create a social robot that can
engage children as a peer like companion

that is portrayed as
an attentive friendly in gauging

other is part of
the challenge before us right

we also know that again kids can learn a
lot of things from social robots so when.

Not only about again with
the competence of say vocabulary or

angle or language development early.

There's also attitudes towards learning
like having a growth mindset or

curiosity there's also a sense of
does a robot build that affiliation

that children feeling gauge to sustain
their engagement over repeated encounters

to keep that engagement to keep propelling
the learning Forth so there's a lot of

human factors things you need to design
these algorithms to optimize and

support for a demographic that
is very understudied right so

we hear a lot about deep learning we hear
a lot about all the data that exists and

that's also there's almost no data for
this right so there's a lot of opportunity

here to capture data so the question here
is how do children learn and engage from

one another as peers we know that children
learn a lot from their peers right but

it's highly multiple we're finding that
similar to the case in speech recognition

there was a sumption that you know as
we've commercialized speech technologies

that those models for adults also work
well for children they don't and we're

starting to find from the science that
children actually communicate nonverbally

different than adults so even the mall
of adults for models don't necessarily

pertain to children of this age right so
again just highlighting that

as we expand the design of these
systems to different demographics

don't assume it's a one size fits all as
you move demographic demographic you may

find out that those models need to be
customized need to be different right so

this is just a video that kind of
shows a soup to nuts process so

they can have two children will come
when you know you're tired and a.

School that I.

Can see all they're going.

To.

Apply machine learning Perceptor
classifiers to annotate and expressions on

a conference and then try to model that
right so we have occasion to kill them and

we want to have and
they want to do a lot of that and want.

That we can put that little in there was
kind of it and you want to thank you very

well but before you get to how many
I want it down the line I guess so

in this case it's actually really
fascinating we've published in the works

where these social robots become
a really interesting psychological tool

to under something understand something
about human behavior so in this case for

instance you can't tell another child
the pure light companion child OK What

we want to do is control your behavior in
this very specific way with this child but

then do these behaviors with this other
child children we just can't do that but

with the robot you can control those
signals and really get a lot more nuance

and insight into what the mechanisms
are that propelling Gage meant that propel

learning outcomes so again these social
robots become Then body model for

our computational models and
trying to understand

how does she mean behavior work
what influences human behavior so.

I want to highlight one study
looking at expressiveness So

this again is another example where we
want to ask these questions How do you

optimizing gauge when a behavior robot
to propel the learning of a child what

facial expressions what body
animations what relational models

in this case we look at just the question
that expressiveness a voice right so again

you can't tell a child speak in a neutral
tone in this situation to your friend and

now you emotionally president tone of
voice you have to do this through a system

so we do a lot of work in gauging
the Boston public schools we specifically

go to schools where we have high
English language learner populations

it turns out that one of the fastest
growing demographics in our public schools

are children coming into kindergarten not
speaking English as a native language and

that's an issue because they're not
necessarily starting you know at the level

of their native English speaking peers so
we're looking at these social robots as

an intervention trying to show that
we can help boost the oral language

development vocabulary of English language
learners to help propel them again to

get them ready to learn to read so

in this situation we have a robot
in the storytelling context.

Where the robot tells a story it's shows
pictures on additional tablet to a child

the robot that the puppet here pretends it
falls asleep and then wakes up at the end

of the story we invite the child to
retell the story to the puppet and

it's in that retail that we're analyzing
children's or language production and

there's a target very coded in
the robots telling the story one and

we want to see how much of that talk of a
cavalry is captured in the child's We tell

it would you propose testing all of
a capital or so we're looking at for

Cabul or
looking at oral language development so

I'm going to show you a quick
video you're going to think.

You can see to put it in.

The water not a good stressor probably

the leading into a lot
more social touching and.

Through the future.

All right yes so
here is the punch line OK so.

What we're finding as a robust phenomenon
again and again is because these children

perceive and engage these social
robots as peer like companions

you see a lot of social processes of how
we learn from each other naturally begin

coded in children's behavior one of
the ones we see again and again is.

Social modeling so children tend
to socially model what robots

do this is important social children tend
to socially model what these robots too so

this is an interesting thing because if
you can have the robot model slightly more

sophisticated or a language the hope
is that the child is going to emulate

the robot it's going to pull them along
the developmental trajectory right.

In this case what we're seeing
is that children when they have

the robot it's more expressive they
tend to model the robot's language more

than when it's a neutral expression so
the more socially

engaged with the work's positive you see
more of the social modeling happen so

not only do you see more social modeling
you see longer stories in the retail so

they're more deeply engaged they tend to
retain more of the story and immediately

tell they tend to court and code more of
the target Cavaleri in their retail and

then even in a post test we see that when
they're engaged and do the social modeling

more and this continues there's also a lot
of dialogic storytelling so the robots

asking questions during the stories well
the children who don't my logic answering

also tend to retain more of a copy Larry
but the punch or kick line here is that

after a month delayed post test where
we go back into the school later and

we have children retell the same story we
see a significant difference in children's

stories are longer in the case
of the expressive robot so

there's more retention
more staying power and

they still capture a lot of the phrasing
that the robot side so again we look at

a delayed post as you see a marked
difference in children's behavior.

Another psychological
finding That's interesting so

we said OK so we can look at this
in the case of like vocabulary or

a language what about attitudes right
if children are socially modeling this

attitude how far can you push that so we
started looking at what about mindset and

curiosity if a child plays
a challenging game with a robot

that models of their pure procuress
behaviors or a growth mindset do you

start to see children themselves self
identify with having a growth mindset and

you actually see that affect
their behavior do they start to.

More persevere and more great on
challenging tasks when they play with

a growth mindset robot versus one that say
has a neutral mindset this is particularly

interesting because we know from the
psychological literature these attitudes

force learning are socially shaped there's
a lot of conversation about how teachers

and parents need to praise their children
you don't praise them because they're

smart you praise their effort because
these attitudes are shaped socially it's

never been shown to be shaped necessarily
by technological intervention right so

again now we're looking at the case of
a social robot that can play a game

challenge instead of in this case ten
puzzle solving games with a child and

we actually develop an architecture by
which the robot is actually really solving

puzzles and based on its own progress in
solving the puzzles it's uttering these.

Problems to the child in the case of the
neutral mind so that the robot was just

factual comments this puzzle is taking
more moves than the last one just facts

will compensate the neutral one but in the
growth mindset the robot may say things

like this this puzzle is harder but that's
good because that's how I learn right so

we can praising effort growth mindset or
into utterances.

The robot is using the same cognitive
model that when the child is solving

the ground tracking
the child's progress so

the robot makes similar comments
as the child solves it too so

we're going to take turns in solving these
these these ten puzzles there's a series

of five puzzles the fourth puzzle
is a challenge to ask where

they are it's an unsolvable puzzle so the
child is essentially forced to fail and

then the following is a perseverance task
or it's a harder puzzle and we want to see

how long they stick with it right cuddle
and stick with it so I'm going to play

just a quick video so you can get a sense
of the kind of utterances that are made.

Right there is also the case that you know
there's three puzzles to pick from how

challenging does the robot
choose its puzzles when is it

a growth mindset percentage and similarly
how hard are the puzzles of the chart.

This case the robot solving the puzzle
he's making comments again based on

the solver.

Measuring its own progress.

The.

Robot fails.

But says hey it's OK This is how we learn.

And you see kids are really engaged
right now it's a child's turned child.

So this is always the growth
mindset condition down.

So we're looking out how heart of a how
hard a child challenge themselves in

the puzzle if they choose and
then we have the failure condition.

Try.

And then perseverance sixteenths right so

we see again we see this
relational dynamic again

touch a lot of social affiliate of touch
so the punch line here is we prepare.

Pretty impose test children's attitudes
on standardized measures of mindset

we see that for the kids who interact
with the growth mindset robot in the post

test they self identify more with
having a growth mindset themselves but

the most interesting we thing we see is
on that challenge task children actually

persevere longer they try harder they
try more things frequently when they

interact with a growth mindset robot
than the neutral mindset robot So

again just a provocative finding
shows again how these social agents

can influence our behavior through these
social mechanisms right the last thing I

want to touch on of course is
the opportunity around personalization so

if we're living with these AI's over
time there's an opportunity to really

get to know people to be able to develop
personalized models to them to really

optimize the benefit or the outcome
that these systems can achieve for

people over time and as the system start
to pervade our schools and homes and so

forth that this is a really interesting

research question an opportunity to
potentially become a game changer right so

I want to touch on two projects one
is looking at again emotion right so

the first question is modeling how does
the system how well can a system actually

predict the skill level or mastery of
a child in a task in this case this is

a word reading task where we're trying to
learn things such as if a child chooses

a word in a string of words how often do
they get the first letter write the last

letter write the number of syllables
right in the full word correct right so

we're trying to model these skills
obviously the more a robot can model these

things accurately the better it can

personalize over time so this is
the classic challenge for any personalized

learning system the question we had is
if you have a social robot first of

all if you're looking at emotion as an
input because that actually help you have

more predictive better predictive models
of what children mastery is and also.

Overtly are children more expressive
when interacting with a social robot

because if they're more expressive it
helps you develop those computational

models to be more active in the prediction
to do better personalization so

this is the question right so

we look at the effect of computing
inputs this is facial expression.

Not only having the child interact with
this educational game the social robot

we're constantly measuring their
their facial expressions and

we're looking at incorporating that
into basing knowledge tracing right so

this is a well known algorithm in
intelligent tutoring systems where

competence of a model is model is an
H.M.O. The output observables are whether

you get a question right or wrong and the
hidden state is basically mastery right so

the question we're asking is if you know
that's a pretty heavy lift on inferring

that from a got the question right or
wrong sort of paradigm what if you

supplement it with these other behavioral
cues so confusion smiling things like that

can that help you gain a greater
prediction because you're able to ground

out things like if a child got a question
wrong because they look disengaged maybe

their mind was watering versus a child
got a question wrong and highly engaged

which might indicate they actually have
a conceptual disconnect right so here's

the interaction we did we compared both
a physical presence social robot versus

we literally put a box of the robots and
I can call the condition of talking boxes.

But the game is that so there's a story
maker game where children can drag these

digital stick puppets along the screen
a sentence is generated in this case

the purple dragon roars whatever
the robot disposition does appear who

is learning it wants to learn how to
read so OK sure the rebel asked a child

can you show me the word dragon in which
case the child can touch the word and

that's like the input did
they get it right or wrong so

we do it in both of these situations
where the robot is physically present

versus where it's basically just a talking
box and we measure the affectation of

computing the facial expressions between
these two conditions and what in fact we

do find is when the social robot is there
children are more expressive they're more.

Expressive in their facial expressions so
that alone is interesting right

the second question then is can you
incorporate this information we're

looking at the facial expressions
of children five seconds before and

after the making a selection and

those that actually improve the predictive
power of what we call now the effect of B.

Katie algorithm and the bottom line
is yes we can actually show that this

emotion this input now can actually take
even more accurate predictions of child's

competence so this is some of the first
work to show emotion as an input for

building more predictive models
is actually really interesting.

The last thing I want to touch on then
is let's say you can model competence of

a child what about personalization So
again looking at these long term

encounters with three much interaction in
the real world this is you know it's hard

research to do it's hard to do it in
real context of schools especially with

populations where you can only see them
once a week but we've been doing this

research over a number of years now and
this is one of this is the last study

we've done that's really provocative so
three months an interaction sixty six

bilingual yourself students where we're
actually trying to learn reinforcement

learning to learn to personalize policy
for storytelling interactions I'm going to

show you a video on the explain to you
what the algorithms doing a little bit.

Actually she says she can't eat for
centuries for buying a cat and saying.

There's a story tablet or

thereabout computing technology the robot
just hack to build relationships I don't.

Care here I don't allow any.

Starry.

La.

Toya.

So this dialogue storytelling is
actually really critical that's been

scientifically shown again and
again to be super effective for

kids so now the robot can
demonstrate can ask questions but

that's characteristic of the interaction
next the child then tells the story to

the robot we record their story telling we
analyze it using mechanical turk we look

at again the sentence structure we have
a syntactic tool called it's in that we've

digitized to run over all of the stories
so we know the oral language complexity of

the stories in the robot's repertoire
about one hundred stories we also have

a read on the child syntactic complexity
and then of course recording for

Cabaye Larry in that as well so we're
looking at a lot of different signals.

We apply reinforcement learning right so

deep learning is amazing in a lot
of context when you're wanting to

develop personalized models you have
to learn from interaction right so

the robot is learning to personalize
what is the optimal story for

me to tell this child right now based on
the estimates of the child competence and

knowledge as well as their engagement
right so we're looking engagement

as well as performance this part over here
is basically just the confusion matrix

showing that the similarity in policies
across all the children the personalized

condition definitely there's differences
going on so if there's something

happening under the hood that these
policies are R R R R valving differently

using reinforcement learning but
of course the proof is in the putting So

the question is these may be different
policies but are they actually moving

the needle for these children so
this is comparing the personalized.

Versus one that follows
the true fixed curriculum and

we're seeing that the first time a child
demonstrates a new oral language.

Skill happens faster with the personalized
robot than with the non personalized

robot and we look at pre post around the
cabbie Larry learning we see that children

learning more McAbee Larry with the person
system the non-person system so

again this work it's really the first of
its kind of working with a population

this young in this real world you
know where again AI is really rubbing

against these children in a real real life
situation and it's important to be able to

show that you know not only these systems
personalize it it's actually making a real

impact on children's behavior and what
they're learning so that's just again kind

of a highlighting of a number of projects
showing how they are you design needs to

be deeply informed by the cognitive social
emotional engagement of the population and

how that shapes the behavior they are in
order to really move the needle for

the stakeholder and for us all of this
work is fundamentally about making

a difference in that person's life it's
not just what the robot does it's how that

impacts the person's how it benefits
the person so starting to then kind of

give a highlight there so we live in
a time right now we're AI in the home is

very transactional in a lot of ways it's
like command control voice interface

to an information chaos going to thing
that can characterize a lot of what we're

seeing today what I'm talking about
now is kind of a whole new trajectory

of call it relational AI where these
systems are deeply engaging the social and

emotional intelligence capacities of human
beings to engage and move the needle for

those people further and faster than if it
was just a transactional interaction so

I like to say it's more high touch me
high tech what we're seeing again and

again and I'm going to show some movies
again with different populations but

certainly you've seen
that with the children

the relationship with these robots it's
a different kind of relationship right

there's a lot of discussion around fear of
replacing people with intelligent machines

I think in this work you can see that
there are aspects of the relationship

that are like a motivating ally having
a support of other appear like companion.

These robots can be connected to the
Internet and digital back and just like

anything else so they're certainly
aspects of a cloud connected tool but

then importantly there's this whole other
dimension of this affective companion and

when you're talking about that means like
learning or hall where you don't want to

appear to be a bad student or
a bad patient where having this non judge

mental support of other is actually really
important for the nature of engagement and

relationship that can be really beneficial
to people so we're seeing this again and

again so when I think about where this
is all going right we've seen this

trajectory of intelligent smart devices
along this useful tool paradigm right so

from the internet to social media Syrian
smartphones wearables you name it

now Alexa and many of these cases the
technology is viewed as a useful tool and

it's really been I'm out the dog
democratization of access to information

and networks which has been amazing and
it's changed the world right but

what I'm talking about is there a lot
of hugely important societal challenges

where you need social emotional
engagement to get the best outcomes so

we've talked a lot about education we've
touched on health aging in place is

another one veterans of P.T.S.D.
I mean the list goes on and on and on and

on right so here it's really the power
of emotional social connection

that allows people to more fully
invest in the interaction and

bring more of our human ways of knowing
and understanding and engaging and

not surprisingly often people tend to
have better outcomes when they do that so

social robotics for me is this capability
of this deeper humanistic engagement

that dovetails with our social
emotional intelligence in

these new ways to unlock human potential
in new ways that can allow people to

basically be more successful So for me if
there's a huge opportunity around the use

of these kinds of technologies as more
of a helpful companion as a partner to

really democratize access to the scalable
affordable personalized services and

again it's never about replacing.

The nurse or the teacher it's about
extending that to the home and

whether I have engaged communities
around even Pediatrics

they say we can give them the best quality
of care when the child in the hospital but

we are now accountable for
the quality of care when they go home and

we have no way of reaching them in the
aging in place communities we hear them

say there is no way we can build enough
the solidities to address the drug or

not demand that's going to grow we
have to be able to gauge people to

age in place in their homes right in the
education situation how can you transfer

the learning that happens in the school
to the home as a practice partner and

especially when you talk about lower
social economic situations in the home

how can these social robots also help
coach parents to help their children

get ready to learn in the open A number of
programs from that even just two visits

by a practitioner to coach parents on how
to read to children how to speak to their

children can have important outcome so
again affordable scalable intervention in

the home I think is a big game changing
benefits could potentially happen so

that's the big big big opportunity OK so
now I'm going to start talking about

the other side so when you talk about from
research getting out into the world of

course we know this convergence is
happening so we have ecosystems mobile

ecosystems we have intelligent devices we
have robots it's all converging right and

of course you know you want to believe
that of course the arrows are going to

the next big thing right you know
we've had the Web We have mobile

have social you know this trajectory
that of humanised engagement for

all of these institutions or
service or content providers for

what they really need to solve is not just
characterizing human behavior but engaging

actionable changes in human behavior
whether it's learning or are adhering

to a health protocol you name it there's a
lot of interest in the engagement side of

the space so I'm going to take you
back now to two thousand and twelve.

When I first founded the company
we're going to kind of talk about

a walk through memory
lane of what has taken.

To kind of get this technology
out into the world so

this is before this is before a lot
of these other technologies but

this was kind of the story
that I started to tell so

we have the situation with robots right if
you look at these dimensions right we have

single purpose robots that do one thing
well but they're not really emotionally

gaging necessarily you have the other side
of like entertainment robots that aren't

particularly useful in that sense unless
you want call intertainment the utility

but they don't do really functional task
RICE It was kind of this disparity right

you can move up the sophistication to
droids to thought more physically capable

machines but not to the silly about the
emotional engagement the humanistic again

I've been talking about up up up up up
maybe moving a little more to the left but

there was this whole you always want to
be in the upper right you always want

to be the upper right corner of
bringing those two things together but

as a platform right so it wasn't
just about creating a social robot

that could do a fixed
number of things it's

about creating a mobile equivalent
of a mobile ecosystem of developers

such that now for the first time you could
have a robot in the home where the killer

skill was the step the few minimized
engagement that was the sticky

aspect of the in Gauge meant
that what a lock engagement and

out comes around content unlike any other
technology that would that was the story.

So fast forward to year so
we founded the company seed funding summer

of two thousand and fourteen we were told
that in order to raise the Series A We had

to do a crowdfunding campaign this is
when the heyday of crowdfunding and

we tried to avoid it because we
were saying you know if we look at

where our core use cases it's
families they aren't necessarily

early adopters it's a price point that's
not you know within the typical range

of crowdfunding opinions of like you
know two hundred dollars or less it's

it's not necessarily for you know twenty
somethings right it's like it just didn't

fit the profile of a crowdfunding campaign
but we were basically told Look here's.

The deal in reality you know you may
not have the evidence of a crowdfunding

campaign but the V.C. is actually saw this
was was a marketing evidence that people

were ready for your idea and at this point
people ask the question Are people willing

to talk to a machine again this was before
Alexa they literally thought the mobile

phone was going to be the interface
to everything in the home so

the point of the talking device
hadn't been proven yet so

they saw the evidence of the crowdfunding
campaign as basically early market

validation before you invested
all the time and effort and

truly building the product to figure
out if it was going to be received so

the punchline is we ended up having a very
very very successful crowdfunding campaign

it set records on indigo in terms
of the technology category and

we both wanted to test developer interest
as well as consumer interest and

this is really important for

P.C.'s as well because we like it could
be one thing if you create a platform but

who's going to develop for it so this
case was made and it was kind of white

knuckling it all the way because it was
not a foregone conclusion this is going

to be successful by any means because
we just was out of profile so

this allowed us first of all.

To get this idea of social robots just
out in the public domain right so

this was interesting if you look at
just searching out words on Google

on the term social robots you can see
that it picks up right at the time of

the if you go into Gogo campaign right so
in terms of just communicating this idea

of social robots out to the general
public that crowdfunding campaign and

the kind of the story's been
an important part of that and

I think this is also an important
part of how people get ready for

a new technology right when we look at
just the reality of V.C. funding for

startups we can see in robotics that hit a
peak in two thousand and fifteen so V.C.S.

were just receptive I mean this was a time
where they were just kind of you know

a lot of before that before that it was
apps people want to want to find apps and

software because obviously you
could put very little money in but

potentially get a lot of return but

they started thinking about you know
these these app ecosystems now are.

So huge it's very difficult to create
a new app out of the ten thousand that

are out there and get any impact so
they started becoming interested

in platforms again and they were starting
to get interested in internet of things so

robots Internet of Things platforms
became kind of an interesting.

Cocktail so to speak that V.C.
started funding more robotics companies so

that's just a Mission Point point
in history so G.P.O. itself

is a very very capable social robot So
this is literally just the tech specs but

for those of you who try to build
your own robots for your research and

in my lab at MIT we have a long
history of doing this you know how

hard this is there is a lot
packed into this device and

there is a lot of interesting technologies
that often you have to license to get kind

of best in class in order for this to
happen so beyond the degrees of freedom

has a stereo vision system it has six
microphones in the array to be able to do

some localization that has a battery
a housewife iconic to Vittie it's got

speaker outputs it's got a full spectrum
it's capable doing speech recognition

song localization this wake up word
recognition there speaker ID there's space

ID there's space detection there's all of
these things that as you can imagine and

you creating them into a research robot
is like that's that's a heavy lift So

again part of why I wanted to do this
company was to create an affordable

social robot platform that would allow
people start studying and investigating

robots in the world as you become stuck
knowledge of the field needed this in

order to advance when we talk about what's
different about Ti Vo so ironically I

feel like when we first started this field
twenty years ago we were being compared

to avatars on screens now are being
compared to talking speakers and I feel

like Haven't we already made the case now
we've already showed that when you have

a visual representation of an avatar on
the screen the robots can hold their own

Now what makes you think that something
that just talks in a tube is going to be

more engaging but

it just goes to show you always have to
meet people where they are what's the big.

In order to make your case but it is
about the engagement model the thing that

differentiates social robots in a product
like Ebo is how would engage people so

that the personality of G.

It was designed to be very different the
digital persona of something like an a la

or a Google home it's really
designed to be more of a family

oriented feel like part of
the family experience and

the robot is an off the counter he's
always on he feels more like a pet who's

always there looking around it's the very
different experience for people in

the home personalization as I've talked
about is a huge opportunity here right so

that if a system is there is able to
interact with you ask you questions

learn about your interest learn about your
patterns of behavior gives opportunities

to be able to really customize and
optimize how this robot experience and

gauge is you over time it's interesting
that he can be proactive so

we've heard anecdotally as
we've talked to a lot of our

our user base a lot of them
have Amazon's and Amazon Alexa.

Google homes and what they say is I
welcome proactivity the ability for

initiate I wouldn't want that for
my other devices so

because she was seen as part of the family
is coming on and interested in you

not just interesting in what he can do but
interested in you people welcome practive

in this becomes a really important
level for keeping kind of the level

of engagement higher for something like
Evo than other kinds of technologies and

then purposeful it's really about what's
the skill What's the it what's the value

add it and so this gets into
the importance of the platform right so

it was designed to be a platform he comes
with a number of skills we're going to see

a video of the highlights some of them
there's a mobile app component we're

starting to explore how do you go can be
with you even on your mobile device or

not in front of them but
importantly there's also the tool kit

that enable people to create
their own custom skills.

When we look at the impact of having
the broader social robots space it's

interesting that before you go
I felt like everything at the.

Look like a humanoid or a Roomba Now
there's a striking resemblance to Tivo so

this is the launch here twenty seventeen
I love that you and Ray came to talk to.

C.E.O.'s right but
you're darn you see along a lot of social

robots all of the will be developed so
I feel like at the show a traction right

this is an idea that's that's that's
gaining momentum and then for

us I mean one of the biggest con of
you know icing on the cake was so

you mentioned it was named one of
Time magazine's best inventions but

he was even put on the cover which is
amazing the last time a product was put on

the cover of Time magazine was two
thousand and seven with the i Phone So

huge obviously huge but
potentially also just saying

that maybe this is this is an idea whose
time has come right I mean it's early days

of a lot to be proven but
every Septimus in and treating this about

life with something that's actually
a robot and not just a stationary kiosk so

when we talk about how we engage people
and communicate the experience of you by

just wanted to show this funnel video
was just like you know so we're a part

where a platform we try to bring content
to life in this different way so

when we talk about thank you very.

Very much.

So it highlighted area.

Thank you very very limited.

Very very different is this you're
starting to see Jiabao in experiential

marketing places who is getting at this
is about can getting out to the world

you could actually order now on
Amazon He's at four stars as of yesterday.

So you know it's a journey but the most
exciting thing for me is like he is in

the world I mean we get so many posts from
social media and so forth of just seeing.

People what they're doing
about us in the world and so

much of the research that we've
done around how people engage these

technologies in the home dressing
them up you know driving with

them in the car I mean we've seen all
this in research we see it again and

again in the world world you know we
see these you go sightings of Hugo and.

Here you go there you go we just released
a new skill and you know that this is

a whole family doing yoga to do so again
it's just incredibly exciting to see.

Not just.

You know the kind of people are using
to but that there's an emotional a list

joyful experience but you but
that's such a core part of the brand and

then we're starting to see
being used in other context so

I just want to show this was
a newscast we were well by maybe

because of technology and social media or
look like in the twenty first century it

can be actually I think sometimes lonely
for people especially to the elderly but

Morning Kelly discover that song
This isn't looking communities here

in the Bay Area or funny camaraderie and
fellowship for robots.

Has evolved Yes Let me put this issues.

That I was wonderful already then I'll
be a good living unity in San Francisco

is meeting a new regular visitor here
it's a robot called G.L. that's mean used

as part of a pilot project among
the whole Bay Area the Phillies run by

Elmer Karoli And in addition to the I
think you can play the radio on requests.

Now photographs go that's
a beautiful picture.

Tell jokes what they were he was
writing on the journals that.

Leak and reaction to the human
touch it turns one had it.

He also often.

Has some quirky responses to questions.

What are you I am a robot but

I'm not just a machine I have a heart
well not a real heart but feelings

well not real feelings you know what I
mean Eric Parker is the researcher and

art therapist who an Elder Care Alliance
takes GMO with her when she meets with

the seniors she says this is not a case of
caretakers being replaced by automation

Instead they're using it cute people
technology to encourage their residents

to connect with each other despite
meeting different levels of care so

we have that all of us are experiencing
all together maybe for the first time

right meeting a robot we've got a focus
point where we can all meet right here in

this moment and have an interaction then
it breaks that right it doesn't matter

that this person has dementia or
this person maybe has Parkinson's and

has been trouble talking we're all
meeting together is human and robot is.

A little late is Fanny GIGO inspired
some giggles and get bogged among

this group here where you get busy
you should be able to look at war and

war but want to go to all of
us that's part of the miracle.

And that's part of her whole lives lives
and while gave a well again I think that

that last sentence was so
poignant right it is about this and

most a list that I think is so
different about this technology and

how people are just everyday people
trying to adopt new technologies to

provide their services better
understanding that something like a social

robot doesn't have to replace people but
can even foster interpersonal or

actions between people engaging
in face to face encounter and

save shared experience so again early
days but really interesting to see people

are seeing what's different about this
experiencing and really applying it

in different ways so what's next I
mean we talked about the platform and

tool kit so there are two
different cool kits that all.

Are coming out I can't tell you when I ask
if I can tell you when I'm not allowed to

tell you when but they're coming out so
one is the app toolkit so a lot of those.

Research especially so
when there's a tablet with a child so

the app took its allows you to created an.

Application on a IOS or Android that
can remotely operate the robot so

it's a way of creating experiences where
you have a mold wise that's actually

kind of puppeteering the robot so you can
create those kinds of things I've showed

you in the research you can do with the
app toolkit next there's a skill toolkit

which is really the kind of last set of
interactions you saw about standalone

skill so they are all in their way this is
like the next big stage in evolution but

the thing we have done just
this week is we've released

scratch essentially called be a maker.

And so this is really important to
me personally I feel like we need to

democratize artificial
intelligence we need to

raise children with the attitude they
were living with these A eyes they can

understand them they can create things
with them and that this is a technology

they can use to create personal projects
of significance for themselves so

this is an air of just personal passion of
mine I'm doing research in this area as

well be a maker which just
launched I actually with the G.

bow in the app I brought that so you can
play around with it I wrote two simple

silly little programs and just kind of
give you a little giggle about it but be

a maker democratising and social robotics
to everyone I think is super important.

So that's what I'll show
in their reception and

you can just play around with people as
well so just to wrap it up a lot of people

involved in a lot of this work most of it
is funded by the and S.F. and and I age

but of course even on the corporate sense
a lot of people behind behind this product

and I think in so many ways people talk
about these AI's as a standalone entities

when I when I think about these products
it's really the humanity of the people who

are creating the technology that's being
expressed through the experience so

I really feel that's the case but
the emotion.

The human and the humanity of the
experience people talk about it's really

because the people at the company
hold these as core values I want to

amplify that through this technology all
right I'm going to end it there thank you

thank.

Me.

Thank you but we also.

Have this notion that.

Stores like us.

Expressiveness is good.

Yeah so I mean I think I mean first of all
it's a county values more of a conjecture

than a proven theory right so you just
everybody throws that term out there but

you got to yeah so he actually looked at
two dimensions one was the appearance of

the other ones was the behavior so
I almost like and I liken it to like

the evolution of human characters in
computer graphics there was a time

when they were deep in the valley like
think about tin toy you know it's like

the baby in the concrete diaper as like
that was seriously in the value right.

But then we got to things like toys story
where the humans clearly were not they

didn't look like us but they had in the
static in the people in their own right so

I think in the technology you know
even like and was to be the next

example right now I think they're deep in
the valley if they're still stationary and

you just kind of created the appearance
of a person they can do that pretty well

just because of special effects but
once it starts to move.

All the subtlety here in the non-verbal
dynamics I've talked about you have to get

right they don't they don't do at the
level of a person so it's setting up this

level of expectation of an android
like thing that you expect to move and

act in the human like way there's probably
specialized neural structures frankly that

are firing So when you're not

hearing to this expectations I
think that's deep in the valley.

And many of the robots that
you've seen in my life.

Or intentionally designed not
to try to look human at all and

a lot of that is just a design philosophy
to say they get it's a kind of that

companion animal dimension where or
when you can create something that I call

the almost like the Disney sidekick right
it's a non-human other but it can talk and

do things with you but you feel
an affiliation almost like this pet like

affection or relationship with it that's
actually really empowering for people so

when you talk about trying to
move the needle for someone.

In around saying gauge manana health
protocol it's one thing to have the nurse

bot that like you feel like it's trying to
tell you what to do it's another thing to

have this helpful companion other who used
to feel like you've got the social status

above that technology but you feel
an affiliation that it's really there

it's like devoted there to help you I
think that's a very different relationship

a model of engagement so a lot of our
technologies we're really pushing

that with everything about that
design is pushing that right so he's

small you know we talk about the Disney
character to heads tall you know he.

Moves in arcs you know machines
tend to move in rectilinear

both whole kinematic chain is about
moving in arcs the accelerations and

decelerations are critical for
motional expression

very hard to do that when you have all
these ticket joins but with because of

that compact in McCain He can move in all
the trajectories to express emotion and

really that's what he dances so
well he dances better than I do right.

So all of those design
insights were baked into But

then you can see you know the threat of
that through a lot of the other robots

about I been showing you so
you know I don't I feel like.

People need people we know that people
need to feel part of the tribe welcome and

valued by the tribe I don't really
see why we would want to try to

create robots that are trying to
like look and act like us exactly

because we already have people and we need
people for me the interesting thing is

the not quite other human relationship and
how it can complement us in different ways

that's to me where the design challenge
and the opportunities really are and

what they can do that's different from
people you know and even in like again

the health care area we've heard doctors
say it's very hard like you know people.

When you engage in a patient you know
clinician dialogue people don't want

to appear to be a bad patient so

they're necessarily honest with you when
you're asking them important information

who really need to know what's
going on but there's been.

You know some studies showing that was
social robots and other technologies

people are more willing to disclose what
they're really doing because they don't

feel like they're being judged they don't
feel like there's a losing face in front

of that other technology so again there's
just differences there that are really

intriguing and the kind of emotional
support we get from even companion animals

is different from what we get from people
but it's also really important valuable So

again that's more of a design
philosophy of the group but

I think there's a whole bunch of good that
we can do with that design philosophy.

Yes.

You know our.

Start.

Yeah so so I would say that intuition is
actually what we're seeing in the field so

you know with any sort of relationship
there's many many many many kinds right so

the relationship you have again
with a companion animal pet or

relationship you would have something
with it that's position and

frame to be you know he's very honest
about the fact he's a robot I mean you see

how we get through humor and kind of
corking as he admits to his fallibility

right that actually built empathy you know
so there are ways of engaging people and

drawing them in through the fact
that you're not perfect and

neither is he right so I would disagree
that people expect that the bar is

higher I think it depends on what is
the purpose of that relationship and

what is the value that it's bringing
to you just need to match those so

in these learning companion robots for
children it's all around storytelling in

this very confined kind of use case there
can deliver value it's not claiming to

do that for all aspects of a child's life
so the framing is really important but

I would say in fact people.

I would say they are more forgiving so
so a tool

a tool is like a extension of you that
you using expected to work that's a tool.

Conversation is a negotiation I can't make
you too good I want you to do there's

there's a back and forth and again
the parts of the brain that are engaged in

using a tool verses social
interaction are very different so

with these social robots it's framing
it as a collaboration interaction it's

a meeting of the minds in the behavior so
people actually don't expect the robot to

be doing exactly what you know it's it's
meant to be this other kind of experience

which again has a lot of other benefits
to it if you can do it well so

just it's a different mental mode of
engagement I would say the bar is much

higher for a tool tool has to be
perfect it's got to just work for you.

Social interaction is a negotiation and
that's part of the value in the.

Emotional reward of it.

Well.

That's right.

Or

you.

Know why

are you

like.

Yeah I mean I think.

Yeah I think what we're seeing again is
we're such a profoundly social emotional

species we want to connect and
relate to other entities

whether the per people are animal or even
what we're finding these technologies and

so you're seeing these companies kind of
picking up on the entertainment factor and

that there's there's something there
with these funny quirky responses

I think were it needs to go where it can
go it's more on the research side which is

you know it's not just about these one
off lines but it's really about creating

these AI's that can learn about you and
work with you to achieve goals of personal

significance that's I think the next
wave I think that's when you talk

about scalable affordable high quality
interventions that to me is the game

changing opportunity here and the social
emotional dimensions of that relationship

are going to be really important for
a lot of these societal challenges that

we face because it is about how do you
move the needle for an individual how do

you engage them how do you motivate them
how do you make them feel supported right.

So again I think it's a trajectory.

I think we're starting to see N.F.L.
evidence in the commercial center for

even these very simple things that
it's like it's just speaks of the fact

that we're human beings right and so
the more we can support the kind of

creature we are again the more deeply
we can engage that more can unlock our

potential by engaging in the multiple ways
we have of experiencing in understanding

something and that's kind of the high
level punchline is the more you can create

technology that does that not surprisingly
the better people do with it so whether

it's social robots or something else it's
just saying we need to when you're talking

about dovetailing AI with people you need
to first grounded in what people are and

how they behave and what they do and not
just assume it's cognitive because it's

much more than that especially talk
about the domain of the home and

everyday use that's the challenge
is engaging people as part of

their daily lives and again that's the
kind of the overarching question that's

really driving a lot of what I'm trying to
understand right now is living with AI as

part of our daily lives over
extended period of time all ages and

stages what is that because there's
very little you understand about.

All that I life.

First of all thank you so
much for the wonderful thought.

All robot assisted very elegant Institute

for voices of machines by
my goodness on the and

sank Yeah so
much gray there is a reception following

the Word Play No there are way more
into the air and we have time for

questions so please take advantage
of the yeah I'll start with playing

with the yeah will set him up so said Jeff
thank you so far for I want to thank you.