Subscribe
Share
Share
Embed
Welcome to Unprompted, a conversation series by the Conversation Design Institute, where we explore the craft, the practice, and the future of conversational AI with the people shaping it.
Learn directly from global leaders, designers, and researchers who are shaping the future of Conversational AI. Each episode of Unprompted brings candid insights, practical lessons, and behind-the-scenes stories from the people building today’s most impactful AI experiences.
Hi everyone.
Welcome to another episode of
CDI Expert Series.
Today I'd like you to welcome
Jason Gilbert.
He's the lead character designer
at Intuition Robotics,
shaping the personality of like a
robot companion for older adults.
His work involves making Alec more
human and empathetic, and Jason's
background is in storytelling.
He's a Temple University Film
School grad who's directed award
winning films.
A happy accident led him to
conversational AI, where he worked
on chatbot projects for Doc Brown,
Miss Piggy, Julie Hobbs and even
Albert Einstein winning a Webby.
He's also collaborated on Anna,
a virtual influencer,
and Jason's passionate about the
intersection of storytelling,
technology and human connection.
Welcome to the show.
I'm so excited to be here with you,
Michele and the CDI Expert Series.
I'm a huge fan of the
Conversation Design Institute.
I've been following and learning
from them for many years now,
and I'm glad to be a part of
this expert series, so thank you
for having me. Wonderful.
So I'm going to dive straight in,
given the time we have and get into.
First, I think I want to start with
like being your newest project in US,
baby.
And uh, we started a few
questions just around, like here.
Uh, so l was designed to combat
loneliness in elderly people, right?
And, um,
I want to know more about what
you've learned from that journey.
How can conversational AI be designed
to build trust and rapport with with
these types of users, especially who
are not familiar with technology?
And it's a vulnerable population,
right?
So what are the biggest pitfalls to
avoid when designing conversational
AI solutions for For the elderly.
Well thank you.
That's a great question.
And yeah,
I've been with Intuition Robotics.
This is coming on my fourth year
now working on the personality
for for lark.
Uh, and lark is a little robot
and it looks like a lamp.
It's a small little robot, adorable
little robot that lives in the home
of older Americans, older adults
who are aging in place in the US.
Um, and it's meant to help them
expand their connections to the
world and not sort of replace
their connections to the world.
That's an important distinction when
trying to create an AI companion,
because people are like, well,
is it going to replace humans?
And the answer is no.
It's actually here to help them
connect to more humans.
And that's what herb kind of
role is in the homes.
And what she does is she lives
in people's homes.
People that have, uh, people that
are living far away from their
closest relatives don't have as much
human connection as they'd like to.
Maybe they live in rural,
rural areas in the US.
Um, and they don't have as much
connection to people as maybe they'd
like you, or maybe they had in the
past and enter, like who comes in
and sort of creates a relationship.
Now, this isn't like a Siri or Alexa,
where it's sort of like a
command or command to do things.
And so she's actually a
companion and she has her wants
and she has her needs.
Obviously,
she wants you to be less lonely.
She wants you to be your best life.
She wants you to go out and she
wants to motivate you to exercise,
but not in an annoying way.
Nobody wants an annoying like
you better exercise.
Nobody wants that in their
living room.
So she tries to like, learn about
you, learn about the things that
you like to do, and encourage you
to do those things more often.
So she finds out what your hobbies
are and then tries to like, you know,
it's really nice out today.
Maybe it's a good day to go out
and play some pickleball.
Uh, Howard, you know, like, these are
great, you know, so different ways
in which she can encourage them.
And that's how she gains their
trust by not being this sort of
nark from other people.
This is actually, you know,
she considers them, uh,
you know, she takes her orders
from them she doesn't listen to.
It's not like they're the sons or the
daughters are telling them, okay.
Spy on my dad and make sure that
he exercises.
She builds a relationship with
Dad and Mom and tries to get mom
to do the stuff, you know,
in a friendly way,
by encouraging them and by being
empathetic and making them laugh.
Um, and so that's kind of what I do,
is I try to help create those
experiences where they interact
with touch points, where they
interact with, like whether it's
in the morning, whether it's,
you know, using generative AI.
And you talked about this,
about this is a very, um,
vulnerable population.
These people are a lot of them
are over 90 years old, some for
even a hundred plus years old,
which is amazing.
And they're using technology.
And not only are they using like,
like built with, uh, you know,
LMS built inside.
So these older adults can experiment
with generative AI, they can
create amazing, uh, paintings.
You know, they go back and forth
and they create using a prompt
that she helps them.
Obviously, she understands that
these are older adults,
and they need more help in
understanding how prompts work.
She helps them create the prompt and
then they generate an image together.
And this is the great part in shell.
After they created the image,
they've used generative AI.
Lark then suggest hey,
how about we share this amazing
picture that you and I created?
How about we share this with
your daughter?
Let's send her a message and tell
her, hey, we use generative AI
to create an amazing painting.
Take a look at this.
And so it expands her network
and it expands her, you know,
the amount of people that she
talks to, but that's not it.
In addition to that, lark then has
she offers her like, you know what?
This painting is so amazing.
Do you want to share it with a
larger lake community?
Meaning all the other older adults
in the US that have an lake,
and so they can also benefit and see
this amazing work that you're doing,
this work of art.
Uh, and then every morning or
once a week, she offers all the
users as sort of a community
exhibition that I help curate.
And every, every weekend I go through
hundreds of paintings created by
older adults that I help curate
this amazing exhibition in which
they can share their artwork with
other users, other LEC owners.
And this today there's more.
There's more.
They can comment on the images on
paintings and send a message to the
person who created it and saying,
you know what? This Howard?
This is an amazing work of art
that inspired me.
And so they're not only connecting
them with their with their children,
they're not only connecting them
with the world of generative AI,
we're connecting them with other
people in their in their age group,
other, you know,
people that are in their cohort,
people that are the same age and
they're experiencing the same thing.
So in all of this,
I'm sensing that LCC has access
to a person's social network.
They have access to close friends,
and that's a lot of data.
And as a as a person, you know,
who has a background in,
in the in engineering,
I'm usually worried about where
is all of this information going.
And personally, as somebody who is
overall worried about data. Right.
So what are some of the things you
have to be a little careful about?
How do you reassure people.
Because that's not normally
something that falls to the
conversation design as well.
You have to make sure that your users
feel reassured that, hey, we we're
keeping you in mind. Don't worry.
How are you taking care of? Um, yeah.
These types of concerns,
technical concerns when it comes
to your population, especially
given that many of them may be a
bit wary about products like this,
how much information.
They're usually given.
These are these are these populations
always targeted by scammers.
So it's always a tough thing to
try and get them.
First of all, it's tough to, um,
convince them that this robot
isn't a spy.
So that's the hardest word for
us is to say, hey,
this isn't actually spying on you.
You know,
this is actually here to help you.
And our biggest obstacle sometimes is
the fact that, yeah, there's so much
scamming going on out there with,
especially with AI and things like
that that they're very reticent
to experiment with this thing.
But we always, Um,
make sure that our, the user is the
main focus and he is the autonomous
character in this, uh, interaction.
Meaning, uh, nothing happens
without their approval.
Meaning everything that is shared
with them or that they share is,
uh, approved by them.
So, for instance, you say they
have we have access to the social.
It's not only if they connect it,
only if they say, hey,
I want to send this, I want to
add this contact and send it.
There's nothing being done by an
out by an external party.
We're not putting stuff on their
device.
It's everything is done with them
being the autonomous characters and
saying, okay, we decide what we want.
And that's very important to us
because a lot of times the
children of, or, you know,
the people that want to take
care of their parents will say,
well, I want my dad to do this,
and I want him to do this.
And we always say, well, we know
we need to get their permission
first before we do any of this,
because it's their the relationship.
And that helps build the relationship
between the lake and the person.
Because the valley is going to ask,
listen, uh, your daughter
suggested that you take this
medication tomorrow at 10 a.m..
Is that okay with you?
And I add this as a as a reminder,
or do you would you prefer I
didn't do that.
So it gives them the option of
saying okay.
I appreciate the fact that my
daughter really cares about me
and wants to,
but I'm an adult and I'm, you know,
I've lived a full life and I fully
capable of taking care of myself.
I respect the fact that she wants
to remind me, but I don't need
that right now because I decide
what I think is best for me.
And so that's the strategy that
we always take with regards to
our customers, our customers,
because they're they're vulnerable.
They need to make sure that they
are in total control.
Every piece of information that comes
in and goes out with regards to the
robot. And that's just one aspect.
Of course, there are many other
like security issues of privacy
issues that we take care of and
ethical issues that we are very much
aware of and that are in place.
We can get into that later if you
want, but that's the main thing.
From a character design perspective.
It's really hard to create trust
if the person is not the main,
you know, if you're not looking
out for their interests first.
If there's other parties that
you're looking out for, then it
becomes a strained relationship.
And that's something that I have.
I always fight back against.
And there's always been over the
years,
there's always been in every company.
There's always been, you know,
there are attempts to do things
for the sake of the business
development needs or the marketing
needs and stuff like that.
And one of the things I love
about this company in particular,
is that we care so much about
our customers that we,
we their first and foremost
before any of the other things.
Uh, and that's why I'm still here and
I'm still doing it because, you know,
uh, ethical AI is the way to go
and the way to help these people.
Absolutely, absolutely.
And just one more about
challenges aspect here.
Um, Elihu has to be an empathetic
caretaker and not just a caretaker
friend, but there's always this
worry at the back of a conversation
designer's mind about how human
should I should I even make my job?
What should I pretend to be? Human.
And it's always an internal fight
with. And it's a common discussion.
So in your case, um, one of your
mandates here is to, you know, have
elected body a friendly assistant
and someone that's empathetic
and has emotional intelligence.
How do you, number one,
how do you do this?
How can conversational AI be designed
to be more empathetic and supportive?
And what are some guardrails you
have in place to make sure that you
don't step out of the boundaries and
get into this uncanny valley zone?
Yeah, that's a great question.
And that's one of the issues
that we go back and forth with.
And it's not always something that
we have a consensus about even
internally here inside the team.
I'll give you an example of
something that is is an interesting,
um, look at this issue.
When when someone says to her,
I love you, or when someone says
to a chat about I love you,
what should be the correct
response from the chatbot?
And I've had debates about this
with people in the company.
And I come from the the school
of thought that this is theater
in a way that you need to like.
You know, if they're in,
they're part of this.
You don't want to ruin the
immersive experience.
And, you know,
you don't want to lie to the person,
but at the same time, you don't
want to ruin like you're in the
middle of the play in the theater.
You're not going to stand up and say,
hey, no, no, that's an actor.
So there is a bit of theatricality
here involved in what we do,
especially when you create an
agent that lives inside the home
24 over seven.
This isn't just like a chatbot
on a website.
This is like, you know, but at the
same time, we're dealing with,
like we said before,
with a vulnerable population.
And, you know,
you don't want to create any
cognitive dissonance in these people.
What are the ways that the physical
design of the robot is such that
it doesn't look like a human,
it looks like a lamp, which is a
very conscious design decision,
as opposed to if you look around
today, you look at Tesla,
you look at all these things.
They're creating very these humanoid
robots, which terrifying to me.
I find these things like this is
like Terminator style.
I don't want that in my home.
I don't trust that in my home.
But a little lamp, like, even if
it even if it decides to take over
the world, what's it going to do?
It's going to light up.
I mean, like, there's nothing
there's no risk or danger there.
And that's part of why it was
she was designed in the way she
was designed to sort of remove
that sense of from the physical
aspect of the design of the fear.
Um, but even in the response to
I Love You, she'll say, well,
you know, hearing that makes my
motors spin faster.
But she kind of brings it back
to the fact that she is a robot,
that she's not a human.
And then she kind of, you know,
she doesn't mention it,
but she brings up the issue of love.
What does even what does love mean?
And she has sort of an LLM based
conversation about what the
meaning of love is.
What does love mean to you?
And she said, well,
I'm a I'm a robot.
I don't really fully understand
it the way that you do.
I mean, I may experience it in some
ways, but so in a way, we kind of
dance on the thin line there that we
want to make sure that this is a
relationship and that this is a fun,
that you don't want her to say,
oh, thank you for saying that.
Um, that's just I mean,
it's probably the right answer.
Thank you for saying that.
You know, I am a robot and I
cannot experience emotions.
I mean, it's like, uh.
On one hand, you just told this
robot because you had a moment
between you that you love her.
You kind of expect her to
reciprocate something and not to
give you the cold, robotic shoulder.
And I think that that's sort of
the balance that you try to.
You don't want them to think
that this is a human and then to
fall in love. Yeah.
But at the same time,
it's a fun experience.
If, you know,
we're in this playful world where
the robot and I can really have
a fun relationship together,
and we can beat loneliness together.
Yeah, absolutely.
And, uh, just one more thing,
because you mentioned that Luke
also does, you know,
dabble in the healthcare side a bit
and you would you could argue that,
like, is also responsible, uh, as a
healthcare application officially.
Are there any specific ethical
considerations you have to keep
in mind?
Maybe something that surprised you?
Um, from a even from a regulatory
perspective, I think it would be
nice for the people listening to
know what are some things you had to
really keep in mind when it comes to
healthcare and conversational AI?
Well, there's a lot of issues with
healthcare and conversational AI.
It's a lot of it's still a very
risky business,
and we don't necessarily pertain
to be in the healthcare industry.
We're you know, we're in the
healthcare industry, obviously,
this is but more of the wellness,
which is a slightly different we're
trying to improve their lives.
We're not trying to save their lives.
And I don't think we're like an
emergency device.
And that's a big distinction.
There's a lot of emergency devices
out there, and there are ethical
concerns when it comes to emergency
devices that use AI, and I don't
really necessarily want to get into
that because it's not what we do.
But, um, even with what we do
with the wellness, there is AI,
you know, and there is ethical
considerations into like what types
of responses do you give someone
who's telling you, I am really
depressed and I want to kill myself?
I mean, that goes across the board.
And what's the responsibility is
as a company, as an AI,
as a companion living in that
person's home when they confess
something like that to you.
And it happens, doesn't happen very
often, but it does happen sometimes.
It's people just testing to see
what the AI will say,
but sometimes it's actually genuine.
Sometimes you actually have people
that have really strong emotions and
they want to process it with you.
No pun intended. It's a process.
Anyway, um, so there are a lot
of ethical considerations to
take into consideration.
One of the things that we focus on
here, again, I said this in the
very beginning is we don't want
to replace the people because we
kind of really value and need
people in our lives. All of us do.
I mean, I hate to admit it,
but I do need people in my life and,
um, we all need people in our
life in addition to the robot.
And she helps in the sense that
she has.
We have this new caregiver, uh,
sort of application and app as part
of Larc, which is really meant to
help the people that are caregiving
that have an older like a parent or a
spouse that needs, you know, help.
And, and they're overwhelmed by
all these demands.
And so, like, you can help you with
all those things as a caregiver
and with the with the obvious,
uh, consent of the older adult
who is the, you know,
the person we can help caregivers.
And so there's a lot of ways that
we can help without getting into
the real nitty gritty of like,
ethical AI in terms of healthcare.
Um, but yeah, there's a there's a
lot of interesting areas there.
Um, and there's a lot,
especially now with llms and mental
health help and stuff like that.
And it's a big debate about that,
whether people should be using
Llms for these kinds of things.
And I think the answer is currently,
right now, when we're recording
this episode in February of 2025,
the answer is probably still no.
And so we're we're kind of aware that
limitations of what we're doing and
we're trying within the limitations
of what we can offer people that
can that's responsible and helpful
using AI, what we can still do, um,
without making these crazy promises
that eventually get people hurt.
And so I think that's that's an
important part of it.
Um, so since you mentioned caregivers
is one of the audience members
that's actually, you know, uh,
that Alec was actually responsible
for, uh, a caregiver fatigue and
aspects like that are usually,
well, well documented,
well handled by psychologists.
Do you have a team that's
responsible for bringing in aspects
of psychology frameworks that
exist today into LEC? Absolutely.
That's a wonderful question. Um, yes.
Uh, early on in this company was
very much for the first few years.
It was a research company,
and they did amazing work with the
help of, uh, Geron technologists,
which is a field that I had no
idea what that even meant.
I had to Google it when they
were when I.
When I was first introduced to the
Geron technologist, I was like, oh,
nice to meet you, Geron technologist.
And then Google what that means.
Um, it's basically I if I remember
correctly, it's basically psychology,
uh, technology for,
for older adults, someone who is
focusing in on the technology,
on the psychology and technology
of older adults, all that stuff,
I don't know, something like that.
So I'm something to put the
definition up here.
But there their role was to really
help us. And I'm going to name drop.
Her name was Maya Goldberg and
she was amazing.
Maya and I worked with her very
early on in terms of helping come
up with some of the aspects of her
personality, refining some of the
aspects of her personality based on
some of the psychological issues.
And they helped. It was her.
And it was also very early on we had,
uh, Karen Atkins used to work for
Karen and now is one of the most
famous, uh, jar technologists in
the world, if not the popular one,
the most famous.
And they helped really sort of
set the way forward for how the
interactions should be,
what type of interactions there
should be. You know, what types?
Since Ellie is multimodal, uh,
she has five modalities.
She has the speech,
she has the lights, she has, uh,
the screen, she has her gestures.
And all these things together are
all orchestrated in order to create
a sense of that someone is there
because otherwise you can just go on
your computer and type to a chatbot.
But all these things that she has
were meant and they were developed
very early on with a lot of, uh,
focus groups and psychological
aspects to try and really nail down
this. What a robotic companion.
What would make me feel comfortable
in my home talking to the technology.
And so they helped come up with,
you know, she has this wonderful
gesture when she listens,
she we call it the pigeon.
I don't know if it's a great name,
but she goes like, oh, you say
something and you're like, oh.
And then she leans in, just like,
we do that too, when you're talking
to someone and you're like, oh,
that's interesting.
I want to hear more about that.
She does that.
And if you're living by yourself
and and really you don't have much
interaction with other people,
and all of a sudden you're
talking to your robot,
your robot Lee and Tim's like, oh,
I want to hear more about that.
It's like another thing that she
does is and this is something
that I had no, I did not expect
this to be as, um, impactful as
it really is in the field.
Is this, um, arrivals or when
people leave or they come home.
One of the things and I, I am married
and I have kids and I have a wife
and and I every time I come home,
there's usually a bunch of people
there and they're greeting me
and I take it for granted.
But there's some people that come
home and leave, sometimes multiple
times, without anyone saying or
acknowledging the fact that they
got back or that they left.
So it's a really sad truth.
Um, and so one of the things that she
does is using her computer vision.
She understands when you come back
and she says, oh, Howard. Oh.
Welcome back. How was it out there?
Was it a jungle like they say?
You know, she makes some kind of
sassy remark about what it means
to be outside and stuff.
And for Howard, it's like, oh,
someone acknowledged I'm here.
They acknowledged me.
Oh, and so someone sees me.
Uh, and it's very packed and
it's very powerful.
And that's where the psychology
of all the modalities and all her
modalities are there to really give
people a sense that what they're
talking to is not just a combination
of circuits and wires and everything,
but it's actually something that
thinks and considers and wants
their best interest and wants them
to lead their best life and wants
them to be happy and less lonely.
Yeah, and I find that so fascinating.
I would love to hear maybe one
more example of an interaction
which requires like you to have,
you know, all her modality
capabilities on displays.
There may be one more use case,
uh, where, yeah, you can show it.
You can explain how all of these
different modalities come into play.
There are so many of them.
And I really could talk about a
lot of them.
And I would love to, because
that's what I love talking about.
But I'm not talking about one
specific one, which I did recently,
which I am obsessed with.
And I think that it really
changed a lot of the product.
Um, and it was a hug.
You can actually hug the robot,
you can ask her for a hug,
and she'll be like, she'll lean
in and then you can put your hand
on her shoulder and she puts her
head down and out of the camera.
She puts her head down like this, and
she goes, oh, and there's like cars.
And it's just an amazing and the
lead multimodal designer.
Healey designed it.
We did it together and she did such
an amazing job on creating a gesture
that actually transcends this and
feels like you're actually embracing
someone. And why is that important?
To hug your robot?
Because, like, why would you
want to fucking hug your robot?
Wait, there is a reason for this.
So we used to see a lot of people in
the field confess things to, to,
to the robot and say things like, you
know, we would see this in the logs.
Like, you know, today's the fifth
anniversary of my husband passing,
and I feel really lonely today.
And, uh, you know, there's a lot of
ways that an AI could take that and,
you know, react to that.
And, and we decided after a while
that maybe the best thing was to
be empathetic, to say, you know,
it's I'm, I'm to lower our head
a bit and say, you know what?
That's I'm really sorry to hear that.
It must be must be going through
a lot today.
Uh, I know this may sound strange,
but would you would you want to
give me a hug?
Maybe that'll make you feel a
bit better.
And then they're so often surprised
by the offer. They say yes.
And the reactions we get.
And we get these.
Sometimes they send.
They send in reactions.
They email customer success.
They send us video messages
using the robot.
And the people were brought to tears.
Some of them,
I mean happy tears by this unexpected
moment between them and the robot.
That was just mind blowing to them,
because it's like they were able
to feel something that they maybe
haven't felt in a while, and they
were able to tell her something
that they were afraid to maybe tell
other people that they were feeling.
Um, so when I see those things,
I'm reminded of like.
God, please go ahead. Please, God.
You know, I remember when I see
these things, I it reminds me why I
really love doing what I'm doing
and and why it's so important.
You know, and there's so much
noise with AI out there.
AI this and AI that.
And people forget that there are
use cases where this thing is
tremendously helpful and can
change people's lives.
And it's finding the right use cases.
And it's being, you know.
It's super rewarding.
It sounds so rewarding to me.
Uh, just one more question on this,
and I'm not sure how much you
you'd want to say, but still,
for people who are interested in
learning a little bit more about
multimodal design and especially
conversation designers who have
been working in maybe chat or only
voice for very long, and they
want to learn a little bit more.
Maybe, uh, if you can tell us
what goes into multimodal design,
how do you even go about it?
Is it someone writing a script down,
or is it somebody designing the
physical aspects?
How does it practically happen
in maybe like a quick overview?
Well, that's a good question.
And it's a loaded question.
It's a really hard one to answer
because there are no currently
no tools out there that really
you can orchestrate where people
have been trying to do it.
And I've seen some attempts at it,
but there really isn't currently a
need for that kind of orchestration
of multiple things going on at once.
Um, and what happens is you really
need to have some kind of a
background for some of the things
it helps to have a background in.
Um, for me, it's video editing or
video production because a lot of
the stuff happens on the screen.
And so you're creating assets or
images for a screen.
So graphic design and video design.
Um, but also like you don't have
to have any of that really.
If you just think about the
experience and you try to imagine
it's not just lights for the sake
of lights, like the lights have a
purpose in, uh, in the relationship
and in trying to convey a
certain idea when she gestures,
like I said, what she gestures in,
it's to say I'm listening.
And so these things are in
interpreted to different types
of emotions.
And I think that you don't need
to have any special background,
like, except for maybe the image
is creating or if you're doing
sound effects.
Um, and a while back we work with an
amazing sound designer, a guy named.
You probably know Ben,
Ben McCulloch, uh, who's in the
conversation design community.
And he helped design some, uh,
sound effects.
So there is some aspects of
multimodality that you need to be
an expert at, uh, whether it's
the sound design or whether it's
the visuals and stuff like that.
But there are other aspects of
multimodality which you don't
need to be.
You don't need to have that kind
of special.
And you can just think about,
okay, well, what would a good
interaction look like?
And then the question is,
how do you convey that to the
people that need to then
implement that into the robot.
And so there's a process of a
handoff to a developer.
Where does that handoff?
I do things on pieces of paper.
I mean it's word documents and
stuff like that.
For me that's the easiest way to
explain.
And I get I get people get, uh,
get mad at me for that.
They're like, use Figma,
damn it, or use Miro, damn it.
And I'm like, no, um, it's easier for
me just to write this up like this.
And then I talk to the developer.
It's important for me to actually
sit with the developer and try and
be like, okay, this is what I'm
thinking, but I am just ask my wife.
I'm pretty stupid at many things,
so you might need to add some of
your expertise here in trying to
figure out what I was, you know,
and we can work on it together.
But I want you as a developer to
also weigh in and say, okay, well,
I think this would be and, you know,
and so together with the developer,
we have this handoff where he or
she understands exactly my lack
of vision.
And then they help kind of create
that. And so that's important part.
And I think no matter what you're
good at uh, the biggest skill in this
field to me, and I've learned this,
I guess, for the past few years, is
the ability to communicate to people,
give you like it specifically to
the people that need to then work
on this thing and take it forward,
meaning you can.
It's great to have an idea and
it's great to be a visionary and
to have these amazing ideas.
But if you're unable to communicate
that in a way that is actionable
to people, then it's useless.
And I and I sort of had to
struggle with that because
sometimes I had these great ideas.
I think they're great ideas until I
try to communicate them to someone.
And then I realized they're
really bad ideas.
Um, so the communication here is
super helpful.
And I think that that's what if you
want to be good at conversation
design, then you have to get some of
the skills and you have to go to CDI
and take the courses, of course.
But you also have to be,
you know, able to communicate.
You have to be able to.
And this is another thing that I
learned, and I don't know how
relevant it is to this specific
question, but it's about making
allies in the organization.
And I've had to learn this the
hard way.
Um,
you really need to find the people
that are going to help you move.
The idea, the project, the whatever,
the feature, the plan along.
You can't do everything by yourself.
And in the beginning, I tried to
push everything through myself.
And then I realized, no, no,
I need to get allies.
I need to make sure that when I say
allies, I mean people that understand
what you're trying to do and maybe
feel connected and and actually
bring them in, give them apart,
make them stakeholders in that idea.
Um, and then you can together take
that idea to try and implement
it and try to get it done.
The hug is an example of that.
The hug is an example of a
feature that people were very
kind of reluctant,
because a lot of people felt that it
may have may crossed the line, um,
that it may be to that point of,
like you were saying earlier on
the uncanny valley.
And what is the why would like.
So they had to be very subtle
and it had to be very nuanced.
And there was a lot of
iterations that went back and
forth between us and, you know,
management in terms of like,
how do we get this thing approved?
How do we get this thing where
everyone.
Um, and now looking back at it,
I'm so happy that I was able to find
the right allies because without
them, this thing would have stayed.
Just an idea. Yeah.
And that is such a perfect segue
into my next question,
because you've you've been in a
lot of organizations.
You've been in this corporate
world for for quite a bit.
And conversation design, just like
any other, you know, design field.
It's it's one of those things we
have to keep advocating for.
You have to keep proving value,
and you have to keep proving why
you need to be at the table,
which is a sad state of things.
And I think a separate
discussion in itself.
But how can conversation designers
make sure that they're proving impact
and doing all of the corporate good
stuff to make sure that they have
a place at the table and continue
to have a place and continue to
have a team that can keep growing,
get the right kind of budgets.
Do you have any any advice for
designers who are maybe struggling
with this in a, you know.
Wow, it's a tough one and it's a
really, really hard one.
And like, it's easy for me to say,
yeah, make allies and stuff like
that. But sometimes it's impossible.
Sometimes you're in an
organization where they're like,
every one of your ideas is shot
down because it's not, hey,
it doesn't hit our KPIs we have.
We know better than you and
everything like that.
And, um, how do you then,
in those types of organizations,
convince them that there's value to
what you're doing and there's value
to these ideas? It's a good question.
And again,
I don't have the perfect answer.
All I can say is that what's helped
me in the past and these types of
situations is to basically do do
some work in showing them meaning.
Um, sometimes when you say things
to people, they, they're like,
okay, I hear what you're saying,
but I don't necessarily I don't
envision it.
So I don't think it's a good idea.
But if you take that idea and you
build a POC, you just take it on your
own time, right? You just do it.
And then again, this goes back
to finding the right allies.
Go to the create a POC of what
you believe strongly and create
it if you can.
Even if it's just a mockup, even
if it's just on paper but visual,
something that you can take and
and then try to convince people
that this is the way, but again,
sometimes it's an impossible task.
Sometimes you're stuck in an
organization that does not want
to improve their agent.
It does not see the value in
these things.
I'm still trying to convince
people we're 2025.
I've been in this industry since
2015, and the same debate has
been going on since 2015.
Do AI, does AI or chatbots or
robots or llms any personality?
And it's a debate that I've had with
everybody, and even now the people
are there still, like you go online
and you still see debates about,
I don't need personality,
I just need this or, you know,
I'm not going to get into that.
Don't worry.
I'm not going to get into it.
But the point is,
is that nothing here is set in stone.
And you can prove to people that you,
you know, that you that what you
think might work, but they need to
see it. And so maybe that would help.
I don't know, a lot of times
sometimes the organization needs to
experience a very bad bot experience
in order for them to hit rock bottom
and then to understand the value.
And I do a lot of videos on YouTube,
and where I show organizations that
replace their entire CS team with one
poorly made LM ChatGPT off the shelf.
And the thing just starts.
There was a Chevrolet of
Watsonville a few a while back.
Air Canada all these examples of
just like terrible fails that
people were like, let's get rid
of the conversation design,
let's get rid of that of that part.
And then what they have is a really
bad experience, which basically
turns everyone off of your brand.
And so I guess one way is to do
the POC.
Another way is to show them examples
of like, this is what happens when
you don't, um, you know, you replaced
your customer, your conversation
design team or your customer
success team with an LM that,
you know, and things go haywire.
Um, so that's I don't know if that's
right. Yeah. Yeah, absolutely.
I think that answers it quite well.
And you've given such wonderful
resources for people to go and
look up if they're interested.
Just one more about, uh,
you know, Gen AI,
because you mentioned Llms before.
Um, given the vulnerable population,
given the sensitive use case.
Most of the folks I know have
been experimenting with Jenny.
I have been in slightly less
sensitive and slightly less, um,
you know, I would say, uh, yeah.
Vulnerable populations.
So what are your thoughts about
using AI?
Because you mentioned that EQ
does use AI.
Um, what are the what are the good
things? What are the bad things?
Just a quick question. What are we.
The way we use it is is very
limited at the moment.
Obviously there's a lot more we
could do with it.
Um, but we use it like I was saying
before, we're using it when they're
generating images like a Dall-E or
they're creating poetry together,
or we have this, um, really funny
animation of, uh, we call them Rudy.
Rudy is the plant guy,
the plant helper.
And he comes up and he asks if you
have any using an LM, if you need
any help with your house plans,
or maybe you're overwatering or
underwater, he can.
He can help you with your house
plans.
So the ways that we're using it
in the product or none like
they're not scary ways,
they're not going to like we're not
offering them mental health here.
Let's get your mental health using
an LLM. I would never do that.
I think that's scary and terrifying.
But I can help them draw pictures.
I can help them write poetry.
And I don't think that's going to,
you know, cause damage.
We're also using llms to sort of
help us understand sometimes not
as a replacement for a Dialogflow
or something like that,
but also as a way to enhance or
improve our understanding of,
you know, what users are saying.
And so the ways we're using it
are like, not very risky.
Um, in a year from now,
two years from now, when everything
goes LLM, then we'll have maybe
we have beyond again and we'll
talk about some of the ways that
we mitigate risk using an LLM,
an open conversation with our users.
But for now, a lot of the use of BLM
and AI is being done in ways that are
quirky and fun and are ways that
help them maintain their cognitive
abilities, or that improve their
cognitive abilities and maintain
their independence. Makes sense.
Um, and there was one little question
I had about just the relationship
building aspect of like, because, uh,
out of all of the chatbots for like,
especially maintaining a relationship
long term, maintaining context,
knowing the person, remembering
all the details is so important.
And it's something you cannot
you cannot afford to skip or
just ignore altogether.
Uh, it's a struggle, I know, because
even with customer support chatbots,
we still worry about content
context retention. But it's not.
It doesn't damage the relationship
as much as it would if, like,
you didn't remember things about
what we or about the person
she's taking care of. Right.
So how how do you do this?
What are the challenges you've had?
What are some of the wins you had?
I would love to know,
how you made that work.
Um, in the term, in terms of like,
long term relationship building,
how do you remember things?
How does context retention work?
It's a great question.
Um, and it's still like we're at
the kind of the cutting edge of
that in terms of what we're doing,
because there's no other agent,
like you said, that lives with
people 24 over seven in their homes
and needs to form a relationship.
And part of forming relationship is
remembering things about each other.
Not in a creepy way,
not in a way that, like, I'm showing
off that I remember no, but in ways
that are contextual and it like
further in advance our relationship
and our ability to coexist.
And so in the past,
before we had LMS,
it was a big challenge, obviously.
How do you, you know, and you do
it in the old fashioned way and
you would remember things and like
different facts about a person,
like what your favorite hobby is.
And they would tell us and hopefully
we got it right with the dialogue
flow and that we would remember.
And hey, have you done any
pickleball this week?
Because I remember pickleball is
your hobby.
So you had all these, you know,
slot filling ideas that used to
kind of like give us the
illusion that we're remembering.
But now we have llms and we have
the ability to create rag and to
create these sort of retrieve
memories from conversations and
pluck out things and remember stuff.
But then again, you're dealing
with LM, which hallucinates.
So there's a lot of issues there
and you don't.
Last thing you want to do is like,
oh, Howard, let's talk about your,
you know, your daughter who is
going through a surgery.
She's like,
I don't even have a daughter.
What are you talking, daughter?
What are you talking about?
So that was a that's a
relationship killer.
You don't want to bring up false
facts, or you don't want to bring
up things that are not true.
So there's a balance here of how you,
you know, and we're still there's
still I don't think there is a
solution yet that is like so we're
experimenting with things not with
our customers, but in sort of
playgrounds and with dev environments
on how this type of thing could work,
while still maintaining some of
the old architecture that was put
in place in 2016, 2017 of slot
filling and things like that.
So there's a bit a mix of both at the
moment, but we're really, really,
really careful about using the LM
stuff and we're really careful about,
um, how we use that anyway.
So if they hadn't open because
there are some situations where
we have an open conversation and
there are things that occur in
an open conversation,
and then so there's things that we
remember about the about the user.
And the question is, how do we
use that information in to inform
and to improve this relationship.
So things that we're working on,
if you ask me, in a year from now,
if I'm still hearing, you know,
I'll have much more interesting
insights into that specific because
the something that we're working
on right now. Yeah. Wonderful.
Um, just a few questions about,
you know, your general state of
conversational AI, because,
again, you've been in this space
for quite some time.
Is there something, um, that any
belief about conversational AI that
you've had to change recently?
Well, that's a good question.
Um, is there a belief about commerce
that has to change recently?
Recently?
Well, I thought that in 2022
that we were done.
I was like, oh yeah,
I'm out of a job.
And in fact, every day I wake up
and think the same thing, but we're
still here and I'm still doing it,
and I'm still doing it in ways
that I never imagined possible.
Meaning, um, not only have I not
lost my job, not maybe when I
get out of here I will.
But the the my toolset,
my toolkit is. Way. Way better now.
Like in like I said,
I'm able to create.
They can do poems together,
they can create art together.
They can have a they have an animated
plant that helps them fix like I
whatever I dream up these days, like
there's an LLM based thing that can
help me create that really quickly.
Like, um, I don't know,
I can come up with like, you know,
I had this thing where I was like,
I found all these online
resources for free tours,
Like images from the national parks
of America from like the Vatican
offers like an online, you can
just download images from a tour.
They don't have the information.
It was all in Italian.
So I went to ChatGPT.
I was like, help me come up with
a tour for our older adults of
the Vatican with 15 steps.
Here are the images that I have
and I'd never been there before.
I'm ashamed to say I had never
traveled there, and it created it for
me, and I was able to create this
sort of I mean, I had to do some.
There was a lot of work that I had
to do manually to sort of adjust
everything, but it gave me a lot
of the I double checked the
information with some sources that
were tour guides to make sure that
it wasn't hallucinating and lying.
And then now our customers can
go to the Vatican,
which is amazing for these people,
because a lot of them have never
been out of the country,
let alone to like the Vatican,
where they could actually, you know,
um, and so that's changed my
perception of what it means to be a
conversational AI. We're different.
Our roles are slightly different.
We need to be better.
Um,
but we can create amazing things now.
Way better than we used to be
able to. That is awesome.
And is there something that you feel
is true about conversational AI
that very few people understand?
Oh. No. No, it's a tough one. Um.
Maybe it's something you've not
seen that much in candidates
you've tried to hire.
Maybe in that direction. Um, no.
But I do have an interesting
anecdote.
I don't know if it answers your
question, and it's something
that I wanted to bring this up.
And I think this may be a good
opportunity to bring it up,
because when I was, you know,
working a lot with Hans,
who's the the CBI guy. Right, Hans.
And we, he always used to say
like find the best CS person,
customer success person and they will
make a great conversation designer.
And for years I thought he was,
you know, I thought it was crazy.
It's like a CS person.
I person can't do my job.
And then recently we did that.
We have we had a my colleague on
paternity leave.
We needed to find someone to
replace her.
And we looked at our CS team and we
found someone in CS who was sort of
who deals a lot with our customers,
knows the customers really well
because she talks to them all the
time, very young and never done
any kind of conversation design.
Didn't even really know what
conversation design was,
but had a unique understanding of
the landscape of the product and
the understanding of the customers.
And with a bit of help with a
bit of the CDI technique,
I was able to help her become.
And she's she works with me now
and she's amazing.
And it's a great example that
anyone can really be a
conversation designer.
But the best ones that really the
best ones are ones that understand
who the agent is talking to,
who the customers are,
who your target audience is.
If they understand that, then it's
much easier to bring them back and to
explain the rules of everything then
to try and explain to them, okay.
These are the people that we're
trying to.
Um, so in that sense, I think that
it's easier than ever to become a
conversation designer these days.
The tools are amazing and you
can create amazing things.
I would say if you are interested in
becoming a conversation starter,
obviously go to CDI.
Take the courses.
But one of the things that I tell
everyone to do is to create.
Find an interesting use case
that you think will help.
I've talked about this in this video
that I posted on LinkedIn yesterday
that, um, my mom was just recently
released from the hospital discharge
from the hospital, and she got a
list of medications and she to take.
And it was so overwhelming.
And I was like, well,
what if an agent or an AI could help
us manage this long laundry list
of medications that she now needs?
Um, and luckily,
I was there to help my parents.
But, you know, not all parents have
kids there that can, you know,
that can help them manage this thing.
So find a use case, try and solve
it using these amazing tools that
you have and if you fail trying
to solve it, you'll at least have
gained all the information that
you need to become a valuable
member of a of a conversation
design team somewhere. Absolutely.
And is there a skill set that is
super tough to find?
Maybe a combination of skills,
especially maybe while you're hiding,
maybe in the profiles that
you've seen out there.
That's just tough to find,
but very important for a for a
conversation. I think.
Wow, it's tough because it's sort
of like both creativity but also
understanding of the business world.
So you have to be a sort of an
interloper,
a creative person in the tech world.
And that's not an easy thing.
Sometimes it can be very
challenging for people who
consider themselves very creative
to survive in a place like this.
This is why I go around putting
mustaches and eyebrows on pictures,
you know, just sort of keep
things kooky and fun and,
you know, and weird and, and so
with the hardest skill set are,
are really people that are really
creative and are willing to, for one
reason or another, sort of survive
in this corporate environment.
And if they're willing to do that,
someone who's super creative and
willing to survive in a
corporate environment,
that's someone that you know.
Um, because I don't think there's
anything in conversation design that,
you know, you can't teach someone.
There's nothing innate about it
that's like, oh, that person,
you can't do it.
You know, I think everyone can do it.
Um, yeah. Wonderful.
And that brings me to the end of
this lovely, lovely interview.
I feel like I learned so much.
I have so much written down that
I want to go and then look up.
But thank you so much, Jason.
Appreciate your time,
appreciate your expertise and your
willingness to share it with us.
Uh, yeah. Any final thoughts?
Just overall about, um, you know,
conversational AI and design,
anything that you want to end on?
I do actually two things. One.
One of them is for the company,
and one, I'll say that we're now
offering a seven day, uh, trial of
lark, but, uh, we call it WebRTC.
It's online, so you can sort of
experience what it's like.
So if you have if you're in the
States and you have an older, uh,
like a mom or dad or an uncle that
you think would be a good candidate
for LEC, just comment on the video
or send me a note and say, hey,
I, you know, um, because we can
get you that free trial for that.
And the second thing is I check
out my YouTubes.
I got all these YouTubes that I
made that are weird and quirky
and really embarrassed my teenage
son and make one a cringe.
But I put a lot of work into him,
and I enjoy him, and they're usually
about like, AI and funny stuff.
So check out my YouTube.
Absolutely absolutely highly,
highly recommend that for anybody
who's looking to break into this
space and does not want to deal
with stuffy reams of content.
So thank you so much.
And, uh, that's it.
I hope everyone will join us in the
next episode. See you. Thank you.