Trust and Transparency in AI Agents
Key Points
- “AI in Action” is a new IBM series that dives into what generative AI can and can’t do, how it’s built responsibly, and how it solves real‑world business problems.
- Trust and transparency are the foundation of virtual customer‑service agents; users must be told when they’re talking to AI rather than a human.
- Effective virtual agents need to be empathetic, personalized, and timely while safeguarding the accuracy and security of the data they handle.
- Customers feel deceived when a chatbot pretends to be human and then fails abruptly, so honest disclosure of AI assistance improves user acceptance.
- To be useful, virtual agents must provide specific, context‑aware answers and offer smooth handoffs to human agents whenever the issue exceeds their capabilities.
Sections
Full Transcript
# Trust and Transparency in AI Agents **Source:** [https://www.youtube.com/watch?v=1FR5xIOknY4](https://www.youtube.com/watch?v=1FR5xIOknY4) **Duration:** 00:18:30 ## Summary - “AI in Action” is a new IBM series that dives into what generative AI can and can’t do, how it’s built responsibly, and how it solves real‑world business problems. - Trust and transparency are the foundation of virtual customer‑service agents; users must be told when they’re talking to AI rather than a human. - Effective virtual agents need to be empathetic, personalized, and timely while safeguarding the accuracy and security of the data they handle. - Customers feel deceived when a chatbot pretends to be human and then fails abruptly, so honest disclosure of AI assistance improves user acceptance. - To be useful, virtual agents must provide specific, context‑aware answers and offer smooth handoffs to human agents whenever the issue exceeds their capabilities. ## Sections - [00:00:00](https://www.youtube.com/watch?v=1FR5xIOknY4&t=0s) **Untitled Section** - ## Full Transcript
representative that's me into a phone
trying to talk to a human when I have a
customer service issue but instead I'm
only getting canned responses so how do
we get a human on the line when we
really need one or do we need one we're
going to get to the bottom of it today
on AI in action in this series we
explore what generative AI can and can't
do how it actually gets built
responsible ways to put it into practice
and the real world business problems and
solutions will encounter along the way
so welcome to AI in action brought to
you by IBM I'm Albert Lawrence so let's
talk about virtual customer service
agents and how to actually get help
today I'm joined by jeie Walters
customer experience speaker trainer and
podcast host jeie thanks for being here
thank you so much for having me we're
gonna have a good time my other guest is
Morgan Carroll senior AI engineer at IBM
what's up Morgan I am jazzed to be here
oh I know you are I feel it now look I
know you both are wondering why did I
Choose You two for today's episode well
it's because you both specialize in the
future of customer service and virtual
agent experiences big picture I want to
explore trust is it time for us to give
virtual agents a chance can they really
be empathetic personal timely and answer
the questions that we actually have I
know we're asking for a lot but jeie
what do you think I think it's time
customers are expecting this they want
to be seen and recognized for who they
are and where they are on the journey
however they interact with Our Brands
and so now we have a chance to do that
in a way that is automated and personal
and all of those things but we have to
consider the entire Journey at the same
time I think it all comes back to trust
we want to make sure that users know
that the data that they're giving and
receiving is accurate and secure if they
are interacting with an artificial
intelligence agent versus a human we
should let them know that so that they
don't think like okay this is aive human
no this is artificial intelligence that
is assisting us now Morgan you mentioned
that it's important for humans to
understand when they're speaking with a
virtual assistant versus another human
jeie why might that be the case can you
give me an example we have heard over
and over that customers really don't
like it and it's not just customers it's
all of us right we don't like to be
deceived as humans and so it can feel
very deceptive when we think we're
having this very human conversation and
then suddenly it basically says be Boop
I can't do it and we realize that it's
actually not a human that is breaking
trust from the beginning instead of just
being open and honest about the fact
that you know you're you're talking to a
virtual agent who's here to help you
people are very open to that well when I
am talking with a virtual agent then I
guess one of my concerns is that I want
to make sure that the answers that I'm
given are not just broad generic answers
I would prefer for them to actually be
answers that pertain specifically to my
my questions that address my needs how
do you bridge that Gap with generative
AI so we're going to connect to
different data sources generative AI is
going to take all of those different
sources and sort of make an experience
that is unique to the user without the
user having to actually you know press
zero go to a live agent have the live
agent look at their information but
generative AI can do that I think one of
the powers of gen AI that we're seeing
is that for instance in a contact center
if somebody calls in suddenly that agent
has the information just like you're
talking about Morgan at their fingertips
There Are Places still today where
agents have to pull out big binders and
try to find the information based on the
information that they happen to get on
the phone call and we're really seeing
how we're connecting all that data in
the back end to make it easier for the
people who serve the customers and
customers really are responding to that
very positively um you just painted this
picture of like humans pulling binders
and that feels kind of archaic but I
guess it's still happening it's still
happening because don't all have the
capability to put the data in one place
and that's what's so powerful about what
gen can do it can connect some of these
different systems on the back end so we
can actually provide a better front end
for the customer okay so I want to get
into some of that backend discussion
then to now Morgan I'm very curious how
do you end up really building that
personalized generative AI solution then
can you take us through that system yeah
so we want to keep the customer in mind
first and foremost put ourselves in
their shoes and think okay okay how does
this conver how should this conversation
go that's going to be efficient and get
straight to what they need so the first
thing we do is sort of build like an
outline of a dialogue just like a
skeleton like okay this is where we want
to start maybe and this is where we want
to end and then we're going to come in
and add all these different Integrations
so maybe a database that has all the
customer information so when they call
in it's like hi Morgan and you know it's
already got all my nobody has to pull
out a binder and like search through all
this stuff and it just makes it very
simple so now that we've had a chance to
talk some about the technical and
engineering aspect and also really the
front faceing customer service aspect
how are these two things going to come
together yeah so we want to think about
the customer experience overall from
beginning to end let's say we have
someone who they need to get in touch
with HR because they're about to go on
maternity leave well HR only has five
different options to select from and
this doesn't fall into any of those
categories what do they do obviously
we're z00 representative we want to talk
to a live person so these are the that
we have to consider and this is
something that generative AI can help us
with okay yeah and I would just say
think about the overall kind of strategy
and needs of both the brand and
organization and the customer that
you're serving so for example when when
email came out we got very excited about
email and we sent all these emails and
now we're kind of like we're done with
email right and so I think we have to
think strategically about how to best
serve people and when we're talking
about the customer Journey customers
want choice you and I might not want to
go through the same Journey you might
have different needs than I do as far as
if I'm in the car I might want to call
and talk to a person when that is
happening so providing those options for
people even though it might be more
efficient on our end to use the virtual
agent can we give those choices along
the way to the customer as well
different customers might have different
preferences regarding how they would
like to be helped uh what are some of
those preferences like well I mean think
about the different circumstances that
we go through in our daily lives I kind
of get a chuckle when people say things
like well our Digital customer journey
is this and you know our in-person
journey is this because sometimes we go
into a store sometimes we want to shop
digitally sometimes we might prefer a
virtual agent in a chat and sometimes we
might need to talk to somebody and call
them right away and we don't necessarily
fit into the boxes that they give us so
that customer choice of how to be served
and helped is just super super important
so I'd like to Pivot to the agent
experience they're obviously inundated
with a number of emails and there's
plenty that virtual assistants can
automatically respond to but if they
have perhaps something more difficult um
they're getting a lot of these emails
per day and they have to respond to
these generative AI can actually assist
them it's not going to necessarily
respond for them it could but in this
case we're thinking about how can we
make the customer service agents job
easier in gener of AI can help them
draft an email or look up information
stuff like that so even if a customer
isn't speaking directly to a virtual
agent generative AI is still playing a
role with the human- to human contact
exactly but then it makes me think about
this idea of being a proactive versus
reactive why is that important to the
entire process and how might a customer
feel and experience that on the front
end well when you get right down to it
if somebody is contact customer service
something has probably gone wrong
already so yes we want those moments to
be efficient and effective but we also
want to prevent those moments from
happening because it will reduce cost
for the brand and everybody else but it
also makes the customer field more
valued and cared for so if we can
predict for instance Based on data and
AI can look for patterns really well so
maybe AI notices you know a lot of
customers are returning this product
after 3 months what's going on
and then we can figure out well it's
breaking after 3 months that's not great
so let's figure out what is actually
going on and proactively prevent that
from happening or proactively reach out
to the customer and say we think this
might happen in 3 months here's a
proactive solution for you that makes
people feel really kind of recognized
and seen in a way that when they have to
call in when the onus is on the customer
to really solve their problem they don't
feel valued as much well Morgan I know
we've been talking about the customer
service experience but how exactly is an
engineer helping to transform and to
build the customer service experience
okay so this is exactly what I do every
day um think back back in the day when
you would call a customer service line
and you only have you know your five
options whatever you have to push one
two three and that's all you can do now
we can actually use generative AI to
take what the user is saying so actually
I'm going to break it down on a sort of
like a technical level we have our um
our dialogue flow here right and then we
have the user saying like I need help
with this particular thing well if it's
not sort of hardcoded into our answers
we can actually go out call out to a
large language model with some of that
information and say like hey we need you
to provide an answer to this user and
just it's directly connected to the
virtual assistant so we get that data
back and then we just pass it right
along to the user and that works most of
the
time this is why we have guardrails
built into our virtual assistants
because obviously all of this is really
really new and in my experience you know
in building these assistants sometimes
we do get a wrong answer and we want to
make sure that doesn't happen so we have
various tools that um basically if a
customer asks a question and we're not
really really sure that it's correct
then we may just say something like I'm
sorry I can't help with that and that's
where we need a life person well and
this is why it all comes back to
transparency and building that trust so
that we're not presenting something that
isn't quite true truthful to the
customer they know the entire time that
they're dealing with a virtual agent and
that you know what sometimes there are
places where we need to introduce a
human jumping back into the mindset of
being a customer right and if I call on
into a virtual agent I'm wondering what
are some of the common questions that I
probably will be asking as a customer
that a virtual agent can actually
understand and give me valuable
information about well I think right now
this is probably the best use of virtual
agents are these kind of repetitive very
standard questions that customers ask so
think about when you're at a hotel some
of the questions you might have for the
front desk where are the pool towels how
do I get a wakeup call what do I do
about you know these automatic trpes
aren't working or whatever and once you
start seeing those patterns that's
something that a virtual system could
handle very very easily a virtual chat
that something that maybe is accessible
right in the moment then on the flip
side of that these big nuance
complicated things that we run into as
humans sometimes we need humans and we
need both that empathy that a human can
give and also just understanding you
know what this isn't the standard
question this is about something that
went wrong that doesn't go wrong a lot
we have to figure this out the customer
is very upset so once we get to that
place then we're going to start seeing
new patterns and we're going to start
realizing different ways that we can
maybe update customers you know if
they've already talked to a human and
said this is going wrong maybe there's
an automation through the virtual agent
that says hey I'm your virtual agent I
just want you to know we're working on
this here's what to do next here's when
you can expect to hear from us because
that's another thing customers want is
just that proactive communication of we
heard you and we're working on it well I
heard you say automatic drapes and now I
want for my next hotel room to have
those um but but I'm also thinking you
know you gave me some great ideas in
terms of in the future what are some
ways that a virtual agent might be able
to help and what are some of the common
questions that customer might be calling
on in from Morgan do you happen to have
any thoughts in terms of some questions
that you are anticipating for customers
to be calling in for definitely what are
your hours what time do you open um
what's your phone number what is your
address things like that just simple
again repetitive tasks but I actually
have a good example of one that we do
need a human for um and this happened to
me yesterday I got to my hotel and there
was no electricity so the hotel actually
does they do have a virtual assistant in
their app but but obviously that's not
really going to be useful if there's no
electricity so go to the front and
that's a really really simple example
but these things happen in the future
and now you've got me stuck on this
hotel situation both of you um do you
think that we'll be able to speak to a
virtual assistant and just say open up
the curtains that's happening in some
hotels already they have Smart speakers
they have tablets in the room they have
all sorts of ways that we can interact
and I think the other side of this that
maybe we have have to start thinking
about as well is how do we take care of
those humans now who are dealing with
these big complex issues all the time
because in some ways these repetitive
things they don't take as much empathy
they don't take as much brain power and
compassion so they're kind of like a
little breather between these big
complex things I think we really have to
think about the world where we get to
where we're asking humans to just carry
a big heavy lift for every single thing
that they do and look for ways to take
care of them well let's stay on this
train of thought is we're think about
this world where humans don't need to do
that why can't we do that right now
what's holding businesses back from us
getting into Genie's ideal future we're
kind of already there um we have a
concept Called Agent assist so our
agents you know again they we don't want
them to have to search through the
binder they actually may have their own
virtual assistant to help them with
these complex customer inquiries so you
know customer calls in hotel uh my
drapes aren't working and the TV is
broken and there's no water well I mean
a virtual assistant probably could
handle that but I think we're going to
want a live person so they're like all
right let's look up the information we
got to call the water company and
electrician you know etc etc and they
have this virtual agent that can help
and hopefully just put me in a new room
yeah because at that point the very
least wow okay so then that raises
another concern for me which is data
privacy and sharing so how do you
reconcile that right how does data
privacy and sharing end up impacting the
personalization that can also make a
customer feel really valued right so
obviously we don't want someone stealing
our credit card number but it all comes
down to do we trust the model that we're
using we have to make sure that we're
open and transparent how is this model
being trained where is the data where is
my data going you know is it secure so
let's think of um a client we have
client information over here and then we
have our virtual assistant up here in
the cloud we we're not actually storing
the data in the Cloud it is secure in
the customer's environment we're only
sending and receiving the data that we
need in order to have the conversation
Flow versus just here's my credit card
number let's just send everything over I
think we're at the cusp of a really
interesting time around this whole
conversation around data privacy and
protection because for years we've been
talking about the fact that you can have
a more personalized experience the more
data that a brand knows about you that's
why we get suggested purchases that's
why we are told when hey that product
ordered might be running out you might
want to order it again that's all data
driven but I think now consumers and
customers in general are becoming savier
about what that means so I want to give
them more control I want to see a world
where we can have a dashboard where
people can say yes you can use my data
this way this way and this way and it's
all integrated into their experience
from the beginning and they have more
control and like visibility into how
it's being used I like this like I want
to sign up for that like right for that
future but I I got to imagine imagine
though that there must be some sort of
like difficulties with scaling to that
like are there any difficulties within
that and if so what does that look like
let's say you have a your virtual
assistant and it's running on a server
you may maybe let's say for Simplicity
sake one copy what if you have let's say
20 people trying to interact at the same
time well since it's running in the
cloud it can actually handle multiple
transactions if we need you know say
like suddenly there's an outage like
it's an electricity company we've got
500 customers calling since it's in the
cloud it can automatically scale so we
have more instances of this assistant to
handle more people oh well that's
fantastic but you know I have to ask
what if it goes wrong what can go wrong
oh so many things um tornadoes being
from Texas what if a tornado comes
through and takes out our data center
well that's why we have multiple data
centers we're going to have all of our
data backed up redundancy in various
places so what I'm getting now is you
both shared so much today but for my
brain I'm trying to process them so I've
got a few key takeaways that I'm going
to take a stab at okay um keep customers
in mind as you start your build because
this is customer service after all it's
important for humans to feel valued and
seen in their requests even if it's not
another human that you're speaking with
on the other end and we don't like to be
deceived so be transparent with your
customers if it's a virtual agent it's
okay just let me know I'm talking with
the virtual agent is that good that's
awesome awesome I love positive
validation
um jeie Morgan thank you so much for
being here today you've got rain cells
all percolating right now um this is a
fantastic conversation that's it for
this episode so thank you all so much
for listening thank you for watching but
worry not there are a ton more great
things where this came from and we will
see you here again soon
[Music]