This Is How AI Will Revolutionize The Customer Service Industry

foreign
continues to disrupt many Industries but
today we're specifically discussing the
customer service industry I'm Rose Marie
Miller here with Dion Nicholas the CEO
and co-founder of forethought thank you
so much for joining me today Dion
Rosemarie thanks for having me I'm glad
to be here absolutely so could you just
tell us what is forethought yeah so
forethought is the generative AI for
customer support automation we
effectively fine-tune large language
models like GPT and and our own
proprietary models on a business's
specific customer service data so there
are service inquiries their history
their policies and give them this AI
Suite uh that helps automate customer
service that helps augment their agents
and discover insights and do a whole lot
of things on top so it's kind of like
having your own personal chatbt that can
answer questions for your business for
customer service
sometimes I'll go on a website and a
little pop-up pops up is that kind of
yeah exactly that's one of our products
and so we're an entire Suite but our
most popular one uh solve is you can use
it as a little widget so if you've um
you've probably interacted with it in
some websites uh but
um most people so prior to forethought
you've probably interacted with a chat
bot and when I ask people what was their
experience like most often people say it
sucked right because
um chat Bots classically
pre-pre-for-thought pre-gpt pre
everything we're trying to do they've
existed for a while but they were always
manual they were always keyword based
decision tree based you had somebody
hard coding you know kind of like you
know press one for for uh for billing
press two for refunds but in chat bot
form and that was really the state of
the art and so nowadays with a with a
forethought AI it's it's conversational
it actually understands you it's able to
solve problems go back
um say issue refunds or look at Borders
so it's a completely different Paradigm
to
um how chatbots actually operate so
where is forethought getting its its
knowledge for its AI if it's able to
understand me yeah so it's a it's kind
of a multi-layer approach so we start
with foundational models so imagine a
GPT uh chat GPT gpt4 we actually
launched
um our product support GPT back in March
uh which was uh super exciting
um so we start with the foundational
models and these are trained to
understand language and you know all the
all the power that you can imagine from
a language model
um and then second we we fine-tune or we
train on the businesses specific data so
for example if you ask Chachi PT today
hey I just ordered my I don't know
instacart what's instacart's refund
policy or something like that right
um it's not going to know the answer to
that it knows broad internet stuff but
it doesn't know the specific policy of
the business so the biggest problem with
something like Chachi BT is it's just
going to hallucinate it's going to make
up an answer and that's cool when you're
using chat gbt to write blog posts and
stuff because you want it to be creative
but it's not cool in healthcare or in
finances or in customer service where
you need it to give you a correct answer
and so what we do is we take these
foundational models and then we also
have an entire system that trains on the
policy so we integrate into their
systems our system will look at past
conversation history how did agents
respond to this question what are the
main kinds of questions that get ads uh
get asked like you know for some
businesses like e-commerce it's where's
my order it's refund but if you are
um you know a SAS product like an Asana
then maybe it's how do I leverage this
to-do list feature right and so all
these different kinds of questions and
all the ways that agents can answer them
the AI is learning and kind of
fine-tuning on that so it can start to
mimic the agents and give these correct
answers and so you can take these
Baseline foundational models fine tune
and a few other things but basically
constrain them to a business's specific
policies and you get something really
powerful that can actually answer
questions on their behalf so how's
business doing what's your customer base
like oh um it's great so um you know
some of the businesses I just mentioned
um are actual customers of forethought
we work with you know some of the
organizations that are changing the
world including folks like instacart
including folks like upwork
um who are really on the Forefront of
being customer-centric businesses and as
this new wave of generative AI is coming
about they're they're working with us to
to launch and this amazing customer
experience so we're super excited
um and we're growing we've been around
since uh 2018 we launched
um and you know before AI was cool and
we've just been growing ever since you
know it's kind of an entrepreneurial
journey and uh I'm really proud of kind
of where we've come so far and we still
got a lot more to do yeah so 2018 that's
about five years ago yeah now it's a big
deal how are you guys kind of leveraging
the hype oh it's so interesting it's
fascinating how everything is just
shifting and changing so five years ago
when we launched forethought the the
cool thing and the thing I'm most proud
of is that we actually came about it in
this AI First fashion what we talked
about earlier with with respect to most
chat Bots and there's thousands of them
all of them have all been these decision
tree kind of keyword based things and so
they were they were conversational but
weren't really AI right they weren't
intelligent a little bit more artificial
than intelligent is what I like to say
and so when I started forethought and I
have a background in engineering as well
as AI research and I've always been
fascinated with can AI understand
language and can you use that to build
better products right and so that was
actually the reason I started for
Thought part of the the the founding
story there and so generative AI even
though it's really hyped now technically
it goes back years right language models
were invented probably in the 80s or
something I may be wrong about that
specific number but a long time ago and
then what happened was in 2017 in 2018
and pretty much every year since then
there's been a small but meaningful
breakthrough in in how AI can do
language right so
um Stanford launched a question answer
data set in 2017 which was uh the first
time people were starting to compete in
research and build models on how AI can
answer questions uh 2018 Google launched
Bert there was a whole paper on the
Transformers and you know it doesn't
matter the architecture specifically but
each of these were Ground breaking
breakthroughs the T in GPT is is a
Transformer uh generative pre-trained
Transformer so it's actually that that
architecture that led to all of this
right and so we started seeing all this
in 2017 we started seeing it in 2018 GPT
was launched we've been leveraging gpt2
since I think 2019 and so the reason I
say all this is because while there's
this point in time and I think really
chat GPT in November of 2022 is really
the the impetus for this generative AI
boom that that has captured the uh the
attention of the masses of hundreds of
millions of people when you actually
peel back the onion a lot of these
groundbreaking things have been
happening for years and we've
um we've been leveraging a lot of it on
the research side and I think the
biggest change right now going back to
your question with respect to the hype
has been
instead of us banging down doors
explaining to people this is why you
need real AI instead of what you've been
doing before we're now getting people
banging down our door saying hey we need
AI can you help us my CEO wanted this
yesterday and we're like okay I'm glad
so I think it's it's been fun to see the
awareness of AI kick in
um and it's forced us to get real sharp
it's forced us to continue to
um upgrade and focus on our core
differentiators
um but it's also just been such a a fun
time to be an AI company like this is
the this is literally the best time in
history to be involved in AI right and
uh and I don't think it's letting up oh
that's scary you don't think it's
letting up meanwhile I'm sure you saw
the center for AI safety they put out a
statement a few days ago about how AI
could lead to like human extinction or
something like that what are your
thoughts on that so
I I am a personality trait I will just
start by disclaimer I am an optimist a
cautious Optimist in general
um when it comes to these things and so
I'm not actually a huge fan of most kind
of alarmist type things I think I don't
know what the agendas are behind it like
I don't know
um but I'm also not the expert on that
and so the way I think about it is there
have been waves in the past that have
been as powerful as this shift to Ai and
so the only way I can reason about the
future is by trying to find analogies in
the past right and so you know when
um the printing press electricity right
computers
um even nuclear energy right I think
that's probably the the biggest thing to
think about nuclear energy is both
positive and negative in the sense that
you know it has the the potential to
power all of society right like in terms
of our energy and all that but there's
also the the the Dark Side of nuclear
energy which is like nuclear weapons and
stuff like that right I don't even know
if we're allowed to talk about that
stuff but but you're right but um that
is I think the same with AI uh maybe
potentially even to a lesser degree
depending on how you think about it but
AI has this power to radically improve
productivity across every single area of
Our Lives from you know how we get work
done to how we interact with our friends
our family our people
um and I think that's just going to be
net positive heck AI will even help us
you know and figure out the theorems
that will help us figure out nuclear
fusion and fission and like have better
nuclear energy and all these other
things right so I think there's just a
whole
superpower for um Humanity through AI
but at the same time there are real
risks right like we've seen that with
social media disinformation
misinformation now what happens when
basically anything on the internet can
be doctored right like who knows if this
video of us is even real or AI generated
at the at the end of the day and so but
so there are real dangers right and you
know jobs will be changed and things
like that and so I think it's very
important that we think about AI safety
right and so even though I'm not kind of
an alarmist like I'm actually grateful
that there are people saying hey should
we be stopping this should right because
that is what's going to prompt enough
people who are working on the safety
side
um to take these things seriously
um
and
one more kind of like you know
generative AI history lesson the the
earliest generative models
um were called Gans generative
adversarial networks and the way they
were trained was you would train one
model to produce something like
Shakespeare or an image and at the same
time you train another model that's
trying to detect how fake uh trying to
detect if it's an AI fake like AI
generated or not and so what you do is
you'd pick these models against each
other
um and so as this model got better at
producing human-like images for example
this model got better at being like nah
that's a fake and this is what a real
human-like image looks like and they
would keep getting better and better and
better until this model was like really
good and could like outsmart the the
machine
um literally that's how the first
generative models were trained the
reason I I say that is because you know
in the past what they would do is they
would train these models they would
throw out the the Tester the score and
just use the generative model
um but I think in the future when you
when it comes to AI safety not only will
we have all of this AI that can produce
the stuff but we should also be using AI
to help detect is this content fake on
the internet and so I I think a lot of
the AI safety will be you know making
sure that the the stuff that is powerful
in generating and the stuff that's
powerful in safety evolve at the same
Pace in society and there's no there's
no way to understate that and so that's
kind of how I think about it I think
it's very important and I think it's you
got to equally level up our you know how
we think about copyright law how we
think about laws in general jobs all of
those things as the AI advances so while
I am not an expert on this I I believe
some of the conversation I've been
hearing is it's going to get too
intelligent and if it's too intelligent
then it's going to be like I don't know
over more intelligent than people yeah
and so why if AI is so brilliant and I'm
kind of using your model yeah if it's so
brilliant why would it like bring it
self down right right right right so
um
a couple things I think that's actually
it's a good question
um
well okay let's start with AI that's
about as smart as human so before we get
to you know artificial super
intelligence let's start with just
artificial general intelligence okay now
um I again by analogy I like to think
about okay what would the world be like
if you're trying to take it from first
principles what would the world be like
if we all had access to that and there
was AI that was about as smart as humans
well first of all we have a billion
examples of those humans right and and
so what it would almost feel like is
what if you have the world's smartest
human let's say Albert Einstein I don't
know if he was the smartest of all time
but like let's say everyone had access
to their own personal Albert Einstein
right and I think that's actually what's
more likely to happen is you would be
able to again like prove theories or you
know just have this Albert Einstein type
person to help you but there are still
things that are going to be hard for
Albert even if there's millions of
copies of them um like he still hasn't
figured out quantum physics right like
he still hadn't figured out and he spent
his whole life and so just because you
have ai that's like equivalently smart
or even slightly smarter than than
average humans there are still going to
be things that are hard for this AI to
do
um and things that we'll have to figure
out an example another example is like
cryptography right
um people immediately think that if you
have this AI that it's going to be able
to hack the nuclear codes and and do all
of this stuff but the the problem is
those are encrypted and it's actually
really really hard to break encryption
you just but you kind of have this
imagination that like if an AI gets
really smart it's going to be able to
figure it out it's like no it's a
mathematically impossible problem
um to solve and eventually you know our
current encryption will get broken but
then you're going to have Quantum
encryption Quantum Computing and Quantum
Computing algorithms also break all of
our current encryption but again you can
also use quantum encryption to protect
against Quantum Computing right so
you're going to need algorithms that are
more kind of robust that kind of
technology and I think the same happens
with AI and all that so all that to be
said I think there are like
for for the AI to want to overthrow us
it would have to be a as smart as us if
not smarter right which we think is
probably coming it would have to have
access right as in access to all the
internet it would have to be able to
somehow break into things that
um you couldn't mathematically break
into even if you're a super intelligent
Ai and it would have to have motive
that's that's the last piece
um we again because of sci-fi we just
assume that if you have something that's
really smart it's going to want to take
over the world
um I don't think that's true for every
human
right
um who are reasonably smart creatures
intelligent beings
um it's just as likely that you know if
built into the system this AI is just
going to want to you know perform its
tasks and things like that it doesn't
necessarily it could happen but it
doesn't necessarily imply and I think in
most cases wouldn't imply let's say
enslave Humanity or something right like
I think that's just like our imagination
is running well all that being said
though I do think like it is worth
having these conversations I don't like
to think about things in terms of an
alarmist way because from first
principles it's like those extremes are
very rare to happen but if you don't
think about them you won't mitigate
against them and in our case mitigating
might be creating new AI encryption
algorithms right that can now protect
against AI hackers and all of that and
so you gotta you gotta build it out I
don't know never heard that perspective
before just because it's more than us
doesn't mean it wants to take it yeah
like they're not the same thing right
um but I hope I'm right so yeah
like you know we gotta we gotta have the
conversations because it's important so
you said in the future you think it will
be smarter than humans that's likely to
happen when six months a year oh that's
the question anybody who claims to know
is lying right but
um okay I have kind of two contradictory
thoughts ironically on this one
contradictory thought or possibility is
that we already have all the ingredients
for AGI artificial general intelligence
today
um right and because like for example
Chachi it's not AGI but it's so much
more powerful than most people imagined
even a year ago
I bet even the open AI team didn't
expect it to work this well and so
um and it's really again based on a lot
of these architectural changes like
Transformers and some of the
reinforcement learning stuff but
um it's really powerful and it's almost
like they they found the right
configuration
um in the right training data they
figured out how to scale it up that
created this cool thing that is now gpt4
right and so there may be an argument
again just by analogy or or just
thinking it through that
we already have technology
some through some combination of which
is probably as smart as the human brain
or greater we just haven't put it
together yet
um in which case yeah in some degree AJ
is already here right
um
one contradictory thought but the
opposite though is like
um if you then ignoring that just follow
the progression of of AI and AGI
um and recognizing that okay chat TPT
really is an AGI but it's like really
cool
um I don't know five ten years I would
say like I don't think it's going to be
the next six months to a year I think
the GPT model will get so much better
and it'll be able to do what it can but
it at the end of the day it's just a
language model it's just trained to like
produce things that kind of can trick
trick the human into thinking it's it's
real
um
but you know enough advances will
probably get there five ten years but we
could be off it could be 50.
um I I don't I don't think so like I
think you know we're almost there
so
as the CEO of an AI company it appears
you have no plans of slowing down
anytime soon so it makes me think about
these other CEOs
um the openai CEO uh what is it Google
deepmind yep
their calling to slow it down if I'm not
mistaken
would there be an ulterior motive to for
them to do this that's a good question
yeah
um
I don't know controversially I would say
like um
show me the incentive of I'll show you
the outcome right like what I mean by
that is
um I don't think any of these CEOs have
real incentive to slow down their
research institutes and like you know
absent the fact that this is hype and
blowing up
the whole prerogative of these research
institutes is to keep doing research is
to keep pushing the boundaries
um and I I think it's also in Sam
Altman's best interest and he probably
believes it as a human
um that to say you know when when in
front of Congress hey let's regulate AI
right like there's like no reason or
incentive why he would say otherwise but
there's also no reason or incentive why
he would slow down and so
um not to say that it's disingenuous
like I genuinely believe like I have a
lot of respect for Sam Altman
um and how he thinks about the world and
how he's trying to like actually bring
about AI in a safe way
um but I also think like if you're put
in front of Congress and you you have to
you know think about what a policy
should look like you're gonna think yeah
like you should regulate the stuff and I
agree
um and they're they're you know to their
benefit they're doing a few things
they're they're doing like an I think an
AI regulation coding challenge or
competition to see who can come up with
the best AI solutions to regulate Ai and
stuff like that so I think they care but
I don't know if like these petitions and
all of this are actually going to slow
it down
um to some degree like capitalist forces
are actually in many ways a good thing
right like you know open AI part of
literally why we are having this
conversation today is because chat GPT
dropped and it blew up the internet
right and everyone loves it and at the
end of the day it was a product built
like not them thinking about themselves
as a research institute but them
thinking about themselves as a product
company which is a startup right and so
I think the the fact that
um you know that they are they do have
to make some amount of money I don't
know if they're a for-profit or
non-profit anymore I can't keep track
but the fact that they're operating like
a startup and I think a lot of these
other companies that are probably
focused more on safety are operating
like a startup I think that's net
positive because that's what's breeding
The Innovation and I think the
innovation has to happen on both again
the generative side but also on the
safety side
um and you hope that that happens faster
than Bad actors right because the Bad
actors are never going to stop like
they're not going to be like oh yeah
they're slowing down there's regulations
like they're just gonna keep doing their
thing and so I don't know it's maybe a
little cynical but also like powerful in
and of itself that
um The Innovation will keep happening
and if people care enough about it and
are innovating in in us with a startup
mindset that can actually lead to good
outcomes
I asked that question because crypto is
my focus right Sam Bateman freed he was
a Golden Boy Once Upon a Time Golden Boy
yeah right right and he was he was
undercutting everyone at the very end of
the day so that's why I asked him yeah
you're like is this a Sam bankfriend
freed moment it's not good that they're
both named Sam
wow uh no but I think that's a fair
point right
um and like Integrity really matters in
this stuff when you're dealing with you
know people's money people's livelihoods
in the case of
um crypto and also in the case of AI
where there is this powerful technology
I think the good news and again this is
The Optimist in me speaking like I'm
always like okay where's the the bright
side
um but also you think about it from
first principles is AI these folks are
developing AI
because they're fascinated by it right
in the sense of all of this research was
done by people again who had the
progative to push the boundaries to like
actually figure out how does the brain
work how do we replicate that that's
where a lot of these
um Labs initially came from and a lot of
the the research so they're operating
like startups but it's it's more kind of
not altruistic but it's more real
whereas in many cases I actually think
crypto is really important and
blockchain is actually ironically going
to be even more important in this AI era
um but I do think
um a lot of people got into crypto as a
get rich quick scheme and so the
incentives were completely misaligned
and it was very it was a lot easier
um and is a lot easier for people to get
scammed or to scam
um which is unfortunate
um but I think in Ai and I'm hoping uh
there's a lot you know there's a lot
more
um just kind of substance there well
Dion what's next for you and forethought
so what's next for us to forethought is
[Music]
um
we are thinking about radically changing
how customer service is done but how
companies and customers interact more
broadly
there was this this uh I don't know
Trope going around the internet of
everyone freaking out when Google said
we have no moat against uh against open
Ai and and these llms and I actually
genuinely think the same is going to be
true we're going to start hearing more
and more news from big companies right
like what happens
um when you think about the the crms or
or the help desk or whatever the systems
of record in in customer service right
the whole customer service model is
going to change when you have ai first
interactions right
um you're let's say I need to book a
haircut or something in in 2030.
um it's not going to be me necessarily
calling up the barber and trying to book
a haircut right it might be my pocket AI
my my you know my little Albert Einstein
or whatever
um or my you know forethought
um talking to the businesses AI
um powered by forethought
um and then I'll just get a notification
like hey your your haircut is booked
right and
um and so like we think about where we
are today where the world's gonna be in
in 10 20 years and that whole
interaction system the touch points
between people and organizations look
completely differently and so we are
we're so excited to just from a product
roadmap perspective ask ourselves how do
we get there
um day in and day out
um and there's there's just so much more
to come so we're super excited thank you
so much for joining me today thank you
Rosemarie