September 7, 2023 • 45 mins
Imagine a world where AI seamlessly integrates into your daily workflows, driving productivity gains, efficiency increases, and automation.
AI and in particular Generative AI is all over the media, and now companies are looking at how they should be introducing AI into the enterprise.
While Generative AI platforms such as ChatGPT have been trained on publicly available data, they may not be suitable for always-on and mission-critical systems. So what’s the opportunity for enterprise-grade AI?
To answer this question, I’m delighted to have on this episode Umesh Sachdev, CEO of Uniphore, an Enterprise-class, AI-native company that has set out to transform businesses delivering compelling and engaging customer and employee experiences.
As we navigate the exciting yet challenging landscape of AI, we discuss potential pitfalls along the way. Umesh candidly shares insights into vital areas like regulation, data security, and total cost of ownership.
We dive into how regulation is necessary, including guardrails for AI ensuring ethical use of public data, and protecting against biases and inappropriate use.
Umesh also provided three actionable steps to ready your business for the AI revolution.
More on Umesh
Umesh on LinkedIn
Umesh on X
Resources Mentioned
Jolt Effect, The: How High Performers Overcome Customer Indecision
Uniphore website
Thanks for listening to Digitally Curious. You can buy the book that showcases these episodes at curious.click/order
Your Host is Actionable Futurist® Andrew Grill
For more on Andrew - what he speaks about and recent talks, please visit ActionableFuturist.com
Andrew's Social Channels
Andrew on LinkedIn
@AndrewGrill on Twitter
@Andrew.Grill on Instagram
Keynote speeches here
Pre-order Andrew's upcoming book - Digitally Curious
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
Intro (00:01):
Welcome to the Actionable Futurist podcast, a show all
about the near-term future, withpractical and actionable advice
from a range of global expertsto help you stay ahead of the
curve.
Every episode answers thequestion what's the future on,
with voices and opinions thatneed to be heard.
(00:23):
Your host is internationalkeynote speaker and Actionable
Futurist, andrew Grill.
Andrew Grill (00:30):
AI, and, in particular, generative AI, is
all over the media and nowcompanies are looking at how
they should be introducing AIinto the enterprise.
While generative AI platformssuch as ChatGPT have been
trained on publicly availabledata, they may not be suitable
for always-on andmission-critical systems.
So what's the opportunity forenterprise-grade AI?
(00:50):
To answer this question, I'mdelighted to have on today's
show Umesh Sashdev, the CEO ofUnifor, an enterprise-class AI
native company that has sent outto transform businesses,
delivering compelling andengaging customer and employee
experiences.
Welcome, umesh.
Umesh Sashdev (01:07):
Andrew, it's great to be on your podcast.
Thank you for having me.
Andrew Grill (01:09):
I had the privilege of sitting down with
you on a trip to London earlierthis year where we talked about
the current state of AI, and youtold me the fascinating story
of how you started Unifor Next.
You could share this story withour audience.
Umesh Sashdev (01:21):
I founded Unifor in the year 2008 and the first
thing that we set out to do orbuild or develop in Unifor was a
speech recognition engine whichcould understand multiple
globally spoken languages, alongwith its associated natural
language understanding programs,etc.
(01:43):
Over the years, what that hastransformed into is what started
with an AI company focusing onthe human voice is today a truly
a multi-modal AI platform,where today we cover the aspects
of video through computervision of voice, which is what
(02:06):
we originally started to do withspeech recognition, etc.
One text which engages areaslike chatbots, which some of us
are getting very used to rightnow, email conversations, etc.
So Unifor, over the last 16years, has dedicated itself to
creating AI models that helpmachines understand various
(02:28):
forms of human engagement, humancommunication, which span
across the modes of video, voiceand text.
Andrew Grill (02:37):
I think that's really important because we're
obviously playing a lot withchatbots, but humans also speak.
We see vision and I think beingable to have that multi-modal
is a very important part goingforward.
But perhaps you could definewhat an AI native company is.
Umesh Sashdev (02:53):
Well, ai native company is a company that was
born doing AI, as different froma traditional enterprise
software company, which thereare several of today who are now
bringing in AI into thetechnology stack.
So, at Unifor, the very firstprogram we wrote, the very first
(03:14):
technology we built was aspeech recognition with natural
language understanding, whichwas around voice AI.
So a company that was borndoing AI and that's the only
thing they've done from thebeginning and we are a prime
example.
Today we are also, from anenterprise, b2b standpoint, one
of the world's largest AI nativecompanies.
(03:35):
We service over 1500enterprises in 20 different
countries and in thoseenterprises we have 750,000
users, which happen to be theemployees of the organization,
who, every day, actively, arebenefiting from some form of AI
that Unifor deliver to them,either on their voice calls,
(03:58):
video meetings or emails andchatbots, etc.
And that, from an adoption,from a scale, from a ability to
integrate within the enterprise,makes Unifor the largest B2B AI
native company in the worldtoday.
Andrew Grill (04:14):
So I talk a lot about the challenges that
companies will face when theyintroduce AI into the enterprise
and I think people arediscovering what sort of public
generative AI like ChatGPT, whatthey can do, but what are the
unique challenges that theenterprises will face when it
comes to generative AI?
Umesh Sashdev (04:30):
Andrew, this is a very fascinating topic and, as
you would imagine, very close tomy heart right now.
It's one where we've beencalling ourselves conversational
AI and AI native company for 16years and in the last eight,
nine months, the world hasfinally met us where we've been
standing for a long period oftime.
I have not felt this level ofcuriosity, engagement from
(04:56):
Fortune 100, fortune 500 CEOsand their board members, where,
in the last few months, a bunchof folks from the C-suite,
including CEOs of these Fortune500 companies, have reached out
with the very same question.
You just asked me that whatshould we consider when we think
about announcing our AIstrategy at our next earnings
(05:17):
meetings to the world, etc.
And the considerations are verywide and deep, which is why
it's creating such discomfortwith folks who are not used to
this subject.
Let me try to unclutter ordemystify at least the big
boulders that people ought tothink about.
First, the pace at which thistechnology has evolved is
(05:39):
unprecedented.
Every 48 hours, you and I,andrew, are reading something
new about transformers ordiffusion, a new model, a net,
new innovation that we werethinking two days ago.
That has not occurred even inthe era of the internet
revolution or the cloud, whichalso means that regulation
around the world is lagging thepace of innovation, and that's
(06:02):
truly important for anenterprise leader to take into
account.
We can rest assured, given thedemands of the industry and
given where the governments haveindicated already they are
whether it's the US, europeanUnion, parts of Asia there is
imminent regulation around thecorner which very likely is
going to put a guardrails aroundusage of public data to train
(06:26):
these models the need to givecredit to content providers
whose content is going to beused in these models which will
then be serviced, servicing manyuse cases.
As an example, if you'redeveloping an AI model like chat
, gpd and you're going to usepublicly available data sources,
then some of the upcomingregulation is likely to ask that
(06:52):
the people whose content wastaken in to train these models
be given credit.
Now, that's a very hard taskbecause one of the new things
with Generative AI is thesemodels have become large, which
is why they're called largelanguage models, and when you're
dealing with that size of data,it almost becomes impossible to
pinpoint the source of thatdata.
(07:12):
So upcoming regulation is onekey area.
The next key area is we have tohave people educated enough to
demystify the total cost ofownership, of Deploying a full
solution, not just a model, in aproof-of-concept type fashion
in an enterprise.
So if you think about what arethe components you you start
(07:34):
with, these models required alot of compute, silicon chipsets
.
There's cost associated withthat is infrastructure, cloud or
a premise.
Then there's the cost of the AImodel in itself, either API
calls or some form of licensing,which even in the beginning
could feel like it's free oravailable off the shelf.
(07:55):
The minute it hits enterprisegrade, that becomes important.
And Then it's finally the layerof software around that AI
model.
So the whole stack of costneeds to be demystified and
fully understood, especially atthe scales up with these
enterprises want to operate.
The next area is Security,infosick, data security.
(08:17):
These enterprises are usuallyin a regulated environment
financial services, health carevs Consumers do not want our
personal information to be goinganywhere except the place that
we allow the enterprise to useit for.
And you know these enterpriseneed to be extremely careful via
integrations, cyber securitymeasures, infosec, to make sure
(08:39):
these AI models that are comingin to their infrastructure Are
not going to take in or misusepersonally identifiable data for
their customers.
Finally, a bunch of CEOs haveasked me this question in the
last several months and which isvery fascinating, and these are
CEOs sometimes who haveWorkforces of in the hundreds of
(08:59):
thousands of employees andthey've asked me you mesh, what
about culture, ship?
What kind of org design changedo I bring about?
What this technology, what AI,is likely to do is cause major
disruption in the way we operatethe business and I don't think
we have a choice in the matteranymore.
The world is moving towardsthat direction.
(09:20):
Efficiency Driven by some ofthese technology innovations,
especially AI, is imminent.
Now we cannot wish it away, butto get my entire workforce on
board To move them from worriedabout will my job last to this
is exciting.
Let me partner with my CEO inmaking this transformation.
(09:42):
That is truly fascinating, andI'll close this topic with one
anecdote.
One of the CEOs I quote veryoften is one of the largest
companies in France and they'rein the government public
services department and I wassitting down with him and he was
picking my mind on you knowwhat we are seeing elsewhere and
(10:05):
our thousand plus customers howour people applied AI, one of
the pitfalls.
And Then he told me, even beforebringing his executive team on
board with the topic.
He first went to the frontlineworkers, the folks who deliver
posts and mail To residents inFrance.
He went to those tens ofthousands of employees To bring
(10:29):
them on board and when he metthem, short enough, their first
question was are our jobs atrisk because of AI?
And his answer was Well, that'snot the right way to think
about it.
The right way to think about itis let's think about the power
of what AI can be doing for allof us making our lives easier as
we go about delivering mail andcosts, etc.
(10:51):
And if you partner with me, youwill have a seat at the table
and picking which AI tools thiscompany brings on board, etc.
And he says that changed thecomplexion of the whole
organization Embracing AI.
So it's a bunch of involvedtopics that many different CEOs
are being very Involved in and,given the experience we have
(11:13):
with our vantage point of 20different countries and
thousands of customers, we'redoing everything we can to share
what we've learned over theseyears.
Andrew Grill (11:21):
Lots of different threads, in the end one pack.
And I recall when we hadbreakfast in London I talked
about the fact that the chat dptlaunch in November 2022 was a
watershed moment because itremoved the friction.
My parents in Adelaide,australia, are talking about
chat GPT and I said where didyou hear about that?
Oh, it was on the news.
I stood on stage in Abu Dhabi afew months ago with the CTI of
(11:41):
Amazon and he said we've beenusing AI for 25 years.
Now it's become popular.
Which brings me to the pointthat I think what it's done and
that's why the CEOs are askingyou these questions it's removed
the friction.
A CEO, a C-suite person, canactually play with the tool
without having to learn Pythonor run scripts or set up APIs.
They can very quickly see thepower of something like a
(12:04):
generative AI platform.
But I think what that's alsodone is it's raised the
visibility of this Technology tothe regulators.
What do we do about regulation?
Do we regulate the tech or weregulate the use of the tech?
And how do we get theregulators up to speed to
understand what the power ofthis is and when the guard rails
should be?
Umesh Sashdev (12:23):
if you think about all the time that you and
I have been around.
We've seen the internetrevolution, we've seen the cloud
and the mobile era.
I Don't remember a time whenthe creators of technology,
typically Silicon Valley folks,were the first to come out and
ask for regulation.
Usually, the creators oftechnology are very bullish on
(12:43):
the outcome that technologydelivers and folks in government
, whether it's Washington or 10Downing Street or Paris, etc.
Are the ones pushing forregulation.
This is the first time, to myrecollection, that Silicon
Valley, simultaneous toannouncing the innovation, has
also started to ask for Guardrails and regulation, and we
have to understand why.
(13:04):
There have been many theoriesabout why this is happening.
People want to create, you know, competitive Pressures, etc.
I think the ask for regulationis for a different reason, and
that is for the first time we've, as human kind, created a
technology that is capable ofmaking decisions like the human
(13:26):
brain.
In the past, we've createdtechnology that could automate
repetitive tasks, take away somethings that we used to do,
mimic human behavior, but nevermake decisions like Generative
AI can.
And so the examples tounderstand how profound the
power of this technology is youcould instruct a Generative AI
(13:50):
technology bot like a chat GPD,and there are versions of chat
GPD today which are calledauto-GPT, where you can ask it
to book me a restaurant inOxford Street for two on
Saturday night and I'd like totry an American meal.
Without saying anything more,auto-gpt takes in the
(14:13):
instruction and then startsadding new tasks to its list
until it gets the desiredoutcome.
So it will go to a website likeYelp figure out restaurants
which match the description,then go send emails to that
restaurant asking foravailability.
Once it's confirmed, use thecard on file to go hit a payment
(14:36):
gateway, finish the paymentuntil it finally receives
confirmation and emails theconfirmation back to the actual
requestor, the person who madethat request.
Nobody told it how to go aboutthese tasks.
It discovered these tasks asyou went about it.
Now another version of auto-GPTcame about which is called
(14:58):
Chaos GPD.
So somebody took the exact sameprogram and to show the world
the negative effects of thisprofound technology, they
created Chaos GPD and theinstruction given to it was the
programmer created fivefictitious individuals who don't
exist and said go find allpublicly available records of
(15:19):
these five individuals on theWorld Wide Web, use any means
necessary to hack into theircredentials on the World Wide
Web and, once done, erase allknown records of them from World
Wide Web.
Much like the example of Find Merestaurant, the machine now is
(15:42):
relentless.
It keeps adding tasks to itselfuntil the desired outcome is
reached.
So it's truly important for usto understand that this is a
relentless task machine.
It does not stop until whatit's being asked to do is
actually achieved.
And the thing it's lackingtoday, which is the area of
(16:06):
necessary research for a lot ofAI companies, including Unifor,
is it lacks judgment.
We, as human beings, have beengifted by nature.
Not only can we make decisions,but, as making decisions, we
know this is good, this is bad.
Even if we decide to dosomething bad, we're conscious
about it.
Ai is, today not conscious.
(16:27):
It's not sentient, it lacksjudgment, and that makes it a
relentless task machine.
Therefore, in the wrong hands,it has the potential to create
serious damage to sovereigns, tosecurity apparatuses of
governments, to cyber risks, tous citizens and humankind, which
(16:48):
is why there has been an askfor regulation.
That said, given how fast theinnovation is, regulation is
easier said than done.
How do you regulate somethingthat's evolving every 40 hours
without curtailing the power ofinnovation?
You want a teenager now, likeyou said, your parents in
Australia, a teenager in India.
(17:10):
You want all those folks who arenow beginning to enjoy the
technology to come up with newideas.
You want them to innovate.
At the same time, you want toput guardrails on.
Anyone putting out new modelsshould almost go through a
certification process, and thepharmaceutical industry has
created a model for this, wherenothing comes out in the
(17:33):
pharmaceutical industry withoutclinical trials, blind studies.
So we probably need some formof regulation like that for AI,
where a new model will not hitthe world without some sort of
blind studies and clinicaltrials.
At the same time, unlike pharma, we can't take years for a
model to be approved.
So it's one where I don't envythe role of administrators who
(17:58):
are simultaneously trying tolearn the technology, learn the
implications, come up with verythoughtful approaches of putting
the guardrails, which makes itimperative for us in the
industry to partner with thegovernment to show them what we
know, the power, the pitfalls,etc.
(18:19):
And then help them come up withmeaningful regulation on the
subject.
Andrew Grill (18:24):
I've always said it's got to be a partnership.
I've had a couple ofexperiences where I've been able
to present to the regulators,to the lawmakers of governments
here in the UK, and what I sayto them is you need to think
like a startup, you need topartner yourself with people
generating this technology thisis way before chat, gpt and
understand why they want to inmy term break some rules.
Why do they need some freedomto operate?
(18:46):
And I think when industry andregulators get together you made
a good point the regulatorsneed to almost learn faster than
the developers of thistechnology how it all works, to
be able to have the guardrails.
But generally, a regulatorisn't going to go and work for a
government organization that isan AI expert.
So I think that's a real issue,but probably not something we
can solve.
On the podcast today, I wantedto turn more to your own
(19:09):
solutions.
At Unifor, you've got severalsolutions conversation AI,
emotion AI, generative AI andknowledge AI.
Perhaps you could give us aflavor for each of these and
what problems they solve.
Umesh Sashdev (19:20):
We think of these as Lego blocks of technology
and AI models which we'vearchitected as a platform.
What does that mean?
One of the things we realizedvery early on working with even
our early customers and that hasremained true till day Now that
we have over 1,500 enterprisesusing our technology is that
(19:40):
when we approach any of theseenterprises, they get very
excited about the potential ofAI and we go okay, we're
bringing our full prowess of theplatform.
Please give us access to yourdata, because we don't want to
bring in publicly trained models.
We want to use enterpriseapproved data.
Then we realized that theseenterprises have never arranged
(20:03):
their data in a way that's readyfor AI.
A lot of the data we get areunstructured forms of data FAQ
documents, standard operatingprocedure documents, historic
recordings of calls from thecall center, etc.
Which are all unstructuredforms of data.
(20:24):
Several years ago, we startedinvesting in this field of AI
called Knowledge AI, which takesin all forms of unstructured
data, creates a knowledge graphfrom whatever it's seen in those
documents and those callrecordings and outputs a very
structured form of what itlearned from that data.
(20:44):
We then take this as a pipelineand feed the structured output
into our Generative AI models.
In Generative AI we use allT-shirt sizes.
These models come in a small,medium, large and extra large
size, depending on what we wantthem to achieve.
But the data that's going intoGenerative AI is being filtered
(21:05):
through our Knowledge AI models.
We then discovered, like I wasmentioning in the previous
question, that to add a layer ofjudgment, to add a layer of
understanding of human emotion,our state of mind, empathy,
sarcasm, anger, even especiallyfolks when they're calling into
(21:27):
the call center with thecomplain, and so on and so forth
, it's not enough to haveGenerative AI become a task
machine.
Uniform is also known prettywidely in the academic circles
for the work we've done inEmotion AI, when we've taken the
changes of tone via voice.
We measure using computervision if it's a video meeting,
(21:52):
facial emotions, gestures, bodylanguage, etc.
To train the AI model onunderstanding human emotional
states.
We make all these three formsof AI Knowledge AI, generative
AI, emotion AI work in tandem.
They work in real time.
What do they actually do in theenterprise?
(22:14):
We have four types of benefitsthat enterprise derived from our
technology.
First is in the contact centerswhere, for the most part,
uniforms technology is aco-pilot to the men and women
who work in these call centers.
Dan and I are receiving yourand my calls when we have a
complaint, we have an issue, weneed urgent help.
There's somebody on the otherside of a 1-800 number that
(22:35):
picks up the phone.
Uniforms AI is working as aco-pilot on that call,
understanding the issue that thecustomer is making or
verbalizing, then, throughautomation, finding very quick
answers to that problem.
Even before the human agentcould have put the customer on
hold and tried to find thesolution, the AI finds it and
(22:58):
delivers a great outcome.
The call is shorter, it's moreefficient and it's a better
customer experience.
We, then, have applied thisco-pilot technology in sales
automation.
We found salespeople whenthey're trying to sell to their
customers on Webex and Zoom andTeams meetings.
It's very hard for them to readthe room on this virtual setup.
(23:19):
You have a presentation in themiddle, you have 10 people who
are attending the call.
You can't watch if people arereceiving your message or
they're still confused about thetopic, etc.
Once again, uniforms AI acts asa co-pilot to those salespeople
on these virtual meetings.
We, then, are servicing lots ofgovernment departments.
We're servicing policedepartments, ambulance services
(23:42):
in the UK, nhs etc.
Are using us in their phonelines as an AI technology.
Finally, we have customers intrading desk terminals, folks
who are constantly on theirtrading computer screen along
with phone lines.
A multitude of places where theenterprise has either a phone
(24:04):
conversation or a video meetinggoing on in their frontlines is
all places where Uniformstechnology is aiding and
assisting those departments andthose employees to be more
efficient and deliver bettercustomer experiences.
Andrew Grill (24:19):
You make a great point there.
In my AI talks, I labour thepoint about having data that's
AI ready.
I ask my audiences threequestions what's the data you
want, what's the data you needand what's the data that you'd
like?
What format is in?
Who owns the rights to it?
How do you access it?
I see the audience pausing andgoing.
We hadn't thought about that.
We just thought it was allmagic and that the data we've
(24:41):
got in the company can actuallybe ready to be ingested.
I want to talk about training.
Once you've got the data,you've got to start training it.
Ai training bias is becomingmore evident as these public
large language models aretrained on, I would say,
questionable data.
What do you think should bedone in this area to ensure that
training data is lesssusceptible to bias?
Umesh Sashdev (25:01):
Well, you lead me to a very important subject.
This is one that we'll have tobe very mindful of as long as AI
is going to be in our life andfor posterity.
Ai bias and training bias is avery real threatened issue to
the efficacies of these models.
Once again, we spoke about thelargeness of these data sets
(25:22):
large language models, largemultimodal models.
Those are what are poweringmodern day generated AI
technologies.
When you have billions andbillions of parameters training
a single AI model, very likelythat some undesirable form of
(25:44):
data enters your models.
If it were a consumer app, therisks are you could train it on
pornographic materials.
You could train it on age andappropriate materials.
In the enterprise, the risksare even deeper.
They're about biases towardsdiversity, inclusiveness, etc.
(26:07):
Let me give you an example.
One of the models that Jenniferfocuses on is Emotion AI.
We use computer vision modelson video meetings, on Webex or
Zoom, etc.
To pay attention to humanfacial emotions as being
conveyed by the participants onthat meeting.
Now, as we were training thesemodels, if our researchers did
(26:32):
not pay attention to being veryinclusive in their data sets,
right then to include faces andfacial emotions of people with
different races, genders, skincolor types, size of eyes, etc.
If those variations were nottaken into the training model,
very likely that the model, whenit started to work within the
(26:54):
enterprise, would start to givea biased output when, if it's
encountering a type of face witha skin color type or size of
eye that it was not trained tolook at, it might misread their
emotional state during a meetingand that could cause major
problems, potentially even legalissues.
(27:15):
Similarly, these generative AImodels you're teaching them
language, human language, humancommunication, and if you're not
paying attention, they canstart to pick up, for example,
cus words.
Now, when people are angry,they speak to each other in a
certain way, the AI model startsto learn how to be angry at us,
(27:37):
etc.
And so what is becoming a veryfascinating subject?
As you and I know, the socialmedia companies met up or
Twitter and so on and so forth.
We've known them to have verylarge teams of trust and safety
executives, armies of peoplewhose job was it to try and
(27:57):
catch these undesirable forms ofdata and untrain it from the AI
model, and these concepts wereused in social media companies.
The very same concept is nowcoming into AI companies, where
someone like us or a chat, gptor anyone creating their own
proprietary models now need tohave a trust and safety
(28:19):
department within theorganization whose core job
should be to findimproprietaries, to find
undesirable data that'sinadvertently entering the model
and take it out as soon as it'sfound or cited, so that the
bias is controlled.
So it starts with all of usbeing very conscious.
(28:43):
If we are creating a model, wehave the responsibility to be
very conscious that our modelscould have biases of different
types.
And once you're conscious, wehave to make the investment,
whether people investment ortechnology investment, to make
sure that those biases can beminimized as much as possible.
(29:04):
And yet again, because of thenature of this technology, the
pace of innovation, it's almostimpossible to completely make do
away from it.
But as long as there'sconsciousness on the part of the
provider of the model, manymeasures can be taken to
minimize the risk of biases inthese AI models.
Andrew Grill (29:27):
Gartner's latest hype cycle for AI that's just
been released puts Genitive AIat peak hype no surprises.
So when do you think it willmove to the plateau of
productivity and what will ittake to get there?
Umesh Sashdev (29:38):
If you think about what's happening in
today's world.
Unifor is the first companywith examples of over 1,500
enterprises using us forproductivity gains in their call
centers, et cetera.
Our largest customer, whichhappens to be a financial
services organization in NorthAmerica, has 65,000 of their
(29:58):
call center agents using ourtechnology.
Our second largest customer isa large telecom service provider
and 26,000 of their users useus in their call center, and we
have many such examples.
Why do I bring these up?
You and I both know that whenthese enterprises adopt any new
technology, not just AI, theyfirst start with a small proof
(30:22):
of concept which becomes a pilot, and then a different
department is added to thatpilot.
By the time, a technology ishitting scales of 50,000, 60,000
users, like the examples I justcited, which means the
technology is delivering provenreturn on investment.
It's delivering on businessoutcomes and Unifor is one of
(30:44):
the first companies proving itat its enterprise scale that
this technology is ready forprime time.
It is delivering efficiencygains, et cetera, and we're
going to see an acceleratedtrend in the coming months of
more and more enterprises, moreand more vendors, beginning to
find ways to make it happenwithin the enterprise.
You asked me a question whatwill it take for the technology
(31:07):
to move from the peak of thehype cycle to the plateau of
productivity?
I think here's what it willtake, andrew, from my vantage
point.
I talk to many CEOs who arevery forward looking, who push
us, and we think we're a fastmoving Silicon Valley company.
Some of my customers, whohappen to be CEOs of larger
businesses and traditionalbusinesses, are pushing us to
(31:29):
move even faster.
What I know is some of them whoare captains of industries,
whether it's cybersecurity orhealth care, and these happen to
be publicly listed, publiclytraded companies At least some
of them in the forthcomingquarters will, in their earnings
meetings, earnings results,declare productivity gains
(31:54):
because of their use aggressiveuse of generative AI, and all it
will take is one large companyin each of those segments to set
the tone with the investors.
We saw what happened when asocial media company last year
which was publicly traded, gottaken private by one individual
(32:15):
and then that individual, forright around reasons, decided to
have massive headcountdecreases in that organization.
It forced every other peercompany of theirs In social
media, in the internet arena.
The pressure from Wall Streetwas if one company can do it,
you all should do it and we sawa series of headcount reductions
(32:38):
in layoffs.
Generative AI is actually moreproductive than that.
However, once one CEO, onecompany, shows the potential of
this technology in theirearnings call in one of the
upcoming quarters, wall Streetand investors will be the
biggest force putting pressureon every company to use such
(33:01):
technology to drive productivitygains.
And I think we are at the cuspof it.
So I won't be surprised if, inless than six months from now,
you and I are having a chatabout this topic and we are
saying look how fast we movefrom the peak of interest to
this plateau of productivity inthe hype cycle as we know it.
Andrew Grill (33:22):
So you're in a unique position because your
solutions actually work deeplyinside the organizations you've
mentioned to solve theseproblems.
But for others that don't havethat deep connection, for
example, Microsoft are launchingtheir Microsoft 365 co-pilot AI
, which integrates directly withproducts such as Outlook and
Word right in the daily workflow.
So how can enterprise AI getcloser to the day-to-day
(33:44):
activities we're already doingand become more seamless?
Umesh Sashdev (33:47):
It's really important to reinstate that the
real power of this technology istruly profound.
The outcomes that we canimagine are unbounded.
Every facet of an enterprisejob is likely to benefit, with
productivity gains, automation,efficiency increases, and it
(34:09):
could be as simple as how do wetype emails to our customers?
How do we send emails to ouremployees?
How do we communicate withinthe companies?
How do we communicate with ourcustomers?
Do we communicate with themover phone lines and call
centers?
Do we communicate with them onvideo meetings?
Or do we meet them in person atstores and so on and so forth?
(34:30):
And so the day-to-day activityof every single employee, right
from the CEO to the youngestintern who joined the company,
is, in my opinion, very likelyto be impacted and benefiting
from the user-generated AI.
If they're not already doing it, it's going to happen very
(34:51):
shortly.
Like you said, companies likeMicrosoft, et cetera, who work
on these productivity tools likeemails and other forms of
communication, are bringing thattechnology.
We play in the arena ofcustomer experience, employee
experience tools anywhere in thecompany where there's a phone
conversation, a video meetingand so on.
(35:11):
So far, we are bringing thosetools.
There are many companies whoare bringing generated AI into
back office operations,accounting, billing, legal
documentation and then shuttingdown this whole tailwind like
small companies, the mother ofyour use cases in the enterprise
.
You spoke about enterprises notbeing ready with data.
(35:33):
Data is everywhere except it'snot available when the
enterprises need to access it.
Enterprise search the ability toarrange enterprises data from
any part of the company, whetherit's in some silo in a
different branch office or it'sin a CRM or if it's in the
billing system.
The ability to put it all in asingle data domain.
(35:57):
And then, using Generative AIand all the techniques that
we've spoken of today Knowledge,ai, et cetera make it available
through a natural languagesearch interface.
But imagine the ease, insteadof going into file rooms and
trying to look for an old record, if you could chat with the bot
and say I'm looking for thatold file, which might be a 20
(36:19):
year old record of you know,something we did back then, can
you please find it for me?
And boom, in 60 seconds thatdata element is in front of your
machine or your cell phone asan output of that chat you did
with that interface.
So I repeat, the possibilitiesare unbounded and as we speak
(36:43):
right now, andrew, there aremany different innovators, many
companies who are working ondifferent pieces of this puzzle,
and that is the power thatGenerative AI has given us that
we're moving with innovation ata pace that we have not
experienced in any technologyrevolution in the past.
Andrew Grill (37:00):
You quite rightly say that the real power of
Enterprise AI is being able tofind that needle in the haystack
.
And I think back to chat GPT,the fact that people are saying
we can use this on public data.
And I say to my clients imagineif you could do that with your
own data and the light bulb goesoff.
But to your point before.
When people then scramble andsay, okay, let's do an
Enterprise AI project, I thenwarn them about the cost, that
(37:24):
it's probably a 10X in terms ofcompute power.
The data's got to be in theright place.
You need data governance.
So what would be your advice tosomeone listening to the
podcast that's seen the power ofGenerative AI through a chat
GPT wants to move this into theEnterprise.
Where should they start andwhat are the pitfalls to avoid?
Umesh Sashdev (37:41):
Several of the CEOs have come to me with the
same question.
It's first, it's reallyimportant to get a core group of
people, almost a Tiger team,formulated around the CEO, who
are both empowered and excitedto drive the change, drive the
transformation within theorganization with this
initiative.
(38:01):
It's not just a technologyinitiative, it's a change
management issue within theseenterprises.
It then becomes important topaint a big, hairy or decious
goal, almost a big vision, whichit can excite the whole
workforce to drive the not star,to say, in the next few years,
using AI, we're gonna be acompany, we're gonna be an AI
(38:22):
company, we're gonna be acompany that will be very
efficient, we're gonna be acompany that delivers the best
customer experiences, and so onand so forth.
Then break that vision downinto smaller chunks of projects.
What's achievable in the nextsix months?
What are the low hanging fruits?
There are areas like callcenters and customer service
(38:42):
which are extremely ripe.
The employees there are cryingfor tools.
The employees don't likeputting customers on long holes
only to search for the answer totheir questions.
The customers don't like thatlong experience when they're in
a hurry and all they need is aquick answer to a question of
when's the next slide, etc.
(39:02):
Or why was I overcharged on thebailing?
So there are departments andthere are low hanging fruits of
efficiency gains.
And so, having created a team,having defined a vision, having
found the first high impact buteasy to understand area to
deploy this at example, customerservice it then becomes
(39:23):
imperative for these companiesand CEOs to over communicate at
every step of the way, to overcommunicate with the customers,
with their shareholders and withtheir employees.
In my opinion, it takes six toseven quarters and seven
repetitions of the same messageby the CEO For everyone in the
company and the community tounderstand and buy into the
(39:47):
shift that's been caused you.
And so not only doing it onetime, not only saying we have an
intention, but showing itrepeatedly over six or seven
different quarters, six or sevendifferent earnings, six or
seven different messages, etcetera, is what's needed to land
the message.
And then the flywheel rotatesvery fast If done right.
(40:10):
The idea generation of where toapply this next can be
crowdsourced from within theenterprise.
But for that to happen, it'simportant to first sell the
message really strongly and showand lead by example, show with
a couple of true points in oneor two areas and deliver the
results.
Andrew Grill (40:30):
You say the future of enterprise, ai, is human.
What do you mean by this?
Umesh Sashdev (40:35):
Well, this is the first time that we have the
ability to imagine a world wherewe, as human beings usually we
are told to learn a new skilland when a new technology has
been released, we have toretrain ourselves.
We have to unlearn our pasthabits, retrain ourselves on a
new technology and your tool.
And we are just so.
(40:57):
If you think aboutphilosophically that phenomenon,
that phenomenon is us humanbeings adjusting to technology.
We actually think that tounlock the real power of AI
within enterprises, it willtruly have to be that the
technology adapts to the humanbeings who work in those
(41:19):
enterprises.
That is when we will have thefull buy-in of the human
workforce in every company,every department, everyone
cherishing and, you know, beingenthusiastic about the
technology.
And that is why, as our vision,unifor, we've said the future
(41:41):
of enterprise AI has to be human.
What does that mean?
We spoke a lot about GenerativeAI becoming a relentless task
machine.
A relentless task machine thattoday does not appreciate human
emotions, does not appreciateevery nuance of how human beings
communicate with each other.
(42:01):
You and I don't just send eachother chat messages.
When you and I speak on phoneor a video meeting, we change
our tone.
We can be excited, andrew, haveyou seen this new release by
its own company?
Or we say, hey, that is sodangerous, somebody should be
paying attention.
Just by changing our tone, wecommunicate with each other in
addition to our words.
When we meet in person, like wedid in London and we sat down
(42:24):
for breakfast, or we're meetingon a video meeting, we're
looking at each other and we'reresponding to each other's
energies and facial emotions,and it's truly important that AI
meets us where we are as humanbeings, as opposed to asking us
to adjust our habits.
We want the AI to start withappreciating human emotions
(42:46):
along with the words, along withour language, and then,
hopefully, there'll be a daywhere, using Generative AI, we
can teach AI to also generateits own emotions and thereby
being a great companion, a greatpersonal assistant, whether in
our daily lives or in theenterprises.
And so our vision of EnterpriseAI is human Really means that
(43:09):
we will continue to pushboundaries, push the envelope,
not settle with the fact that AIis a great productivity tool
right now.
But AI, for AI to be acompanion for each of us in the
enterprise or personal lives, ithas to fully understand the
nuances Of human communication.
Andrew Grill (43:27):
We're almost out of time.
We're up to my favorite part ofthe show, the quickfire round,
when we learn more about ourguests iPhone or Android, iphone
Window or aisle.
Umesh Sashdev (43:35):
I'm an aisle person.
Andrew Grill (43:36):
In the room or in the metaverse, always in the
room.
What's the first thing you ask?
Chat GPT.
Umesh Sashdev (43:41):
I ask it about what does it know about me?
Andrew Grill (43:43):
Your biggest hope for this year and next.
Umesh Sashdev (43:45):
That we meaningfully put guardrails
around AI and safely introducethis profound power on citizens,
on employees in different partsof the world.
Andrew Grill (43:54):
I wish that AI could do all of my.
Umesh Sashdev (43:57):
Well, it could follow up all my emails and tell
me the ones that really need myattention.
Andrew Grill (44:00):
The app you use most on your phone.
Umesh Sashdev (44:02):
I'm constantly looking at the weather app.
Andrew Grill (44:04):
The best piece of advice you've ever received.
Umesh Sashdev (44:06):
It was from a mentor of mine who taught me
early on that as important as itis to build a business and lead
a company with IQ, EQ isequally important.
So intelligence and emotionalintelligence go hand in hand.
Andrew Grill (44:20):
What are you reading at the moment?
Umesh Sashdev (44:21):
Well, I'm reading this book.
It's called the Jolt Effect andit's a very fascinating book
for all of us who sell tosomebody, and it's about what
causes a customer's indecisionor the inertia to maintain
status quo, even if they knowthat it's right for them to
(44:42):
adopt something new or change tosomething new.
It's very pertinent in this erawhere we're pushing Generative
AI and a new concept to ourcustomers and you see this every
day where the customer willtell you I'm sold, I'm really
excited, let's move.
And then they go on a radiosilence.
And this book, the Jolt Effect,talks about how do you jolt
(45:07):
your customers if you trulybelieve in the topic how you
jolt them into realizing it'simportant for them to make a
decision now.
Andrew Grill (45:14):
Who should I invite next onto the podcast?
Umesh Sashdev (45:16):
Well, I'm a big fan of your podcast.
You've had many interestingpeople you cover.
Your style is very interestingto me.
You cover a range of topics but, given the world we live in,
there are many fascinatingpeople.
I'd love to hear from Elon Musk.
Would be on top of that list.
Andrew Grill (45:31):
It's not the first time someone suggested that.
So, elon, if you are listening,please return my calls.
How do you want to beremembered?
Umesh Sashdev (45:37):
I want to be remembered for somebody who is
passionate about makingtechnology work to make the
world a better place.
Andrew Grill (45:44):
As this is the actionable futures podcast, what
three actionable things shouldour audience do today to prepare
for a world of enterprise gradeAI?
Umesh Sashdev (45:54):
First, it's really important to understand
any subject and move from beinganxious about it to being very
comfortable and excited about itis to read, to educate yourself
.
Andrew Gill's podcasts are onegreat area to learn about the
subject.
There are many books andinformation out there about
generative AI out there.
(46:15):
So first it's really importantto educate ourselves.
Next, it's important tounderstand our rights data
protection rights, etc.
Even before any new regulationis released.
Most countries give tremendousprotection to us as citizens,
employees and so on and so forth, so it's very important to feel
comfortable that we areprotected.
(46:37):
And third then, just be curious.
Be willing to learn somethingnew every day, be willing to try
a new technology that's put out.
I love the story you just toldme, Andrew, that your parents in
Australia, at a late stage inlife, are willing to be curious,
learn something new and changethe world.
Andrew Grill (47:01):
As you know, my next book is going to be called
Digitally Curious, and I hopepeople are curious enough to
look at it and hear from leaderssuch as yourself, umesh, a
fascinating discussion.
How can we find out more aboutyou and your work?
Umesh Sashdev (47:12):
We try to be very public.
The company I run as CEO isUnifor.
We have a simple website,wwwuniforcom.
On LinkedIn, if you put myfirst and last name, pumir, to
such day, you'll find myLinkedIn.
I'm on Twitter, which is nowcalled X, and my email is
available on several of thesesources, so I'm always open to
(47:36):
listening and receiving messagesfrom people.
If somebody has a question,I'll be happy to hear from them
and be very quick in answeringthose messages.
Andrew Grill (47:45):
Umesh, always a delight to talk to you in person
or online.
Thank you so much for your timetoday.
Umesh Sashdev (47:50):
Andrew, it was a pleasure, thank you.
Intro (48:27):
Thank you.
Welcome to the Actionable Futurist podcast, a show all
about the near-term future, withpractical and actionable advice
from a range of global expertsto help you stay ahead of the
curve.
Every episode answers thequestion what's the future on,
with voices and opinions thatneed to be heard.
(00:23):
Your host is internationalkeynote speaker and Actionable
Futurist, andrew Grill.
Andrew Grill (00:30):
AI, and, in particular, generative AI, is
all over the media and nowcompanies are looking at how
they should be introducing AIinto the enterprise.
While generative AI platformssuch as ChatGPT have been
trained on publicly availabledata, they may not be suitable
for always-on andmission-critical systems.
So what's the opportunity forenterprise-grade AI?
(00:50):
To answer this question, I'mdelighted to have on today's
show Umesh Sashdev, the CEO ofUnifor, an enterprise-class AI
native company that has sent outto transform businesses,
delivering compelling andengaging customer and employee
experiences.
Welcome, umesh.
Umesh Sashdev (01:07):
Andrew, it's great to be on your podcast.
Thank you for having me.
Andrew Grill (01:09):
I had the privilege of sitting down with
you on a trip to London earlierthis year where we talked about
the current state of AI, and youtold me the fascinating story
of how you started Unifor Next.
You could share this story withour audience.
Umesh Sashdev (01:21):
I founded Unifor in the year 2008 and the first
thing that we set out to do orbuild or develop in Unifor was a
speech recognition engine whichcould understand multiple
globally spoken languages, alongwith its associated natural
language understanding programs,etc.
(01:43):
Over the years, what that hastransformed into is what started
with an AI company focusing onthe human voice is today a truly
a multi-modal AI platform,where today we cover the aspects
of video through computervision of voice, which is what
(02:06):
we originally started to do withspeech recognition, etc.
One text which engages areaslike chatbots, which some of us
are getting very used to rightnow, email conversations, etc.
So Unifor, over the last 16years, has dedicated itself to
creating AI models that helpmachines understand various
(02:28):
forms of human engagement, humancommunication, which span
across the modes of video, voiceand text.
Andrew Grill (02:37):
I think that's really important because we're
obviously playing a lot withchatbots, but humans also speak.
We see vision and I think beingable to have that multi-modal
is a very important part goingforward.
But perhaps you could definewhat an AI native company is.
Umesh Sashdev (02:53):
Well, ai native company is a company that was
born doing AI, as different froma traditional enterprise
software company, which thereare several of today who are now
bringing in AI into thetechnology stack.
So, at Unifor, the very firstprogram we wrote, the very first
(03:14):
technology we built was aspeech recognition with natural
language understanding, whichwas around voice AI.
So a company that was borndoing AI and that's the only
thing they've done from thebeginning and we are a prime
example.
Today we are also, from anenterprise, b2b standpoint, one
of the world's largest AI nativecompanies.
(03:35):
We service over 1500enterprises in 20 different
countries and in thoseenterprises we have 750,000
users, which happen to be theemployees of the organization,
who, every day, actively, arebenefiting from some form of AI
that Unifor deliver to them,either on their voice calls,
(03:58):
video meetings or emails andchatbots, etc.
And that, from an adoption,from a scale, from a ability to
integrate within the enterprise,makes Unifor the largest B2B AI
native company in the worldtoday.
Andrew Grill (04:14):
So I talk a lot about the challenges that
companies will face when theyintroduce AI into the enterprise
and I think people arediscovering what sort of public
generative AI like ChatGPT, whatthey can do, but what are the
unique challenges that theenterprises will face when it
comes to generative AI?
Umesh Sashdev (04:30):
Andrew, this is a very fascinating topic and, as
you would imagine, very close tomy heart right now.
It's one where we've beencalling ourselves conversational
AI and AI native company for 16years and in the last eight,
nine months, the world hasfinally met us where we've been
standing for a long period oftime.
I have not felt this level ofcuriosity, engagement from
(04:56):
Fortune 100, fortune 500 CEOsand their board members, where,
in the last few months, a bunchof folks from the C-suite,
including CEOs of these Fortune500 companies, have reached out
with the very same question.
You just asked me that whatshould we consider when we think
about announcing our AIstrategy at our next earnings
(05:17):
meetings to the world, etc.
And the considerations are verywide and deep, which is why
it's creating such discomfortwith folks who are not used to
this subject.
Let me try to unclutter ordemystify at least the big
boulders that people ought tothink about.
First, the pace at which thistechnology has evolved is
(05:39):
unprecedented.
Every 48 hours, you and I,andrew, are reading something
new about transformers ordiffusion, a new model, a net,
new innovation that we werethinking two days ago.
That has not occurred even inthe era of the internet
revolution or the cloud, whichalso means that regulation
around the world is lagging thepace of innovation, and that's
(06:02):
truly important for anenterprise leader to take into
account.
We can rest assured, given thedemands of the industry and
given where the governments haveindicated already they are
whether it's the US, europeanUnion, parts of Asia there is
imminent regulation around thecorner which very likely is
going to put a guardrails aroundusage of public data to train
(06:26):
these models the need to givecredit to content providers
whose content is going to beused in these models which will
then be serviced, servicing manyuse cases.
As an example, if you'redeveloping an AI model like chat
, gpd and you're going to usepublicly available data sources,
then some of the upcomingregulation is likely to ask that
(06:52):
the people whose content wastaken in to train these models
be given credit.
Now, that's a very hard taskbecause one of the new things
with Generative AI is thesemodels have become large, which
is why they're called largelanguage models, and when you're
dealing with that size of data,it almost becomes impossible to
pinpoint the source of thatdata.
(07:12):
So upcoming regulation is onekey area.
The next key area is we have tohave people educated enough to
demystify the total cost ofownership, of Deploying a full
solution, not just a model, in aproof-of-concept type fashion
in an enterprise.
So if you think about what arethe components you you start
(07:34):
with, these models required alot of compute, silicon chipsets
.
There's cost associated withthat is infrastructure, cloud or
a premise.
Then there's the cost of the AImodel in itself, either API
calls or some form of licensing,which even in the beginning
could feel like it's free oravailable off the shelf.
(07:55):
The minute it hits enterprisegrade, that becomes important.
And Then it's finally the layerof software around that AI
model.
So the whole stack of costneeds to be demystified and
fully understood, especially atthe scales up with these
enterprises want to operate.
The next area is Security,infosick, data security.
(08:17):
These enterprises are usuallyin a regulated environment
financial services, health carevs Consumers do not want our
personal information to be goinganywhere except the place that
we allow the enterprise to useit for.
And you know these enterpriseneed to be extremely careful via
integrations, cyber securitymeasures, infosec, to make sure
(08:39):
these AI models that are comingin to their infrastructure Are
not going to take in or misusepersonally identifiable data for
their customers.
Finally, a bunch of CEOs haveasked me this question in the
last several months and which isvery fascinating, and these are
CEOs sometimes who haveWorkforces of in the hundreds of
(08:59):
thousands of employees andthey've asked me you mesh, what
about culture, ship?
What kind of org design changedo I bring about?
What this technology, what AI,is likely to do is cause major
disruption in the way we operatethe business and I don't think
we have a choice in the matteranymore.
The world is moving towardsthat direction.
(09:20):
Efficiency Driven by some ofthese technology innovations,
especially AI, is imminent.
Now we cannot wish it away, butto get my entire workforce on
board To move them from worriedabout will my job last to this
is exciting.
Let me partner with my CEO inmaking this transformation.
(09:42):
That is truly fascinating, andI'll close this topic with one
anecdote.
One of the CEOs I quote veryoften is one of the largest
companies in France and they'rein the government public
services department and I wassitting down with him and he was
picking my mind on you knowwhat we are seeing elsewhere and
(10:05):
our thousand plus customers howour people applied AI, one of
the pitfalls.
And Then he told me, even beforebringing his executive team on
board with the topic.
He first went to the frontlineworkers, the folks who deliver
posts and mail To residents inFrance.
He went to those tens ofthousands of employees To bring
(10:29):
them on board and when he metthem, short enough, their first
question was are our jobs atrisk because of AI?
And his answer was Well, that'snot the right way to think
about it.
The right way to think about itis let's think about the power
of what AI can be doing for allof us making our lives easier as
we go about delivering mail andcosts, etc.
(10:51):
And if you partner with me, youwill have a seat at the table
and picking which AI tools thiscompany brings on board, etc.
And he says that changed thecomplexion of the whole
organization Embracing AI.
So it's a bunch of involvedtopics that many different CEOs
are being very Involved in and,given the experience we have
(11:13):
with our vantage point of 20different countries and
thousands of customers, we'redoing everything we can to share
what we've learned over theseyears.
Andrew Grill (11:21):
Lots of different threads, in the end one pack.
And I recall when we hadbreakfast in London I talked
about the fact that the chat dptlaunch in November 2022 was a
watershed moment because itremoved the friction.
My parents in Adelaide,australia, are talking about
chat GPT and I said where didyou hear about that?
Oh, it was on the news.
I stood on stage in Abu Dhabi afew months ago with the CTI of
(11:41):
Amazon and he said we've beenusing AI for 25 years.
Now it's become popular.
Which brings me to the pointthat I think what it's done and
that's why the CEOs are askingyou these questions it's removed
the friction.
A CEO, a C-suite person, canactually play with the tool
without having to learn Pythonor run scripts or set up APIs.
They can very quickly see thepower of something like a
(12:04):
generative AI platform.
But I think what that's alsodone is it's raised the
visibility of this Technology tothe regulators.
What do we do about regulation?
Do we regulate the tech or weregulate the use of the tech?
And how do we get theregulators up to speed to
understand what the power ofthis is and when the guard rails
should be?
Umesh Sashdev (12:23):
if you think about all the time that you and
I have been around.
We've seen the internetrevolution, we've seen the cloud
and the mobile era.
I Don't remember a time whenthe creators of technology,
typically Silicon Valley folks,were the first to come out and
ask for regulation.
Usually, the creators oftechnology are very bullish on
(12:43):
the outcome that technologydelivers and folks in government
, whether it's Washington or 10Downing Street or Paris, etc.
Are the ones pushing forregulation.
This is the first time, to myrecollection, that Silicon
Valley, simultaneous toannouncing the innovation, has
also started to ask for Guardrails and regulation, and we
have to understand why.
(13:04):
There have been many theoriesabout why this is happening.
People want to create, you know, competitive Pressures, etc.
I think the ask for regulationis for a different reason, and
that is for the first time we've, as human kind, created a
technology that is capable ofmaking decisions like the human
(13:26):
brain.
In the past, we've createdtechnology that could automate
repetitive tasks, take away somethings that we used to do,
mimic human behavior, but nevermake decisions like Generative
AI can.
And so the examples tounderstand how profound the
power of this technology is youcould instruct a Generative AI
(13:50):
technology bot like a chat GPD,and there are versions of chat
GPD today which are calledauto-GPT, where you can ask it
to book me a restaurant inOxford Street for two on
Saturday night and I'd like totry an American meal.
Without saying anything more,auto-gpt takes in the
(14:13):
instruction and then startsadding new tasks to its list
until it gets the desiredoutcome.
So it will go to a website likeYelp figure out restaurants
which match the description,then go send emails to that
restaurant asking foravailability.
Once it's confirmed, use thecard on file to go hit a payment
(14:36):
gateway, finish the paymentuntil it finally receives
confirmation and emails theconfirmation back to the actual
requestor, the person who madethat request.
Nobody told it how to go aboutthese tasks.
It discovered these tasks asyou went about it.
Now another version of auto-GPTcame about which is called
(14:58):
Chaos GPD.
So somebody took the exact sameprogram and to show the world
the negative effects of thisprofound technology, they
created Chaos GPD and theinstruction given to it was the
programmer created fivefictitious individuals who don't
exist and said go find allpublicly available records of
(15:19):
these five individuals on theWorld Wide Web, use any means
necessary to hack into theircredentials on the World Wide
Web and, once done, erase allknown records of them from World
Wide Web.
Much like the example of Find Merestaurant, the machine now is
(15:42):
relentless.
It keeps adding tasks to itselfuntil the desired outcome is
reached.
So it's truly important for usto understand that this is a
relentless task machine.
It does not stop until whatit's being asked to do is
actually achieved.
And the thing it's lackingtoday, which is the area of
(16:06):
necessary research for a lot ofAI companies, including Unifor,
is it lacks judgment.
We, as human beings, have beengifted by nature.
Not only can we make decisions,but, as making decisions, we
know this is good, this is bad.
Even if we decide to dosomething bad, we're conscious
about it.
Ai is, today not conscious.
(16:27):
It's not sentient, it lacksjudgment, and that makes it a
relentless task machine.
Therefore, in the wrong hands,it has the potential to create
serious damage to sovereigns, tosecurity apparatuses of
governments, to cyber risks, tous citizens and humankind, which
(16:48):
is why there has been an askfor regulation.
That said, given how fast theinnovation is, regulation is
easier said than done.
How do you regulate somethingthat's evolving every 40 hours
without curtailing the power ofinnovation?
You want a teenager now, likeyou said, your parents in
Australia, a teenager in India.
(17:10):
You want all those folks who arenow beginning to enjoy the
technology to come up with newideas.
You want them to innovate.
At the same time, you want toput guardrails on.
Anyone putting out new modelsshould almost go through a
certification process, and thepharmaceutical industry has
created a model for this, wherenothing comes out in the
(17:33):
pharmaceutical industry withoutclinical trials, blind studies.
So we probably need some formof regulation like that for AI,
where a new model will not hitthe world without some sort of
blind studies and clinicaltrials.
At the same time, unlike pharma, we can't take years for a
model to be approved.
So it's one where I don't envythe role of administrators who
(17:58):
are simultaneously trying tolearn the technology, learn the
implications, come up with verythoughtful approaches of putting
the guardrails, which makes itimperative for us in the
industry to partner with thegovernment to show them what we
know, the power, the pitfalls,etc.
(18:19):
And then help them come up withmeaningful regulation on the
subject.
Andrew Grill (18:24):
I've always said it's got to be a partnership.
I've had a couple ofexperiences where I've been able
to present to the regulators,to the lawmakers of governments
here in the UK, and what I sayto them is you need to think
like a startup, you need topartner yourself with people
generating this technology thisis way before chat, gpt and
understand why they want to inmy term break some rules.
Why do they need some freedomto operate?
(18:46):
And I think when industry andregulators get together you made
a good point the regulatorsneed to almost learn faster than
the developers of thistechnology how it all works, to
be able to have the guardrails.
But generally, a regulatorisn't going to go and work for a
government organization that isan AI expert.
So I think that's a real issue,but probably not something we
can solve.
On the podcast today, I wantedto turn more to your own
(19:09):
solutions.
At Unifor, you've got severalsolutions conversation AI,
emotion AI, generative AI andknowledge AI.
Perhaps you could give us aflavor for each of these and
what problems they solve.
Umesh Sashdev (19:20):
We think of these as Lego blocks of technology
and AI models which we'vearchitected as a platform.
What does that mean?
One of the things we realizedvery early on working with even
our early customers and that hasremained true till day Now that
we have over 1,500 enterprisesusing our technology is that
(19:40):
when we approach any of theseenterprises, they get very
excited about the potential ofAI and we go okay, we're
bringing our full prowess of theplatform.
Please give us access to yourdata, because we don't want to
bring in publicly trained models.
We want to use enterpriseapproved data.
Then we realized that theseenterprises have never arranged
(20:03):
their data in a way that's readyfor AI.
A lot of the data we get areunstructured forms of data FAQ
documents, standard operatingprocedure documents, historic
recordings of calls from thecall center, etc.
Which are all unstructuredforms of data.
(20:24):
Several years ago, we startedinvesting in this field of AI
called Knowledge AI, which takesin all forms of unstructured
data, creates a knowledge graphfrom whatever it's seen in those
documents and those callrecordings and outputs a very
structured form of what itlearned from that data.
(20:44):
We then take this as a pipelineand feed the structured output
into our Generative AI models.
In Generative AI we use allT-shirt sizes.
These models come in a small,medium, large and extra large
size, depending on what we wantthem to achieve.
But the data that's going intoGenerative AI is being filtered
(21:05):
through our Knowledge AI models.
We then discovered, like I wasmentioning in the previous
question, that to add a layer ofjudgment, to add a layer of
understanding of human emotion,our state of mind, empathy,
sarcasm, anger, even especiallyfolks when they're calling into
(21:27):
the call center with thecomplain, and so on and so forth
, it's not enough to haveGenerative AI become a task
machine.
Uniform is also known prettywidely in the academic circles
for the work we've done inEmotion AI, when we've taken the
changes of tone via voice.
We measure using computervision if it's a video meeting,
(21:52):
facial emotions, gestures, bodylanguage, etc.
To train the AI model onunderstanding human emotional
states.
We make all these three formsof AI Knowledge AI, generative
AI, emotion AI work in tandem.
They work in real time.
What do they actually do in theenterprise?
(22:14):
We have four types of benefitsthat enterprise derived from our
technology.
First is in the contact centerswhere, for the most part,
uniforms technology is aco-pilot to the men and women
who work in these call centers.
Dan and I are receiving yourand my calls when we have a
complaint, we have an issue, weneed urgent help.
There's somebody on the otherside of a 1-800 number that
(22:35):
picks up the phone.
Uniforms AI is working as aco-pilot on that call,
understanding the issue that thecustomer is making or
verbalizing, then, throughautomation, finding very quick
answers to that problem.
Even before the human agentcould have put the customer on
hold and tried to find thesolution, the AI finds it and
(22:58):
delivers a great outcome.
The call is shorter, it's moreefficient and it's a better
customer experience.
We, then, have applied thisco-pilot technology in sales
automation.
We found salespeople whenthey're trying to sell to their
customers on Webex and Zoom andTeams meetings.
It's very hard for them to readthe room on this virtual setup.
(23:19):
You have a presentation in themiddle, you have 10 people who
are attending the call.
You can't watch if people arereceiving your message or
they're still confused about thetopic, etc.
Once again, uniforms AI acts asa co-pilot to those salespeople
on these virtual meetings.
We, then, are servicing lots ofgovernment departments.
We're servicing policedepartments, ambulance services
(23:42):
in the UK, nhs etc.
Are using us in their phonelines as an AI technology.
Finally, we have customers intrading desk terminals, folks
who are constantly on theirtrading computer screen along
with phone lines.
A multitude of places where theenterprise has either a phone
(24:04):
conversation or a video meetinggoing on in their frontlines is
all places where Uniformstechnology is aiding and
assisting those departments andthose employees to be more
efficient and deliver bettercustomer experiences.
Andrew Grill (24:19):
You make a great point there.
In my AI talks, I labour thepoint about having data that's
AI ready.
I ask my audiences threequestions what's the data you
want, what's the data you needand what's the data that you'd
like?
What format is in?
Who owns the rights to it?
How do you access it?
I see the audience pausing andgoing.
We hadn't thought about that.
We just thought it was allmagic and that the data we've
(24:41):
got in the company can actuallybe ready to be ingested.
I want to talk about training.
Once you've got the data,you've got to start training it.
Ai training bias is becomingmore evident as these public
large language models aretrained on, I would say,
questionable data.
What do you think should bedone in this area to ensure that
training data is lesssusceptible to bias?
Umesh Sashdev (25:01):
Well, you lead me to a very important subject.
This is one that we'll have tobe very mindful of as long as AI
is going to be in our life andfor posterity.
Ai bias and training bias is avery real threatened issue to
the efficacies of these models.
Once again, we spoke about thelargeness of these data sets
(25:22):
large language models, largemultimodal models.
Those are what are poweringmodern day generated AI
technologies.
When you have billions andbillions of parameters training
a single AI model, very likelythat some undesirable form of
(25:44):
data enters your models.
If it were a consumer app, therisks are you could train it on
pornographic materials.
You could train it on age andappropriate materials.
In the enterprise, the risksare even deeper.
They're about biases towardsdiversity, inclusiveness, etc.
(26:07):
Let me give you an example.
One of the models that Jenniferfocuses on is Emotion AI.
We use computer vision modelson video meetings, on Webex or
Zoom, etc.
To pay attention to humanfacial emotions as being
conveyed by the participants onthat meeting.
Now, as we were training thesemodels, if our researchers did
(26:32):
not pay attention to being veryinclusive in their data sets,
right then to include faces andfacial emotions of people with
different races, genders, skincolor types, size of eyes, etc.
If those variations were nottaken into the training model,
very likely that the model, whenit started to work within the
(26:54):
enterprise, would start to givea biased output when, if it's
encountering a type of face witha skin color type or size of
eye that it was not trained tolook at, it might misread their
emotional state during a meetingand that could cause major
problems, potentially even legalissues.
(27:15):
Similarly, these generative AImodels you're teaching them
language, human language, humancommunication, and if you're not
paying attention, they canstart to pick up, for example,
cus words.
Now, when people are angry,they speak to each other in a
certain way, the AI model startsto learn how to be angry at us,
(27:37):
etc.
And so what is becoming a veryfascinating subject?
As you and I know, the socialmedia companies met up or
Twitter and so on and so forth.
We've known them to have verylarge teams of trust and safety
executives, armies of peoplewhose job was it to try and
(27:57):
catch these undesirable forms ofdata and untrain it from the AI
model, and these concepts wereused in social media companies.
The very same concept is nowcoming into AI companies, where
someone like us or a chat, gptor anyone creating their own
proprietary models now need tohave a trust and safety
(28:19):
department within theorganization whose core job
should be to findimproprietaries, to find
undesirable data that'sinadvertently entering the model
and take it out as soon as it'sfound or cited, so that the
bias is controlled.
So it starts with all of usbeing very conscious.
(28:43):
If we are creating a model, wehave the responsibility to be
very conscious that our modelscould have biases of different
types.
And once you're conscious, wehave to make the investment,
whether people investment ortechnology investment, to make
sure that those biases can beminimized as much as possible.
(29:04):
And yet again, because of thenature of this technology, the
pace of innovation, it's almostimpossible to completely make do
away from it.
But as long as there'sconsciousness on the part of the
provider of the model, manymeasures can be taken to
minimize the risk of biases inthese AI models.
Andrew Grill (29:27):
Gartner's latest hype cycle for AI that's just
been released puts Genitive AIat peak hype no surprises.
So when do you think it willmove to the plateau of
productivity and what will ittake to get there?
Umesh Sashdev (29:38):
If you think about what's happening in
today's world.
Unifor is the first companywith examples of over 1,500
enterprises using us forproductivity gains in their call
centers, et cetera.
Our largest customer, whichhappens to be a financial
services organization in NorthAmerica, has 65,000 of their
(29:58):
call center agents using ourtechnology.
Our second largest customer isa large telecom service provider
and 26,000 of their users useus in their call center, and we
have many such examples.
Why do I bring these up?
You and I both know that whenthese enterprises adopt any new
technology, not just AI, theyfirst start with a small proof
(30:22):
of concept which becomes a pilot, and then a different
department is added to thatpilot.
By the time, a technology ishitting scales of 50,000, 60,000
users, like the examples I justcited, which means the
technology is delivering provenreturn on investment.
It's delivering on businessoutcomes and Unifor is one of
(30:44):
the first companies proving itat its enterprise scale that
this technology is ready forprime time.
It is delivering efficiencygains, et cetera, and we're
going to see an acceleratedtrend in the coming months of
more and more enterprises, moreand more vendors, beginning to
find ways to make it happenwithin the enterprise.
You asked me a question whatwill it take for the technology
(31:07):
to move from the peak of thehype cycle to the plateau of
productivity?
I think here's what it willtake, andrew, from my vantage
point.
I talk to many CEOs who arevery forward looking, who push
us, and we think we're a fastmoving Silicon Valley company.
Some of my customers, whohappen to be CEOs of larger
businesses and traditionalbusinesses, are pushing us to
(31:29):
move even faster.
What I know is some of them whoare captains of industries,
whether it's cybersecurity orhealth care, and these happen to
be publicly listed, publiclytraded companies At least some
of them in the forthcomingquarters will, in their earnings
meetings, earnings results,declare productivity gains
(31:54):
because of their use aggressiveuse of generative AI, and all it
will take is one large companyin each of those segments to set
the tone with the investors.
We saw what happened when asocial media company last year
which was publicly traded, gottaken private by one individual
(32:15):
and then that individual, forright around reasons, decided to
have massive headcountdecreases in that organization.
It forced every other peercompany of theirs In social
media, in the internet arena.
The pressure from Wall Streetwas if one company can do it,
you all should do it and we sawa series of headcount reductions
(32:38):
in layoffs.
Generative AI is actually moreproductive than that.
However, once one CEO, onecompany, shows the potential of
this technology in theirearnings call in one of the
upcoming quarters, wall Streetand investors will be the
biggest force putting pressureon every company to use such
(33:01):
technology to drive productivitygains.
And I think we are at the cuspof it.
So I won't be surprised if, inless than six months from now,
you and I are having a chatabout this topic and we are
saying look how fast we movefrom the peak of interest to
this plateau of productivity inthe hype cycle as we know it.
Andrew Grill (33:22):
So you're in a unique position because your
solutions actually work deeplyinside the organizations you've
mentioned to solve theseproblems.
But for others that don't havethat deep connection, for
example, Microsoft are launchingtheir Microsoft 365 co-pilot AI
, which integrates directly withproducts such as Outlook and
Word right in the daily workflow.
So how can enterprise AI getcloser to the day-to-day
(33:44):
activities we're already doingand become more seamless?
Umesh Sashdev (33:47):
It's really important to reinstate that the
real power of this technology istruly profound.
The outcomes that we canimagine are unbounded.
Every facet of an enterprisejob is likely to benefit, with
productivity gains, automation,efficiency increases, and it
(34:09):
could be as simple as how do wetype emails to our customers?
How do we send emails to ouremployees?
How do we communicate withinthe companies?
How do we communicate with ourcustomers?
Do we communicate with themover phone lines and call
centers?
Do we communicate with them onvideo meetings?
Or do we meet them in person atstores and so on and so forth?
(34:30):
And so the day-to-day activityof every single employee, right
from the CEO to the youngestintern who joined the company,
is, in my opinion, very likelyto be impacted and benefiting
from the user-generated AI.
If they're not already doing it, it's going to happen very
(34:51):
shortly.
Like you said, companies likeMicrosoft, et cetera, who work
on these productivity tools likeemails and other forms of
communication, are bringing thattechnology.
We play in the arena ofcustomer experience, employee
experience tools anywhere in thecompany where there's a phone
conversation, a video meetingand so on.
(35:11):
So far, we are bringing thosetools.
There are many companies whoare bringing generated AI into
back office operations,accounting, billing, legal
documentation and then shuttingdown this whole tailwind like
small companies, the mother ofyour use cases in the enterprise
.
You spoke about enterprises notbeing ready with data.
(35:33):
Data is everywhere except it'snot available when the
enterprises need to access it.
Enterprise search the ability toarrange enterprises data from
any part of the company, whetherit's in some silo in a
different branch office or it'sin a CRM or if it's in the
billing system.
The ability to put it all in asingle data domain.
(35:57):
And then, using Generative AIand all the techniques that
we've spoken of today Knowledge,ai, et cetera make it available
through a natural languagesearch interface.
But imagine the ease, insteadof going into file rooms and
trying to look for an old record, if you could chat with the bot
and say I'm looking for thatold file, which might be a 20
(36:19):
year old record of you know,something we did back then, can
you please find it for me?
And boom, in 60 seconds thatdata element is in front of your
machine or your cell phone asan output of that chat you did
with that interface.
So I repeat, the possibilitiesare unbounded and as we speak
(36:43):
right now, andrew, there aremany different innovators, many
companies who are working ondifferent pieces of this puzzle,
and that is the power thatGenerative AI has given us that
we're moving with innovation ata pace that we have not
experienced in any technologyrevolution in the past.
Andrew Grill (37:00):
You quite rightly say that the real power of
Enterprise AI is being able tofind that needle in the haystack
.
And I think back to chat GPT,the fact that people are saying
we can use this on public data.
And I say to my clients imagineif you could do that with your
own data and the light bulb goesoff.
But to your point before.
When people then scramble andsay, okay, let's do an
Enterprise AI project, I thenwarn them about the cost, that
(37:24):
it's probably a 10X in terms ofcompute power.
The data's got to be in theright place.
You need data governance.
So what would be your advice tosomeone listening to the
podcast that's seen the power ofGenerative AI through a chat
GPT wants to move this into theEnterprise.
Where should they start andwhat are the pitfalls to avoid?
Umesh Sashdev (37:41):
Several of the CEOs have come to me with the
same question.
It's first, it's reallyimportant to get a core group of
people, almost a Tiger team,formulated around the CEO, who
are both empowered and excitedto drive the change, drive the
transformation within theorganization with this
initiative.
(38:01):
It's not just a technologyinitiative, it's a change
management issue within theseenterprises.
It then becomes important topaint a big, hairy or decious
goal, almost a big vision, whichit can excite the whole
workforce to drive the not star,to say, in the next few years,
using AI, we're gonna be acompany, we're gonna be an AI
(38:22):
company, we're gonna be acompany that will be very
efficient, we're gonna be acompany that delivers the best
customer experiences, and so onand so forth.
Then break that vision downinto smaller chunks of projects.
What's achievable in the nextsix months?
What are the low hanging fruits?
There are areas like callcenters and customer service
(38:42):
which are extremely ripe.
The employees there are cryingfor tools.
The employees don't likeputting customers on long holes
only to search for the answer totheir questions.
The customers don't like thatlong experience when they're in
a hurry and all they need is aquick answer to a question of
when's the next slide, etc.
(39:02):
Or why was I overcharged on thebailing?
So there are departments andthere are low hanging fruits of
efficiency gains.
And so, having created a team,having defined a vision, having
found the first high impact buteasy to understand area to
deploy this at example, customerservice it then becomes
(39:23):
imperative for these companiesand CEOs to over communicate at
every step of the way, to overcommunicate with the customers,
with their shareholders and withtheir employees.
In my opinion, it takes six toseven quarters and seven
repetitions of the same messageby the CEO For everyone in the
company and the community tounderstand and buy into the
(39:47):
shift that's been caused you.
And so not only doing it onetime, not only saying we have an
intention, but showing itrepeatedly over six or seven
different quarters, six or sevendifferent earnings, six or
seven different messages, etcetera, is what's needed to land
the message.
And then the flywheel rotatesvery fast If done right.
(40:10):
The idea generation of where toapply this next can be
crowdsourced from within theenterprise.
But for that to happen, it'simportant to first sell the
message really strongly and showand lead by example, show with
a couple of true points in oneor two areas and deliver the
results.
Andrew Grill (40:30):
You say the future of enterprise, ai, is human.
What do you mean by this?
Umesh Sashdev (40:35):
Well, this is the first time that we have the
ability to imagine a world wherewe, as human beings usually we
are told to learn a new skilland when a new technology has
been released, we have toretrain ourselves.
We have to unlearn our pasthabits, retrain ourselves on a
new technology and your tool.
And we are just so.
(40:57):
If you think aboutphilosophically that phenomenon,
that phenomenon is us humanbeings adjusting to technology.
We actually think that tounlock the real power of AI
within enterprises, it willtruly have to be that the
technology adapts to the humanbeings who work in those
(41:19):
enterprises.
That is when we will have thefull buy-in of the human
workforce in every company,every department, everyone
cherishing and, you know, beingenthusiastic about the
technology.
And that is why, as our vision,unifor, we've said the future
(41:41):
of enterprise AI has to be human.
What does that mean?
We spoke a lot about GenerativeAI becoming a relentless task
machine.
A relentless task machine thattoday does not appreciate human
emotions, does not appreciateevery nuance of how human beings
communicate with each other.
(42:01):
You and I don't just send eachother chat messages.
When you and I speak on phoneor a video meeting, we change
our tone.
We can be excited, andrew, haveyou seen this new release by
its own company?
Or we say, hey, that is sodangerous, somebody should be
paying attention.
Just by changing our tone, wecommunicate with each other in
addition to our words.
When we meet in person, like wedid in London and we sat down
(42:24):
for breakfast, or we're meetingon a video meeting, we're
looking at each other and we'reresponding to each other's
energies and facial emotions,and it's truly important that AI
meets us where we are as humanbeings, as opposed to asking us
to adjust our habits.
We want the AI to start withappreciating human emotions
(42:46):
along with the words, along withour language, and then,
hopefully, there'll be a daywhere, using Generative AI, we
can teach AI to also generateits own emotions and thereby
being a great companion, a greatpersonal assistant, whether in
our daily lives or in theenterprises.
And so our vision of EnterpriseAI is human Really means that
(43:09):
we will continue to pushboundaries, push the envelope,
not settle with the fact that AIis a great productivity tool
right now.
But AI, for AI to be acompanion for each of us in the
enterprise or personal lives, ithas to fully understand the
nuances Of human communication.
Andrew Grill (43:27):
We're almost out of time.
We're up to my favorite part ofthe show, the quickfire round,
when we learn more about ourguests iPhone or Android, iphone
Window or aisle.
Umesh Sashdev (43:35):
I'm an aisle person.
Andrew Grill (43:36):
In the room or in the metaverse, always in the
room.
What's the first thing you ask?
Chat GPT.
Umesh Sashdev (43:41):
I ask it about what does it know about me?
Andrew Grill (43:43):
Your biggest hope for this year and next.
Umesh Sashdev (43:45):
That we meaningfully put guardrails
around AI and safely introducethis profound power on citizens,
on employees in different partsof the world.
Andrew Grill (43:54):
I wish that AI could do all of my.
Umesh Sashdev (43:57):
Well, it could follow up all my emails and tell
me the ones that really need myattention.
Andrew Grill (44:00):
The app you use most on your phone.
Umesh Sashdev (44:02):
I'm constantly looking at the weather app.
Andrew Grill (44:04):
The best piece of advice you've ever received.
Umesh Sashdev (44:06):
It was from a mentor of mine who taught me
early on that as important as itis to build a business and lead
a company with IQ, EQ isequally important.
So intelligence and emotionalintelligence go hand in hand.
Andrew Grill (44:20):
What are you reading at the moment?
Umesh Sashdev (44:21):
Well, I'm reading this book.
It's called the Jolt Effect andit's a very fascinating book
for all of us who sell tosomebody, and it's about what
causes a customer's indecisionor the inertia to maintain
status quo, even if they knowthat it's right for them to
(44:42):
adopt something new or change tosomething new.
It's very pertinent in this erawhere we're pushing Generative
AI and a new concept to ourcustomers and you see this every
day where the customer willtell you I'm sold, I'm really
excited, let's move.
And then they go on a radiosilence.
And this book, the Jolt Effect,talks about how do you jolt
(45:07):
your customers if you trulybelieve in the topic how you
jolt them into realizing it'simportant for them to make a
decision now.
Andrew Grill (45:14):
Who should I invite next onto the podcast?
Umesh Sashdev (45:16):
Well, I'm a big fan of your podcast.
You've had many interestingpeople you cover.
Your style is very interestingto me.
You cover a range of topics but, given the world we live in,
there are many fascinatingpeople.
I'd love to hear from Elon Musk.
Would be on top of that list.
Andrew Grill (45:31):
It's not the first time someone suggested that.
So, elon, if you are listening,please return my calls.
How do you want to beremembered?
Umesh Sashdev (45:37):
I want to be remembered for somebody who is
passionate about makingtechnology work to make the
world a better place.
Andrew Grill (45:44):
As this is the actionable futures podcast, what
three actionable things shouldour audience do today to prepare
for a world of enterprise gradeAI?
Umesh Sashdev (45:54):
First, it's really important to understand
any subject and move from beinganxious about it to being very
comfortable and excited about itis to read, to educate yourself
.
Andrew Gill's podcasts are onegreat area to learn about the
subject.
There are many books andinformation out there about
generative AI out there.
(46:15):
So first it's really importantto educate ourselves.
Next, it's important tounderstand our rights data
protection rights, etc.
Even before any new regulationis released.
Most countries give tremendousprotection to us as citizens,
employees and so on and so forth, so it's very important to feel
comfortable that we areprotected.
(46:37):
And third then, just be curious.
Be willing to learn somethingnew every day, be willing to try
a new technology that's put out.
I love the story you just toldme, Andrew, that your parents in
Australia, at a late stage inlife, are willing to be curious,
learn something new and changethe world.
Andrew Grill (47:01):
As you know, my next book is going to be called
Digitally Curious, and I hopepeople are curious enough to
look at it and hear from leaderssuch as yourself, umesh, a
fascinating discussion.
How can we find out more aboutyou and your work?
Umesh Sashdev (47:12):
We try to be very public.
The company I run as CEO isUnifor.
We have a simple website,wwwuniforcom.
On LinkedIn, if you put myfirst and last name, pumir, to
such day, you'll find myLinkedIn.
I'm on Twitter, which is nowcalled X, and my email is
available on several of thesesources, so I'm always open to
(47:36):
listening and receiving messagesfrom people.
If somebody has a question,I'll be happy to hear from them
and be very quick in answeringthose messages.
Andrew Grill (47:45):
Umesh, always a delight to talk to you in person
or online.
Thank you so much for your timetoday.
Umesh Sashdev (47:50):
Andrew, it was a pleasure, thank you.
Intro (48:27):
Thank you.
Popular Podcasts
1. Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
2. Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations.
3. Crime Junkie
If you can never get enough true crime... Congratulations, you’ve found your people.