WEBVTT

00:20.080 --> 00:22.200
Bivek, why act now?

00:22.200 --> 00:23.960
AI has moved so quickly,

00:23.960 --> 00:24.840
it's like a tsunami.

00:24.840 --> 00:27.120
I've lived through so many technology

00:27.120 --> 00:29.720
transformations over the last 25 years.

00:29.720 --> 00:32.120
Cloud, .com before then.

00:32.120 --> 00:32.720
And actually,

00:32.720 --> 00:35.720
nothing like AI's hit the market.

00:35.760 --> 00:38.560
So what we can now do with AI,

00:38.560 --> 00:40.240
with multimodality,

00:40.240 --> 00:41.680
with agenetic framework,

00:41.680 --> 00:43.200
it's just like nothing we've seen before.

00:43.200 --> 00:46.120
Since 1956, we've had traditional

00:46.120 --> 00:48.240
AI machine learning type models.

00:48.240 --> 00:50.280
But with natural language,

00:50.280 --> 00:51.600
there was this click,

00:51.600 --> 00:52.680
you know, in our minds

00:52.680 --> 00:53.800
that we can now

00:53.800 --> 00:55.120
interact with the technology

00:55.120 --> 00:57.000
that we needed to code before.

00:57.000 --> 00:59.120
We're getting a lot more accuracy

00:59.120 --> 01:00.320
out of these models,

01:00.320 --> 01:03.080
so the interaction becomes that much richer

01:03.080 --> 01:05.400
and deeper than it could have ever been,

01:05.400 --> 01:07.160
in the last couple of months.

01:07.640 --> 01:08.720
I love experimenting,

01:08.720 --> 01:10.000
every time a new model comes in,

01:10.000 --> 01:11.640
I'm telling my tech team, get on it.

01:11.640 --> 01:13.440
Let's see what it can do.

01:13.440 --> 01:14.400
It's not good enough

01:14.400 --> 01:15.400
to be on the sidelines,

01:15.400 --> 01:17.040
to be second to market,

01:17.040 --> 01:17.840
third to market,

01:17.840 --> 01:19.080
because entire business models

01:19.080 --> 01:20.040
are going to be disrupted

01:20.040 --> 01:21.000
with the technology.

01:21.000 --> 01:23.520
So again, the why act now,

01:23.520 --> 01:24.720
it's not really a choice.

01:24.720 --> 01:25.800
It's something that people

01:25.800 --> 01:27.360
have to move quickly with.

01:27.360 --> 01:28.960
If you look at what AI can do,

01:28.960 --> 01:30.400
the ability that we have now

01:30.400 --> 01:31.800
to fundamentally change

01:31.800 --> 01:34.200
the way you interact with customers,

01:34.200 --> 01:35.640
the way you deliver services,

01:35.640 --> 01:37.080
the way you deliver products.

01:37.080 --> 01:38.640
I think there's a shift fundamentally

01:38.640 --> 01:39.480
around your risk appetite

01:39.480 --> 01:41.160
as an organisation.

01:41.160 --> 01:42.120
Changing the way

01:42.120 --> 01:43.360
that you look at the market

01:43.360 --> 01:45.480
and you're willing to change yourself.

01:45.480 --> 01:47.120
Finding a different position

01:47.120 --> 01:49.760
for yourself with your customers

01:49.760 --> 01:51.680
and within the ecosystem.

01:51.680 --> 01:53.200
Don't just think about the same

01:53.200 --> 01:55.320
4 or 5 competitors that you've had

01:55.320 --> 01:58.320
for the last 30, 40, 100 years.

01:58.560 --> 02:00.120
Think about the new scale ups

02:00.120 --> 02:01.000
and their thinking with

02:01.000 --> 02:02.760
a customer in mind first.

02:02.760 --> 02:03.600
How can I take out

02:03.600 --> 02:05.040
all the friction points?

02:05.040 --> 02:06.040
They're not thinking,

02:06.040 --> 02:07.440
here's 32 steps

02:07.440 --> 02:08.520
and how do I use AI

02:08.520 --> 02:10.440
to augment those 32 steps?

02:10.440 --> 02:11.280
They're saying, actually,

02:11.280 --> 02:13.360
can we do the entire thing in two steps

02:13.360 --> 02:15.720
and make this an engaged experience?

02:15.720 --> 02:18.280
And they're able to change their technology

02:18.280 --> 02:19.800
to pivot much faster

02:19.800 --> 02:21.640
because they're streamlined

02:21.640 --> 02:22.960
and they're fully digital.

02:22.960 --> 02:25.200
Large organisations are now going to have to look

02:25.200 --> 02:26.280
at the way they operate,

02:26.280 --> 02:27.560
an entirely different way

02:27.560 --> 02:28.920
in an AI landscape.

02:28.920 --> 02:30.200
It's not about protecting

02:30.200 --> 02:31.400
what we do today.

02:31.400 --> 02:33.000
It's evolving and making sure

02:33.000 --> 02:34.240
they're fit for the future,

02:34.240 --> 02:35.880
and they remain not only relevant,

02:35.880 --> 02:36.840
but market leading.

02:41.080 --> 02:42.480
In your experience,

02:42.480 --> 02:42.960
what do you think

02:42.960 --> 02:44.720
the best way is to accelerate

02:44.720 --> 02:46.320
the impact of AI?

02:46.320 --> 02:48.000
I have moved away from

02:48.000 --> 02:49.120
just generative AI.

02:49.120 --> 02:52.320
But actually thinking about agenetic frameworks

02:52.320 --> 02:53.480
and the way that we do it,

02:53.480 --> 02:55.560
is we try to break out the tasks

02:55.560 --> 02:57.160
that are associated with the outcome

02:57.160 --> 02:58.400
that we're trying to drive.

02:58.400 --> 02:59.880
And we're doing that, for example,

02:59.880 --> 03:00.920
in customer services,

03:00.920 --> 03:02.520
where I create different agents.

03:02.520 --> 03:04.760
One agent is a conversational agent,

03:04.760 --> 03:05.880
so it's like you and me,

03:05.880 --> 03:07.320
just having a conversation

03:07.320 --> 03:08.680
and capturing my sentiment.

03:08.680 --> 03:09.840
Then there is another agent

03:09.840 --> 03:12.480
that actually is more of a triage agent.

03:12.480 --> 03:14.640
So trying to pick out the active questions

03:14.640 --> 03:15.840
or queries that I might have,

03:15.840 --> 03:17.440
that then goes and gets routed

03:17.440 --> 03:19.680
to specialist investigation agents.

03:19.680 --> 03:20.840
And all of these agents can be

03:20.840 --> 03:21.840
then linked as well,

03:21.840 --> 03:24.120
for kind of instant decision making.

03:24.120 --> 03:25.680
We recognise this early,

03:25.680 --> 03:26.920
partly why we invested so heavily

03:26.920 --> 03:28.600
in our agent OS framework.

03:28.600 --> 03:29.760
We were one of the first

03:29.760 --> 03:31.520
to make cross-platform

03:31.520 --> 03:33.320
agents talk to one another.

03:33.320 --> 03:36.120
So an Agentforce in Salesforce,

03:36.120 --> 03:38.280
we had an agent for customer servicing,

03:38.280 --> 03:39.880
and then we had IQ,

03:39.960 --> 03:41.440
and they managed to communicate

03:41.440 --> 03:43.160
and exchange information.

03:43.160 --> 03:45.280
That is a massive leap forward.

03:45.280 --> 03:46.200
The interesting thing for me

03:46.200 --> 03:47.200
is when we kind of sit back

03:47.200 --> 03:49.120
and we look at what a genetic frameworks

03:49.120 --> 03:50.120
are going to do in the market,

03:50.120 --> 03:52.960
whether that's recruiting an individual,

03:52.960 --> 03:54.200
whether that's changing

03:54.240 --> 03:55.360
the entire customer journey.

03:55.360 --> 03:56.320
Whether it's changing the way

03:56.320 --> 03:57.320
we deliver health care

03:57.320 --> 03:58.160
and we’ve seen some

03:58.160 --> 03:59.040
of the stats around that.

03:59.040 --> 03:59.840
It can take out

03:59.840 --> 04:02.680
90% plus of the compliance burdens.

04:02.680 --> 04:04.640
And so, social care workers,

04:04.640 --> 04:05.160
healthcare workers can

04:05.160 --> 04:06.400
then focus on people.

04:06.400 --> 04:07.360
Now that's phenomenal.

04:12.400 --> 04:14.160
Bivek, how do we keep up

04:14.160 --> 04:16.400
with AI in a responsible way?

04:16.480 --> 04:17.800
It’s a good question.

04:17.800 --> 04:19.640
Now, we were in a different world

04:19.640 --> 04:20.800
12 months ago

04:20.800 --> 04:22.400
when the use cases we were looking at

04:22.400 --> 04:23.440
were quite straightforward.

04:23.440 --> 04:24.400
What we're now looking at

04:24.400 --> 04:26.200
is these complex workflows,

04:26.200 --> 04:28.840
multiple agents dealing with really complex

04:28.840 --> 04:30.680
business transformation and outcomes.

04:30.680 --> 04:32.720
But what people are not thinking about

04:32.720 --> 04:34.840
is the unintended consequences.

04:34.840 --> 04:36.200
What if the models,

04:36.200 --> 04:36.960
as they're trying to

04:36.960 --> 04:37.760
achieve a certain outcome,

04:37.760 --> 04:38.320
what if they go

04:38.320 --> 04:39.400
into a different direction?

04:39.400 --> 04:41.440
I think these are really tricky things.

04:41.440 --> 04:43.840
When you have so many complex processes

04:43.840 --> 04:45.440
running at the same time,

04:45.440 --> 04:46.800
talking to each other.

04:46.800 --> 04:47.600
It makes me think

04:47.600 --> 04:48.880
a little bit of the game

04:48.880 --> 04:50.200
that we used to have as kids.

04:50.200 --> 04:51.160
Where, you know,

04:51.160 --> 04:52.440
talking into someone's ear

04:52.440 --> 04:53.680
and then you try and propagate

04:53.680 --> 04:55.080
that across the different people.

04:55.080 --> 04:56.240
By the time it gets to the end,

04:56.240 --> 04:58.280
it's not the word that you spoke of,

04:58.280 --> 05:00.960
and it's actually the same kind of concept.

05:00.960 --> 05:03.520
So I think if you take the variability,

05:03.520 --> 05:06.280
the scalability and the security aspect of it.

05:06.280 --> 05:07.200
What we have to do

05:07.200 --> 05:09.400
is we not only testing the scenario,

05:09.400 --> 05:10.520
we're testing the model

05:10.520 --> 05:12.160
under different conditions

05:12.160 --> 05:13.000
and trying to see

05:13.000 --> 05:15.000
whether it still needs the right objective

05:15.000 --> 05:16.240
and the right kind of answer,

05:16.240 --> 05:18.160
but also how does it handle things

05:18.160 --> 05:20.200
that it never was trained on.

05:20.200 --> 05:21.080
And where do you put

05:21.080 --> 05:23.200
the guardrails that are necessary?

05:23.200 --> 05:24.200
These things have to be

05:24.200 --> 05:25.480
baked in from day one.

05:25.480 --> 05:27.240
So it's not about just upfront.

05:27.240 --> 05:30.560
It's continually having this governance built in.

05:30.560 --> 05:32.080
It’s governance as code,

05:32.080 --> 05:34.920
into almost on a daily basis, real time.

05:34.920 --> 05:36.040
The first use case

05:36.040 --> 05:37.760
that I came up with that I thought

05:37.760 --> 05:40.760
was really important is to codify risk.

05:40.920 --> 05:42.800
It's to codify those controls.

05:42.800 --> 05:44.200
So if I was going to take your analogy,

05:44.200 --> 05:44.960
which I really liked,

05:44.960 --> 05:46.040
it's like every time someone

05:46.040 --> 05:47.680
whispers to the next person,

05:47.680 --> 05:48.880
a third party comes along,

05:48.880 --> 05:50.560
just autocorrect a little bit.

05:50.560 --> 05:52.480
So the last person in the chain

05:52.480 --> 05:54.800
has heard exactly what the first person intended.

05:54.800 --> 05:56.960
So certainly the way we're looking at it at PwC,

05:56.960 --> 05:58.360
we've got a responsible AI kit.

05:58.360 --> 06:00.040
There's a huge amount of work

06:00.040 --> 06:01.000
that we've done building

06:01.000 --> 06:02.720
our approach to governance

06:02.720 --> 06:04.280
and implementing it with our clients,

06:04.280 --> 06:05.600
to deal with exactly this.

06:05.600 --> 06:07.040
It's also thinking about,

06:07.040 --> 06:08.560
how does the transformation

06:08.560 --> 06:10.000
of the organisation

06:10.000 --> 06:12.160
responsibly evolve as well.

06:12.160 --> 06:13.920
Which I think is really, really key.

06:13.920 --> 06:15.520
It's first time as a technologist,

06:15.520 --> 06:17.280
that we've been able to solve problems

06:17.280 --> 06:18.640
that we've never even thought

06:18.640 --> 06:19.640
of touching before.

06:19.640 --> 06:20.640
It's phenomenal.