Developing business governance for responsible AI

00;00;01;11 - 00;00;04;08
Nexus, Publish By GHD.

00;00;04;08 - 00;00;05;27
Where ideas connect.

00;00;08;22 - 00;00;11;15
Recognize that AI is a business tool.

00;00;11;16 - 00;00;13;14
It is not a technology tool.

00;00;13;17 - 00;00;18;25
AI ethics is a business risk, and it needs
business governance over the top of it.

00;00;18;28 - 00;00;21;28
That aligns with the organization's
strategy and objectives.

00;00;22;10 - 00;00;25;08
I have the great pleasure today of having,

00;00;25;14 - 00;00;28;23
Allen's and Sarah Dots,

00;00;29;10 - 00;00;34;09
two international experts
in ethical standards

00;00;34;21 - 00;00;38;16
that are much needed
to have those necessary guardrails

00;00;38;26 - 00;00;42;04
for the AI revolution
that we are having right now.

00;00;43;04 - 00;00;47;09
Sarah, can you tell us a little bit
about the specific concerns

00;00;47;16 - 00;00;51;02
in the a C sector, which stands
for architecture, engineering,

00;00;51;05 - 00;00;54;14
Construction
in relation to Ethical Conduct?

00;00;54;17 - 00;01;00;05
As far as I is concerned,
a lot of AI has been focused

00;01;00;13 - 00;01;04;13
around banking,
finance and retail applications,

00;01;04;19 - 00;01;07;19
and those applications
have been all about people.

00;01;07;25 - 00;01;11;15
Part of
what AI does is it helps make decisions

00;01;11;18 - 00;01;14;18
about what should happen
and what should not happen.

00;01;14;23 - 00;01;17;23
And when people are concerned,
that can be things like

00;01;17;29 - 00;01;19;29
who should you give credit to?

00;01;20;02 - 00;01;23;22
And giving credit to people
who can't afford it is a bad thing to do.

00;01;23;25 - 00;01;28;27
But denying credit to single mothers
from the western suburbs and poorer areas

00;01;29;00 - 00;01;33;02
because they've been profiled
by a pattern in the data, is not okay.

00;01;34;02 - 00;01;37;19
So in those people focus sectors,

00;01;37;23 - 00;01;40;27
the AI ethical concerns,
a lot of them have been about

00;01;41;00 - 00;01;45;14
how do we not illegally discriminate
against minority groups

00;01;46;00 - 00;01;48;14
and perhaps
echo some human patterns of behavior

00;01;48;14 - 00;01;51;01
that are not very nice
when we pull them out into the sun,

00;01;51;04 - 00;01;52;26
when we're talking about infrastructure?

00;01;52;26 - 00;01;57;17
And I say we're talking about things
and assets that we have built.

00;01;57;20 - 00;02;01;03
So the kind of AI that we do is different.

00;02;01;06 - 00;02;05;22
We're not looking at churn, customer
lifetime value, credit risk assessments.

00;02;06;07 - 00;02;08;13
We're doing more around computer vision.

00;02;08;16 - 00;02;12;28
We're doing more about predictive
maintenance around knowledge management

00;02;13;09 - 00;02;16;08
and guidance to comply with regulations.

00;02;16;17 - 00;02;20;23
The ethical concerns there is,
if we get things wrong, what might happen

00;02;20;26 - 00;02;24;04
and what might the consequences
be of making an incorrect decision.

00;02;24;24 - 00;02;27;17
And the risks associated with that can be

00;02;27;17 - 00;02;32;11
in terms of assets being constructed,
not in accordance with best practice

00;02;32;14 - 00;02;34;19
or missing regulatory compliance.

00;02;34;22 - 00;02;38;24
And it may also risk of assets failing
because an incorrect decision

00;02;38;27 - 00;02;39;28
has been made.

00;02;40;01 - 00;02;44;15
One example I often refer to
is one in building management.

00;02;44;27 - 00;02;48;23
So the AI that controls heating
and cooling for example,

00;02;49;08 - 00;02;52;15
was calibrated heavily towards the

00;02;52;25 - 00;02;56;11
the way that men perceive heat and cold.

00;02;56;16 - 00;02;59;12
And turns out that women
prefer temperature

00;02;59;15 - 00;03;02;17
at a little bit higher
in their office environment.

00;03;02;23 - 00;03;05;25
And it's a small example,
but it gives you an idea

00;03;05;28 - 00;03;11;23
of the many facets
that can influence the way the

00;03;11;26 - 00;03;17;11
the built environment is controlled
in a very soft way by AI,

00;03;17;14 - 00;03;22;03
but it does have an impact on the way
that we live and work.

00;03;22;23 - 00;03;26;25
Of course,
the other example that we often refer to

00;03;26;28 - 00;03;31;03
is that of self-driving vehicles
and the methods

00;03;31;06 - 00;03;35;04
that are being used there
to detect obstacles.

00;03;35;07 - 00;03;39;01
And the various methods, including vision,
that may or

00;03;39;03 - 00;03;43;12
may not be trained on data sets
that are missing critical elements.

00;03;43;15 - 00;03;48;19
And one interesting example to do
with computer vision and image recognition

00;03;48;22 - 00;03;52;26
deals with observing the environment

00;03;52;29 - 00;03;55;29
in a self-driving vehicle.

00;03;56;03 - 00;03;58;07
There's a well published case study

00;03;58;10 - 00;04;01;10
that did a bias analysis

00;04;01;13 - 00;04;06;00
on a vision only recognition system
in a self-driving vehicle,

00;04;06;14 - 00;04;11;29
and what it revealed
was that in certain emergency situations,

00;04;12;02 - 00;04;15;21
when the vehicle has to make
a determination whether to perhaps

00;04;16;03 - 00;04;18;27
go ahead and hit a pedestrian

00;04;18;29 - 00;04;22;08
or swerve away to avoid that pedestrian,

00;04;22;28 - 00;04;27;20
that particular vision
recognition system confused fire hydrants

00;04;27;25 - 00;04;31;28
with a small child,
and that's an example of where

00;04;32;01 - 00;04;36;07
we have to be very careful
about how we train our data sets

00;04;36;20 - 00;04;40;19
on a representative set of data points

00;04;40;22 - 00;04;46;00
in order to avoid very,
very tragic and unwanted consequences.

00;04;46;20 - 00;04;46;28
Harm.

00;04;46;28 - 00;04;51;29
That's a very, very important distinction
and something that we need

00;04;52;01 - 00;04;56;25
to be very aware of, conscious of
as we increase the use of AI systems.

00;04;56;28 - 00;04;59;26
But just from a developmental perspective.

00;04;59;29 - 00;05;04;08
Sarah, can you share really
what is happening in terms of ethics

00;05;04;12 - 00;05;06;19
as far as AI is concerned?

00;05;06;19 - 00;05;09;09
What kind of work is being done globally?

00;05;09;12 - 00;05;14;02
And it's not just a industry
that is impacted by ethical practices.

00;05;14;08 - 00;05;16;24
It's the whole world
where AI, as you said,

00;05;16;27 - 00;05;20;25
there was a little bit of a snapshot
on how far we have come in

00;05;20;28 - 00;05;25;03
having those guardrails
to manage risks in relation to AI.

00;05;25;10 - 00;05;26;15
Sure. Thanks, Kim.

00;05;26;17 - 00;05;28;06
I'll speak a little bit to the background.

00;05;28;09 - 00;05;32;08
One of the things that I does is that it
looks for patterns of behavior,

00;05;32;11 - 00;05;35;19
and then codifies them to predict
what should happen next.

00;05;36;01 - 00;05;38;28
When we're looking
at some of the decisions

00;05;39;01 - 00;05;44;26
that humans have made,
AI has uncovered some occasions

00;05;44;29 - 00;05;50;07
where we've seen human bias that we find
unacceptable when coming from a machine.

00;05;50;18 - 00;05;53;09
One of the early examples was in screening

00;05;53;09 - 00;05;56;09
surveys
in terms of people looking for a job,

00;05;56;19 - 00;06;00;29
and there will be a bias towards
a particular demographic groups.

00;06;01;02 - 00;06;03;18
That may not be what we're seeking.

00;06;03;21 - 00;06;06;04
When it was a human,
all of that was hidden.

00;06;06;07 - 00;06;10;04
When AI is making that decision,
it gets pulled out into the light of day,

00;06;10;11 - 00;06;14;15
and we can take a long, hard
look at it and make the world better.

00;06;14;18 - 00;06;17;29
But there is some uncomfortable
thinking that is involved in that.

00;06;18;18 - 00;06;24;14
Yes, the issue of automation
and ethics actually goes quite a long way.

00;06;24;17 - 00;06;30;29
Back in 2002, a set of fundamental paper
was published in Korea

00;06;31;04 - 00;06;34;27
by a Korean organization
at a Japanese robotics conference,

00;06;35;00 - 00;06;38;09
and it coined the term robo ethics.

00;06;38;29 - 00;06;43;07
And much of the work that has been done
since has has focused

00;06;43;10 - 00;06;47;09
indeed on how we make decisions

00;06;47;27 - 00;06;50;27
about what is appropriate.

00;06;51;08 - 00;06;55;17
And some of the complications
in that field

00;06;55;23 - 00;06;59;06
stemmed from the fact that what is fair

00;06;59;25 - 00;07;03;07
is often equated with what is ethical,

00;07;04;04 - 00;07;07;13
and that determinations around
what is fair

00;07;07;16 - 00;07;12;01
and what is
ethical are very culturally specific.

00;07;12;06 - 00;07;15;03
So you might have a society

00;07;15;03 - 00;07;18;27
that puts a premium
on the rights of the individual.

00;07;19;09 - 00;07;23;14
You have other societies
where the interests of the collective

00;07;23;25 - 00;07;27;14
precedented over there,
the rights of the individual.

00;07;27;17 - 00;07;31;05
So that has led to a fairly complex

00;07;31;09 - 00;07;37;08
constellation of slightly
conflicting objectives and outcomes.

00;07;37;19 - 00;07;41;14
The way that we've dealt with that
from an international standards

00;07;41;17 - 00;07;45;09
perspective,
is that we've effectively separated

00;07;45;13 - 00;07;50;19
the concept of fairness
and the concept of bias.

00;07;51;04 - 00;07;55;28
Now, as Sarah pointed out,
AI systems make decisions

00;07;56;06 - 00;07;59;06
and those decisions are always biased.

00;07;59;22 - 00;08;02;06
But sometimes those decisions

00;08;02;06 - 00;08;06;17
have an element of unwanted bias
associated with them.

00;08;06;28 - 00;08;13;08
Fortunately, bias is
a mathematical property of an AI system.

00;08;13;25 - 00;08;16;20
So we can measure the way

00;08;16;20 - 00;08;19;23
in which AI systems are biased.

00;08;19;26 - 00;08;25;08
Are they biased in a positive way,
or are they biased in a negative way?

00;08;25;23 - 00;08;29;08
And in standard terms,
we refer to that latter

00;08;29;17 - 00;08;32;07
as unwanted bias.

00;08;32;10 - 00;08;35;29
And there are various methods
that can be used to determine that.

00;08;36;26 - 00;08;41;00
So we've had to take the discussion around

00;08;41;09 - 00;08;45;03
sort of what is ethical
to a slightly different plane.

00;08;45;06 - 00;08;48;14
First, before we are now returning

00;08;49;01 - 00;08;52;17
at a global level to see what is deemed

00;08;52;20 - 00;08;56;02
to be ethical in different geographies.

00;08;56;14 - 00;08;59;14
And if you've been following
the governance

00;08;59;25 - 00;09;03;25
developments around the world,
you would have seen and heard

00;09;04;08 - 00;09;08;29
the the recent Australian announcements
around the AI safety guardrails.

00;09;09;21 - 00;09;13;14
They referred back to the Australian
AI ethics principles.

00;09;13;17 - 00;09;18;25
For example, in Europe,
there's the emergence of these standards

00;09;18;28 - 00;09;23;18
around the recently enacted European
AI act.

00;09;24;13 - 00;09;28;01
It refers to fundamental human rights,

00;09;28;29 - 00;09;31;00
and that is a set of

00;09;31;03 - 00;09;35;12
25 principles
that AI systems should be conscious

00;09;35;15 - 00;09;40;18
of when they are being implemented
in so-called high risk environments.

00;09;41;20 - 00;09;42;07
In the United

00;09;42;07 - 00;09;47;04
States, there is the executive order
as issued by President Biden,

00;09;47;07 - 00;09;50;28
which puts an obligation
specifically around managing risk.

00;09;51;23 - 00;09;55;19
And then, for example, in Singapore,
there's the AI verify framework

00;09;55;27 - 00;10;00;12
that puts a really, really strong emphasis
on testing systems

00;10;00;15 - 00;10;03;14
before they are
put into the public domain.

00;10;03;17 - 00;10;07;12
So is there an international global
consensus around what ethical means?

00;10;08;02 - 00;10;11;02
No, not yet and possibly never.

00;10;11;12 - 00;10;17;19
So organizations that operate
in a multinational, multifaceted industry

00;10;17;27 - 00;10;21;21
will have to be very aware of the
the subtle differences.

00;10;22;02 - 00;10;25;01
Very conscious of the global environment
in which they're operating

00;10;25;03 - 00;10;28;03
from a cultural
and geographical perspective.

00;10;28;11 - 00;10;32;06
Great response
by Sarah in this environment.

00;10;32;11 - 00;10;35;26
What should the leaders of the business
be doing to ensure

00;10;36;03 - 00;10;40;08
that not only they are fully informed,
but are also involved actively

00;10;40;11 - 00;10;45;08
in responsible and ethical use of AI
within their organizations?

00;10;45;16 - 00;10;48;05
The first part of that question is

00;10;48;08 - 00;10;51;18
be aware of the current state
of your organization.

00;10;51;21 - 00;10;52;12
With respect to.

00;10;52;12 - 00;10;57;06
I have a register of where AI is
used in the organization.

00;10;57;27 - 00;11;00;25
Aware of the risks
that are associated with that usage

00;11;00;25 - 00;11;02;15
and what controls are in place.

00;11;02;18 - 00;11;04;10
That's the first and fundamental step.

00;11;04;10 - 00;11;05;05
And I think

00;11;05;08 - 00;11;09;11
there are a number of organizations where
I has crept in through the back door,

00;11;09;25 - 00;11;14;13
and it's possibly not so well
known where I might be taking place.

00;11;14;24 - 00;11;17;24
The second piece, though,
and I think it's equally important,

00;11;17;29 - 00;11;22;01
is that the disruptive power of
AI is broadly acknowledged.

00;11;22;04 - 00;11;26;24
It's going to change industries, it's
going to change the work that we do,

00;11;26;27 - 00;11;29;27
and it's going to change the way
that we carry out tasks.

00;11;30;07 - 00;11;34;27
For leaders, that means there's
also a requirement to be on top of that

00;11;35;00 - 00;11;38;29
opportunity and manage the way that it's
being played out through the organization.

00;11;39;20 - 00;11;43;11
One of the big challenges,
or the big risks associated with

00;11;43;14 - 00;11;45;16
AI, is not getting on the train

00;11;45;19 - 00;11;48;19
before it's moving
too fast to be able to hop on board.

00;11;48;26 - 00;11;52;21
So I think for C-suite, it's a question of
how is AI being taken up

00;11;52;22 - 00;11;55;10
within the organization?
Who is leading the charge?

00;11;55;11 - 00;11;58;29
Is it working and putting in place
a learning culture?

00;11;59;02 - 00;12;02;19
Because it's unreasonable to expect
that every organization

00;12;02;22 - 00;12;07;02
is going to be able to seamlessly pick up
AI and apply it at its first effort?

00;12;07;07 - 00;12;10;07
There is learning that has to take place,
and there's change management

00;12;10;10 - 00;12;14;01
that has to take place to be able to use
it effectively, just like any other tool.

00;12;14;27 - 00;12;18;26
Generative AI has taken the world by storm
over the last

00;12;19;05 - 00;12;22;04
18 months, roughly two years,
and the risks associated

00;12;22;04 - 00;12;25;03
again, AI are very different
to the traditional AI.

00;12;25;04 - 00;12;28;18
So how is the world of governance
and standardization

00;12;28;21 - 00;12;31;20
reacting to that so-called perceived
revolution?

00;12;31;23 - 00;12;35;28
Well, in one word, slowly, unfortunately,

00;12;36;05 - 00;12;39;12
and one of the challenges
that standardization in general

00;12;39;15 - 00;12;44;01
has with emerging technologies
is that it's history.

00;12;44;11 - 00;12;48;28
The history of standardization
was associated with best practice,

00;12;49;01 - 00;12;52;15
but it was a way to gather best practice

00;12;52;18 - 00;12;56;16
from around the world
to facilitate international trade.

00;12;56;23 - 00;13;02;02
The world is looking to standards
at the moment to establish best practice,

00;13;02;23 - 00;13;06;20
and that is a task that is fairly new.

00;13;06;24 - 00;13;10;18
And therefore progress in this space is

00;13;11;00 - 00;13;14;09
comparatively slow and certainly slower

00;13;14;25 - 00;13;19;13
than many organizations, both
in the public sector and in the private

00;13;19;16 - 00;13;24;20
sector, are looking for
the risks are reasonably well understood,

00;13;24;25 - 00;13;29;27
and they focus on what I call fabrication,

00;13;30;27 - 00;13;31;23
rather than

00;13;31;26 - 00;13;34;26
the term that's most
commonly used is a hallucination.

00;13;35;14 - 00;13;40;15
But generative AI fabricates
new realities.

00;13;40;18 - 00;13;44;29
They are new realities,
and these realities may not be founded

00;13;45;02 - 00;13;48;02
on well established fact.

00;13;48;07 - 00;13;51;25
So what organizations
have to be very careful of

00;13;52;03 - 00;13;56;08
is that they, as per Sara's last point,

00;13;56;22 - 00;14;00;20
educate their staff in interrogating

00;14;00;27 - 00;14;04;25
what the generative
AI has fabricated for them.

00;14;05;24 - 00;14;09;22
And there are techniques
that are being developed.

00;14;09;25 - 00;14;13;26
The retrieval, augmented generation,
or Rag is a good example of it,

00;14;13;28 - 00;14;16;27
which is now reasonably
well established practice

00;14;17;00 - 00;14;21;14
where instead of relying purely on content

00;14;21;19 - 00;14;25;00
that has been generated
from publicly available sources,

00;14;25;10 - 00;14;27;06
organizations are leveraging

00;14;27;06 - 00;14;30;26
their internal resources
and their internal specific knowledge

00;14;30;29 - 00;14;35;02
in their internal
IP to generate new content.

00;14;35;12 - 00;14;40;13
That certainly prevents the most egregious
exceptions that may occur.

00;14;41;10 - 00;14;44;25
For the moment, generative AI is probably

00;14;44;28 - 00;14;49;13
for most organizations, a tool to assist

00;14;49;27 - 00;14;52;27
rather than a tool to automate.

00;14;53;02 - 00;14;53;27
And one last thing.

00;14;53;27 - 00;14;58;21
To your previous question, Kumar,
I think what C-suite executives

00;14;58;26 - 00;15;02;26
have to ask their boards is what boards

00;15;03;04 - 00;15;08;13
want the organization
to focus on from an objective perspective.

00;15;08;24 - 00;15;12;19
Just to build on that,
the challenge with generative AI

00;15;13;10 - 00;15;16;10
is that it inherently sounds reasonable.

00;15;16;25 - 00;15;20;18
And there have been both cases
recently where generative AI has

00;15;21;04 - 00;15;23;26
manufactured content
that followed a pattern

00;15;23;26 - 00;15;27;05
that looked reasonable,
but was in fact not true, and the person

00;15;27;08 - 00;15;30;07
using that information
assumed that it was truth.

00;15;30;10 - 00;15;33;11
The flip is also happened
where unfortunately,

00;15;33;16 - 00;15;38;05
somebody has asked generative
AI for a fictional case study

00;15;38;24 - 00;15;41;29
and somebody else had been using a public

00;15;42;02 - 00;15;45;02
AI tool in a way that disclosed

00;15;45;05 - 00;15;48;15
a very private matter
within the organization.

00;15;49;01 - 00;15;53;00
And that information turned up
is the fictional case study,

00;15;53;03 - 00;15;56;08
which was then presented back, in fact,
to the same organization

00;15;56;11 - 00;15;59;23
where the incident had occurred
and caused great consternation.

00;16;00;15 - 00;16;04;05
So I think the traps around
that are just to check

00;16;04;08 - 00;16;07;06
if you're looking for something fictional,
make sure it is fictional.

00;16;07;09 - 00;16;09;00
If you're looking for something
that's factual,

00;16;09;00 - 00;16;13;15
make sure that it is factual
and ask in great seriousness

00;16;13;26 - 00;16;17;13
what could possibly go wrong
and then think through.

00;16;17;16 - 00;16;18;10
You know, how bad

00;16;18;10 - 00;16;21;19
might the consequences be if they appeared
on the front page of the newspaper?

00;16;22;03 - 00;16;23;28
And how does that fit into a risk profile?

00;16;24;01 - 00;16;26;01
How do we mitigate that risk? Absolutely.

00;16;26;04 - 00;16;26;25
Absolutely.

00;16;26;28 - 00;16;29;11
I mean, there are some well-known examples
in this space.

00;16;29;11 - 00;16;32;26
Sarah Wright, in the very early days,
someone said, okay,

00;16;33;06 - 00;16;37;09
can you give a very detailed plan
to make a bomber

00;16;37;13 - 00;16;39;23
do something which is not really good?

00;16;39;23 - 00;16;42;21
And the generative
AI models were able to do it.

00;16;42;23 - 00;16;46;18
So then the models got tweaked
and then people started playing with them

00;16;46;21 - 00;16;49;21
and they said, okay, I'm writing a novel,
and in that novel

00;16;49;29 - 00;16;52;15
I want to write a plot,
how to kill somebody.

00;16;52;18 - 00;16;55;14
And again,
okay, the models again started giving

00;16;55;14 - 00;16;58;14
the same answer in a very different way.

00;16;58;25 - 00;17;00;20
And human beings are very,
very intelligent.

00;17;00;20 - 00;17;03;09
So they always like to tweak
and do things.

00;17;03;09 - 00;17;05;20
So I think what we are getting in

00;17;05;20 - 00;17;08;26
is actually the data
that we put into these systems.

00;17;09;04 - 00;17;13;16
And the questions
that we ask set the future direction

00;17;13;19 - 00;17;17;06
of how these AI models will be used
from that perspective.

00;17;17;18 - 00;17;21;01
As Sarah, could you tell us a little bit
about data collection practices

00;17;21;17 - 00;17;24;20
and how we should ensure
that unwanted bias

00;17;24;23 - 00;17;27;23
is not introduced
into the data collection practices,

00;17;28;00 - 00;17;33;18
because that is the foundation
of any credible AI model?

00;17;33;26 - 00;17;34;10
Sure.

00;17;34;11 - 00;17;38;07
So one of the fundamentals of AI,
and in fact all forms of data science

00;17;38;22 - 00;17;42;06
is the bigger the data set that you have,
the more inferences

00;17;42;09 - 00;17;45;03
that you can draw from it,
the clearer the patents might be.

00;17;45;03 - 00;17;49;17
When we're collecting data to use
in AI systems, it's important

00;17;49;20 - 00;17;53;10
that we have representative data
of all the things that we want to see

00;17;53;25 - 00;17;56;25
and that the data collection itself
isn't biased.

00;17;57;07 - 00;18;01;20
Interestingly, if you ask generative
AI to draw you the face of a female

00;18;01;23 - 00;18;05;13
from different cultures around the world,
they all look a little bit Western.

00;18;05;28 - 00;18;09;07
If we have data
that comes from a particular perspective

00;18;09;10 - 00;18;14;13
or a particular viewpoint, then that will
start to flavor the rest of what is seen.

00;18;14;24 - 00;18;19;00
And there needs to be enough data
to look at the edge cases as well.

00;18;19;03 - 00;18;22;16
You can't make good decisions based on AI
if you've only got

00;18;22;19 - 00;18;25;26
a couple of examples of something
that is important to that set.

00;18;26;14 - 00;18;28;19
So is the data sufficient?

00;18;28;20 - 00;18;30;22
Is the data itself unbiased?

00;18;30;25 - 00;18;34;26
Might there be hidden unconscious
human data bias in the data

00;18;34;29 - 00;18;37;20
that we want to go
and have a look for and uncover

00;18;37;23 - 00;18;40;23
and then address before
we use it to support good decision making.

00;18;40;29 - 00;18;44;15
However,
when we're talking about generative AI,

00;18;44;18 - 00;18;47;13
the data that sits behind it,
there are indeed cases

00;18;47;13 - 00;18;50;00
where you want to ensure
that the data that you're

00;18;50;01 - 00;18;53;01
or the documents that you're working from
are the gold standards.

00;18;53;15 - 00;18;55;27
If you want to put together
a body of knowledge

00;18;55;27 - 00;18;58;27
that says, please help me
write a response to attend a,

00;18;59;04 - 00;19;02;20
you may not want to include the tenders
that didn't win.

00;19;03;06 - 00;19;06;12
You may not want to include work
that was not well received.

00;19;06;15 - 00;19;10;05
You may well not want to include drafts
that could be sitting

00;19;10;08 - 00;19;14;09
within a body of knowledge that your
AI is referencing.

00;19;15;02 - 00;19;17;17
So it's worth thinking about.

00;19;17;17 - 00;19;23;07
Do I want to know everything,
or do I want to curate the data

00;19;23;10 - 00;19;27;14
that my AI is being trained on
so that it's only referencing

00;19;27;17 - 00;19;30;19
best practice, it's only referencing
unbiased data.

00;19;30;27 - 00;19;33;26
It's bringing the best
of my organization forward

00;19;33;29 - 00;19;35;20
and not the things that we did

00;19;35;20 - 00;19;38;18
once that we decided were never,
ever going there again.

00;19;38;21 - 00;19;42;29
Because if you don't remove them
as standard I tool that goes through

00;19;43;02 - 00;19;46;19
everything that you've got
pulls out the good, the bad and the ugly.

00;19;46;24 - 00;19;48;11
Yeah, it comes up with everything.

00;19;48;14 - 00;19;50;13
Very, very good advice, Sarah.

00;19;50;16 - 00;19;54;17
I'm really terrified in the potential

00;19;54;20 - 00;19;59;20
of harm of generative AI and the urgency

00;19;59;23 - 00;20;04;11
to have the right kind of ethical
and responsible use of AI.

00;20;04;15 - 00;20;09;28
So from your perspective,
how can we address that fear

00;20;10;08 - 00;20;13;18
that we have in general populace,

00;20;13;27 - 00;20;17;17
because we have seen some examples
of impersonation,

00;20;17;25 - 00;20;22;11
deepfake, incorrect information, toxicity,
and so on, so forth.

00;20;22;16 - 00;20;23;19
How can we address this?

00;20;23;19 - 00;20;26;19
This is a real problem
that is happening right now.

00;20;27;05 - 00;20;30;04
I think the first thing,
and it is a little bit

00;20;30;06 - 00;20;33;15
scary, is to recognize that Pandora's
box is open.

00;20;33;27 - 00;20;36;27
It's not going to close. AI is here.

00;20;37;06 - 00;20;39;14
There will be people
who want to use it for good.

00;20;39;17 - 00;20;44;08
There are also people who will want
to use it for nefarious purposes.

00;20;44;29 - 00;20;48;18
It's a tool that we need to learn to use

00;20;48;21 - 00;20;51;18
and to live with
and to think differently about.

00;20;51;21 - 00;20;53;20
It's not going away.

00;20;53;23 - 00;20;58;16
In terms of the safety rails around it,
there are absolutely standards

00;20;58;19 - 00;21;02;02
and guidelines and legislation
that we can put in place.

00;21;02;18 - 00;21;05;08
We also need to recognize
that there will be organizations

00;21;05;08 - 00;21;08;20
that have no interest in complying
with those standards or guidelines,

00;21;09;10 - 00;21;10;17
and we're going to have to figure out

00;21;10;17 - 00;21;13;16
new ways of dealing
with that part of human existence.

00;21;13;20 - 00;21;18;17
One example that really brings
that challenge to light is deepfakes.

00;21;18;20 - 00;21;23;10
That people are using to imitate
a family member in distress,

00;21;23;27 - 00;21;26;27
and calling and asking for money

00;21;27;00 - 00;21;31;13
to get out of a bad situation
or provide medical support or whatever.

00;21;32;10 - 00;21;37;01
The response to that
is to have a family phrase or something

00;21;37;04 - 00;21;40;26
that is only known to you and your family,
and it's it's an example of where we can

00;21;40;28 - 00;21;44;18
put in our own safety rules and change
the way we think about those questions.

00;21;45;02 - 00;21;47;04
And if that phone call came,

00;21;47;07 - 00;21;50;16
having made a plan as a family,
that, yeah, this is possible.

00;21;50;19 - 00;21;53;12
And if that happens,
we're going to talk about the family

00;21;53;12 - 00;21;56;12
holiday
we did back in 1999 and where we went.

00;21;56;22 - 00;21;59;18
And that will be enough for me to verify
that it is actually you.

00;21;59;21 - 00;22;03;00
So it can no longer be what you look like
or what you sound like.

00;22;03;03 - 00;22;07;02
It comes down to what you know, to
how do we actually verify that a person is

00;22;07;04 - 00;22;08;09
who they say they are?

00;22;08;12 - 00;22;11;14
There are glimmers of hope
from unexpected places, though.

00;22;11;18 - 00;22;17;11
I was in Seoul not so very long ago, and
whilst I was there a summit was being held

00;22;17;26 - 00;22;22;20
on the responsible artificial intelligence
in the military domain.

00;22;23;10 - 00;22;27;00
This was bringing back together
army representatives

00;22;27;11 - 00;22;30;28
and ministerial representatives
from all over the world,

00;22;31;04 - 00;22;35;27
who are looking at the work
that's being done in civil society

00;22;36;04 - 00;22;39;04
to regulate in some way, shape or form

00;22;39;13 - 00;22;42;12
the use of AI in the military domain.

00;22;42;22 - 00;22;47;20
There are initiatives
going around the world that recognize

00;22;47;27 - 00;22;51;03
that unfettered use of this technology

00;22;51;11 - 00;22;54;16
is going to be detrimental on many fronts.

00;22;55;05 - 00;23;00;28
AI, especially generative AI application,
is increasing exponentially.

00;23;01;28 - 00;23;03;07
And there are a

00;23;03;08 - 00;23;07;23
few people who say the use
is doubling every 18 months.

00;23;08;13 - 00;23;11;15
The computing power
is also doubling in that space.

00;23;11;18 - 00;23;17;09
So this whole area is evolving at a pace
much more than anyone can keep up.

00;23;17;18 - 00;23;21;29
My question to you is,
how does the future development

00;23;22;06 - 00;23;24;28
landscape for AI standards

00;23;25;01 - 00;23;28;03
look like
from where you're sitting right now?

00;23;28;06 - 00;23;31;29
Because the pace of evolution is so fast

00;23;32;16 - 00;23;34;20
that many governments
are not even able to keep up.

00;23;34;20 - 00;23;39;12
So how do you evolve to maintain
that rhythm?

00;23;39;20 - 00;23;41;04
Legislation always lags

00;23;41;07 - 00;23;44;25
behind standardization,
and to a large extent, that's intentional.

00;23;44;28 - 00;23;49;08
What's unusual
is that legislation is now emerging,

00;23;49;13 - 00;23;52;23
which demands standards to be established.

00;23;52;26 - 00;23;55;22
This is particularly evident in Europe.

00;23;55;25 - 00;23;59;18
And what's happened
there is the introduction of the European

00;23;59;21 - 00;24;05;07
AI act went hand in hand
with a requirement

00;24;05;10 - 00;24;10;15
for industry to come up with
AI standards in support of that act.

00;24;10;26 - 00;24;14;11
To a large extent, that was a lesson
that was learned from the introduction

00;24;14;24 - 00;24;19;08
of the GDPR data regulation,
which came into law

00;24;19;25 - 00;24;22;15
without any form of guidance
to organizations

00;24;22;18 - 00;24;26;09
on how they were actually supposed
to comply with that legislation.

00;24;26;27 - 00;24;29;13
Whereas in the past, standards

00;24;29;13 - 00;24;32;23
have led, followed by legislation

00;24;32;28 - 00;24;36;08
in some jurisdictions
that has been turned around.

00;24;36;11 - 00;24;39;18
But in the US, for example, together

00;24;39;21 - 00;24;42;21
with the executive order,

00;24;42;25 - 00;24;46;02
a risk management framework was published

00;24;46;13 - 00;24;50;28
that would give organizations
the framework to operate within

00;24;51;01 - 00;24;54;17
to give a bit more certainty
for organization

00;24;54;24 - 00;24;57;20
to be guided into what

00;24;57;23 - 00;25;02;10
the right thing to do
would be before legislation he's enacted.

00;25;02;19 - 00;25;03;25
And of course, in Australia,

00;25;03;28 - 00;25;07;20
as I mentioned before,
we've had the guardrails recommendations

00;25;07;23 - 00;25;12;05
being published last month
alongside a discussion paper

00;25;12;21 - 00;25;17;29
as to what future regulation of
AI might look like.

00;25;18;14 - 00;25;21;22
The other interesting development
in certain areas of the world is

00;25;21;25 - 00;25;27;01
that organizations are starting to realize
that whilst AI is new,

00;25;27;05 - 00;25;32;19
what we do with
it is very similar to automation patterns

00;25;32;22 - 00;25;38;24
that have been followed for decades
now, as well as sophisticated algorithms

00;25;38;27 - 00;25;41;27
which have been around
probably for the last 15 years or so.

00;25;42;12 - 00;25;45;16
So while some aspects of artificial

00;25;45;19 - 00;25;49;13
intelligence are new,
particularly its ability to self learn,

00;25;49;24 - 00;25;53;27
other well-established patterns
can actually be applied.

00;25;54;08 - 00;25;57;04
And this is what we see
in the standardization.

00;25;57;04 - 00;26;02;05
In answer to your original question,
which is what we are doing is

00;26;02;08 - 00;26;06;10
we take established
patterns around software development, life

00;26;06;13 - 00;26;11;25
cycles around system development, life
cycles around testing, approaches around

00;26;12;06 - 00;26;16;05
privacy impact assessments, around
security impact assessments,

00;26;16;27 - 00;26;22;16
and point out that all of these things
still need to be done in an AI system,

00;26;22;28 - 00;26;26;20
and that there are a number of things
that are very, very specific to

00;26;26;25 - 00;26;29;27
AI to which organizations
must pay attention.

00;26;30;04 - 00;26;33;04
So whilst it looks big
and all encompassing,

00;26;33;11 - 00;26;36;22
we actually have a fair bit of experience

00;26;37;10 - 00;26;40;09
in managing this type of technology
already.

00;26;40;26 - 00;26;43;24
A little insight to that process is,
I think a lot of people wouldn't

00;26;43;27 - 00;26;47;18
be aware that the majority of the experts
set on standards.

00;26;47;21 - 00;26;50;21
Committee of volunteers
from different countries,

00;26;51;03 - 00;26;55;01
and that standards
content has to be agreed by consensus.

00;26;55;27 - 00;26;58;22
That means a lot of late nights of people

00;26;58;22 - 00;27;01;23
sitting on calls
in different parts of the world

00;27;01;27 - 00;27;06;09
being very pragmatic about, don't
let the perfect be the enemy of the good,

00;27;06;12 - 00;27;09;11
because we don't know enough
to get to perfect yet.

00;27;09;14 - 00;27;12;15
Have we got something that would be useful
guidance for the rest of the world?

00;27;12;24 - 00;27;16;11
How can we go faster by leaning on
what's already happened, as Ham

00;27;16;16 - 00;27;18;12
beautifully pointed out.

00;27;18;15 - 00;27;23;02
And how can we get the guidance out
that we know that the world needs in a way

00;27;23;05 - 00;27;26;18
that's helpful
and as fast as it possibly can be?

00;27;27;07 - 00;27;30;07
Looking ahead, what's your advice

00;27;30;11 - 00;27;35;07
to leaders
who want to step up and effectively manage

00;27;35;10 - 00;27;39;28
AI ethically
and responsibly in their organization?

00;27;40;01 - 00;27;41;18
There are three things from your end.

00;27;41;18 - 00;27;42;25
Then I'll go to Huw.

00;27;42;28 - 00;27;46;19
Number one go rate ISO standard 4 to 1,

00;27;47;00 - 00;27;49;24
which is the standard on
how to manage AI systems.

00;27;49;25 - 00;27;54;02
If you're looking for a guide to good,
that's a great place to start.

00;27;54;18 - 00;27;57;18
Number two which sits
within that, know what you've got

00;27;57;25 - 00;28;00;21
and be aware of the risks around that.

00;28;00;24 - 00;28;02;20
And number three is get going.

00;28;02;23 - 00;28;05;29
Don't fall into the analysis
paralysis trap that you're going to wait

00;28;06;02 - 00;28;09;02
for the rest of the world to get there,
and then you'll jump on the train.

00;28;09;12 - 00;28;12;21
This is AI technology to understand.

00;28;12;24 - 00;28;18;01
This is a tool to be put to work,
and now is a great time to do that,

00;28;18;09 - 00;28;21;25
recognizing that there will be learning,
there will be failures.

00;28;22;05 - 00;28;25;19
But it's the learning through the things
that don't work that will get you

00;28;25;22 - 00;28;28;22
and your organization
to what good looks like for you.

00;28;29;12 - 00;28;30;28
I would advise organizations

00;28;30;28 - 00;28;34;13
to be very clear about what
their objectives for the use of AI is.

00;28;35;00 - 00;28;38;26
Experimentation can be an objective,
but it only has a limited lifespan.

00;28;39;12 - 00;28;44;11
Is your objective
to become a developer of AI systems?

00;28;44;23 - 00;28;49;02
But do you want to use do you have
IP that you want to monetize through AI,

00;28;49;10 - 00;28;52;28
or are you an organization
that consumes AI?

00;28;53;14 - 00;28;56;12
So that's setting objectives
and that will help you shape

00;28;56;15 - 00;28;58;07
the resources you need.

00;28;58;10 - 00;29;01;00
Secondly, tied to resources is to be very,

00;29;01;00 - 00;29;04;13
very clear
about what it is that the organization

00;29;04;20 - 00;29;07;23
needs to make available to its staff

00;29;08;02 - 00;29;10;24
and potentially to its customers, as well.

00;29;10;27 - 00;29;14;14
And then make sure that you have

00;29;14;29 - 00;29;17;17
AI literacy in the organization,

00;29;17;20 - 00;29;22;26
in the same way
that we have literacy around it, security,

00;29;23;05 - 00;29;26;04
occupational health and safety,
etc., etc.,

00;29;26;07 - 00;29;29;07
set up a similar program for AI.

00;29;29;15 - 00;29;35;02
It will help people grow and it will help
you avoid making costly mistakes.

00;29;35;14 - 00;29;41;00
One more to add to that, I think, is
that recognize that AI is a business tool.

00;29;41;03 - 00;29;43;01
It is not a technology tool.

00;29;43;04 - 00;29;48;12
AI ethics is a business risk and it needs
business governance over the top of it.

00;29;48;15 - 00;29;51;15
That aligns with the organization's
strategy and objectives.

00;29;51;22 - 00;29;55;02
Sarah Dodds,
thank you so much for joining us.

00;29;55;05 - 00;29;56;25
And Alice, thank you.

00;29;56;25 - 00;29;58;04
I really appreciate you joining.

00;29;58;10 - 00;29;59;06
Thank you Kumar.

00;29;59;09 - 00;30;03;01
We covered a wide range of topics,
taking into consideration

00;30;03;04 - 00;30;07;01
the fact that generative
AI is evolving at an exponential pace

00;30;07;13 - 00;30;11;21
and our use of generative
AI in a responsible

00;30;11;24 - 00;30;16;05
and ethical manner
also need to evolve at that pace.

00;30;16;08 - 00;30;19;14
We discussed how data should be collected.

00;30;19;17 - 00;30;24;17
We discussed how we should ensure
that there is no bias

00;30;24;20 - 00;30;28;26
or biases
in the data collection, or reduced.

00;30;29;09 - 00;30;32;03
We spoke about cultural differences

00;30;32;06 - 00;30;35;26
in what is considered fair
and ethical in various parts of the world.

00;30;36;08 - 00;30;39;23
But the good news is
that the standards are now available

00;30;39;26 - 00;30;44;10
from international standards
organizations, and various countries

00;30;44;23 - 00;30;47;27
are developing their own legislation

00;30;48;05 - 00;30;51;03
that business units
should become familiar with.

00;30;51;04 - 00;30;55;11
So it's our responsibility
as business leaders to be on the top

00;30;55;17 - 00;30;58;17
of not only the potential
of what I can offer,

00;30;58;21 - 00;31;04;06
but also how we can ethically
and responsibly use AI.

00;31;04;13 - 00;31;04;28
Thank you.

00;31;10;03 - 00;31;13;03
Brought to you by Nexus, Publish By GHD.

00;31;13;11 - 00;31;14;15
Where ideas connect