<begin subtitles>

00:00:09:14 00:00:10:21
Hello, you're listening to

00:00:11:09 00:00:13:04
Leadership in Extraordinary Times.

00:00:13:17 00:00:15:10
I'm Peter Tufano, the Dean

00:00:15:12 00:00:17:06
of the Said Business School at the

00:00:17:07 00:00:18:11
University of Oxford.

00:00:19:10 00:00:21:05
This podcast is curated from a

00:00:21:06 00:00:23:00
series of live online events

00:00:23:10 00:00:25:03
which began in response to the

00:00:25:04 00:00:26:08
covid-19 outbreak.

00:00:27:06 00:00:28:16
It's never been more important to

00:00:28:17 00:00:30:17
understand and respond to the issue

00:00:30:18 00:00:31:17
shaping our future.

00:00:32:13 00:00:34:01
We want to share our insights and

00:00:34:02 00:00:35:12
our research from the front lines of

00:00:35:13 00:00:36:13
business.

00:00:36:24 00:00:38:14
And as a transformational business

00:00:38:15 00:00:40:04
school, we want to promote new

00:00:40:05 00:00:42:00
thinking when we need it the most

00:00:42:06 00:00:44:00
and reveal how to find a path to the

00:00:44:04 00:00:45:12
opportunities of the future.

00:00:46:17 00:00:48:04
Please do explore our growing

00:00:48:05 00:00:49:17
library of past episodes if you

00:00:49:24 00:00:50:24
haven't already.

00:01:00:12 00:01:01:12
Episode two, How

00:01:02:13 00:01:03:23
can businesses use artificial

00:01:04:06 00:01:05:18
intelligence responsibly?

00:01:07:09 00:01:09:01
The University of Oxford is at the

00:01:09:05 00:01:11:02
cutting edge of the revolution,

00:01:11:22 00:01:13:13
we live in a world in which change

00:01:13:18 00:01:15:09
is driven by new technologies

00:01:15:17 00:01:17:11
and artificial intelligence, or A.I.

00:01:17:23 00:01:18:25
is at the heart of this.

00:01:19:23 00:01:21:19
AI, predictive analytics

00:01:21:23 00:01:23:12
and other digital disruptions are

00:01:23:13 00:01:25:02
transforming business models,

00:01:25:06 00:01:27:20
processes and customer relationships

00:01:27:21 00:01:29:03
across every industry.

00:01:30:01 00:01:31:14
Organisations increasingly find

00:01:31:23 00:01:33:11
themselves confronting moments of

00:01:33:12 00:01:34:16
existential change,

00:01:35:09 00:01:36:23
and in this landscape of digital

00:01:36:24 00:01:39:05
disruption, leaders need both

00:01:39:09 00:01:41:13
the understanding and the tools

00:01:41:19 00:01:42:20
to integrate A.I.

00:01:42:21 00:01:44:02
into their overall strategy.

00:01:45:18 00:01:47:19
In this episode, Oxford Said's

00:01:47:20 00:01:49:07
L'Oreal Professor of Marketing,

00:01:49:09 00:01:50:18
Professor Andrew Stephen, Dr

00:01:51:11 00:01:53:12
Natalia Eframova and Dr

00:01:53:13 00:01:54:24
Felipe Thomaz are joined by Dr

00:01:55:17 00:01:57:20
Yasmeen Ahmad, Vice President

00:01:57:21 00:01:59:07
of strategy at Teradata,

00:01:59:19 00:02:01:02
to discuss the principles for

00:02:01:03 00:02:02:20
ethical uses of in business

00:02:03:06 00:02:04:21
based on a new report from Oxford

00:02:05:01 00:02:06:24
Said and the International Chamber

00:02:07:00 00:02:07:24
of Commerce.

00:02:08:25 00:02:10:09
Now over to Professor Andrew

00:02:10:11 00:02:11:11
Stephen.

00:02:12:00 00:02:14:00
Hi, and welcome to today's

00:02:14:01 00:02:15:05
episode of Leadership in

00:02:15:15 00:02:16:15
Extraordinary Times.

00:02:16:19 00:02:18:14
Today, we are talking about

00:02:18:15 00:02:19:25
how businesses can use AI

00:02:20:11 00:02:21:11
responsibly.

00:02:21:20 00:02:23:11
We're going to try and unpack that

00:02:23:12 00:02:25:03
and make it as practical

00:02:25:17 00:02:26:23
as we possibly can

00:02:27:18 00:02:29:12
throughout today's discussion.

00:02:30:05 00:02:31:13
I want to welcome all of you

00:02:31:23 00:02:33:07
wherever you happen to be.

00:02:33:14 00:02:35:01
Thank you for joining us.

00:02:35:06 00:02:36:09
And I want to extend a special

00:02:36:10 00:02:37:21
welcome actually to our

00:02:38:09 00:02:40:13
Oxford Executive Diploma

00:02:40:15 00:02:42:13
in AI for Business students, some

00:02:42:14 00:02:44:02
of whom are here in the building

00:02:44:12 00:02:45:17
here at the Said Business

00:02:46:12 00:02:47:24
School and others who are around the

00:02:48:00 00:02:49:12
world joining us today, because

00:02:49:19 00:02:51:06
actually today is the first day

00:02:51:19 00:02:52:23
of one of our modules of

00:02:53:15 00:02:55:01
that Executive Diploma programme.

00:02:55:02 00:02:57:02
And what better topic to talk about

00:02:58:01 00:02:59:20
for an AI programme than

00:02:59:24 00:03:01:18
responsible AI in business.

00:03:01:25 00:03:03:16
So what we're going to do over the

00:03:03:17 00:03:04:18
course of the next 45

00:03:05:21 00:03:08:02
minutes or so is break

00:03:08:03 00:03:09:13
down this topic.

00:03:09:23 00:03:11:07
We're going to hear from

00:03:11:24 00:03:13:24
Dr. Yasmeen Ahmad, who is the

00:03:14:00 00:03:15:11
Vice President for Strategy at

00:03:15:14 00:03:17:17
Teradata, as well as Dr.

00:03:17:18 00:03:18:17
Natalia Eframova, who

00:03:19:14 00:03:21:21
is the Teradata Research Fellow

00:03:21:22 00:03:23:00
in A.I.

00:03:23:01 00:03:24:14
and Marketing here at the Said

00:03:24:21 00:03:25:19
Business School, working in the

00:03:25:20 00:03:26:23
Oxford Future of Marketing

00:03:26:24 00:03:27:22
Initiative.

00:03:27:23 00:03:29:24
And so we'll hear from them in

00:03:30:05 00:03:32:05
a little while to really talk

00:03:32:06 00:03:34:07
about this from both a technical

00:03:34:08 00:03:36:07
and a practical standpoint.

00:03:36:08 00:03:38:05
But first, I want to bring

00:03:38:06 00:03:40:01
in my colleague, Dr. Felipe Thomaz,

00:03:40:10 00:03:41:18
who's an Associate Professor of

00:03:41:19 00:03:43:15
Marketing here at the Said Business

00:03:43:16 00:03:45:19
School. And Felipe has been heading

00:03:45:20 00:03:47:00
up a project

00:03:48:00 00:03:49:06
that a number of us have been

00:03:49:07 00:03:51:14
working on over the last few months

00:03:51:23 00:03:53:12
with the International Chamber of

00:03:53:13 00:03:55:01
Commerce in

00:03:56:06 00:03:57:17
drafting some guidance for the

00:03:57:22 00:03:59:25
private sector around responsible

00:04:00:00 00:04:01:01
AI. So I thought we'd

00:04:01:22 00:04:03:11
kick things off by having a chat

00:04:03:18 00:04:05:18
with Felipe about this

00:04:05:19 00:04:07:10
this project, which is about to be

00:04:07:16 00:04:09:02
released next week, in fact,

00:04:09:23 00:04:11:23
they'll be releasing the report

00:04:12:05 00:04:13:14
to come out of this research

00:04:13:19 00:04:16:03
project. So, Felipe, welcome.

00:04:16:04 00:04:17:09
Thank you for joining us.

00:04:17:10 00:04:18:20
So why don't you start by telling us

00:04:18:21 00:04:20:21
a little bit about this project

00:04:20:22 00:04:22:10
that we've been working on over

00:04:22:21 00:04:25:03
the last few months with the ICC.

00:04:26:15 00:04:27:20
Thank you, Andrew. It's a very
exciting project.

00:04:27:23 00:04:29:23
The challenge was how do we

00:04:30:07 00:04:32:07
synthesise what we know

00:04:32:08 00:04:34:16
about responsible AI applications

00:04:34:17 00:04:36:03
and ethical considerations and

00:04:36:08 00:04:38:07
ethical risks in AI, and

00:04:38:12 00:04:40:06
organise it in a way that is

00:04:40:15 00:04:41:15
business sensible and

00:04:42:14 00:04:43:18
it is in a way organised

00:04:44:21 00:04:46:23
to allow the manager to implement

00:04:46:24 00:04:49:00
and actually put it to use so

00:04:49:01 00:04:50:12
that they can just go ahead and take

00:04:50:13 00:04:51:23
those initial steps.

00:04:53:11 00:04:54:23
All right, so so what are some

00:04:55:10 00:04:56:18
of those steps that need to be

00:04:56:19 00:04:58:14
taken, because part of this is

00:04:58:15 00:04:59:24
very much about coming up with

00:05:00:10 00:05:02:08
things that businesses can do

00:05:02:18 00:05:04:03
to try and ensure that the work that

00:05:04:04 00:05:05:14
they're doing around AI and machine

00:05:05:15 00:05:07:13
learning is is responsible, is

00:05:07:14 00:05:09:08
ethical, is is appropriate.

00:05:09:25 00:05:12:01
What are some of the, I guess, the

00:05:12:02 00:05:13:11
elements in your framework?

00:05:14:07 00:05:15:07
Right. So to start

00:05:16:06 00:05:17:11
out, one of the things that we had

00:05:17:12 00:05:18:14
to do in order to force

00:05:19:08 00:05:20:09
it towards a business

00:05:21:05 00:05:22:24
oriented and actually a managerially

00:05:23:03 00:05:24:21
useful approach

00:05:25:19 00:05:28:07
was actually to force a hierarchy,

00:05:28:13 00:05:30:19
also some complex ethical ideas,

00:05:31:09 00:05:32:13
basically organising the

00:05:33:11 00:05:34:22
questions around what is right, what

00:05:34:23 00:05:36:21
is correct, and organising the trade

00:05:36:22 00:05:38:14
offs that managers are going to face

00:05:39:10 00:05:40:12
in a way that allows them to

00:05:40:13 00:05:42:00
actually analyse their environment.

00:05:42:01 00:05:44:05
So still very much into

00:05:44:19 00:05:46:04
ethereal land and discussion here.

00:05:46:13 00:05:47:23
But just to say that to go to

00:05:48:15 00:05:49:15
a framework, we

00:05:50:14 00:05:52:01
are actually forcing a way for

00:05:52:06 00:05:54:08
managers to make trade-offs

00:05:54:09 00:05:55:16
and decisions and assist in

00:05:55:18 00:05:56:21
decision. So to do that

00:05:57:14 00:05:58:16
is starting to go from

00:05:59:12 00:06:00:23
the most complex ideas to the

00:06:01:09 00:06:03:08
more simple ideas that

00:06:03:09 00:06:05:02
make up those more esoteric

00:06:05:06 00:06:07:07
concepts. So ethics is

00:06:07:09 00:06:08:13
essentially a combination of

00:06:08:14 00:06:10:12
responsibility and accountability,

00:06:10:17 00:06:12:19
and then responsibility becomes

00:06:12:20 00:06:14:06
human centricity, fairness and

00:06:15:00 00:06:16:16
the ability to be harmless in your

00:06:16:17 00:06:17:16
execution.

00:06:18:03 00:06:19:07
And the report is going to have the

00:06:19:08 00:06:20:10
full kind of components

00:06:21:11 00:06:23:00
available for people to dive in.

00:06:23:06 00:06:24:22
But the idea is to say, how do the

00:06:24:25 00:06:26:07
pieces fit together?

00:06:26:18 00:06:28:16
How do we organise them in a way

00:06:28:17 00:06:30:08
that say what has priority?

00:06:30:09 00:06:31:18
What is more important than another

00:06:31:22 00:06:33:11
is to allow for these trade offs.

00:06:33:22 00:06:35:09
Now, when I take it to reality

00:06:36:05 00:06:37:08
and I go, let's go, how

00:06:38:02 00:06:39:19
do I do this as a manager?

00:06:39:23 00:06:40:23
How do I organise my

00:06:41:24 00:06:43:15
business to start leveraging these

00:06:43:16 00:06:44:13
things?

00:06:44:14 00:06:46:05
We start to get to those practical

00:06:46:07 00:06:47:01
points.

00:06:47:02 00:06:48:10
And what we're trying to do is say,

00:06:48:11 00:06:49:22
OK, you're going to start from

00:06:50:08 00:06:52:05
the technical aspect of

00:06:52:13 00:06:53:23
you're going to go to your workflow

00:06:54:23 00:06:55:24
from that workflow.

00:06:56:00 00:06:58:02
You're going to overlay an ethical

00:06:58:03 00:06:59:15
design component onto it, which

00:06:59:24 00:07:01:12
is essentially saying, here are the

00:07:01:13 00:07:02:16
steps that I'm going to take.

00:07:02:17 00:07:04:10
So very broadly listening to

00:07:05:09 00:07:07:00
here's my data concerns, here's my

00:07:07:01 00:07:08:16
algorithmic choices and here's my

00:07:08:17 00:07:09:24
business case. This is what

00:07:10:12 00:07:12:11
I'm going to use my outputs for.

00:07:13:01 00:07:14:14
And you just map all that out and

00:07:14:15 00:07:16:13
that overlay of ethics comes in

00:07:16:14 00:07:18:10
and starts asking questions of

00:07:18:11 00:07:19:20
the manager saying, do I have

00:07:20:18 00:07:22:23
specific threats or risks

00:07:22:24 00:07:24:06
or concerns arising from my

00:07:24:22 00:07:25:18
data sourcing?

00:07:25:19 00:07:27:15
Do I have ethics concerns,

00:07:27:20 00:07:29:06
risks associated with my data

00:07:29:18 00:07:31:04
cleaning up, pre processing,

00:07:31:24 00:07:33:13
how I use it, how I absorb it into

00:07:33:14 00:07:35:03
my company, all of those things,

00:07:35:13 00:07:37:03
and start getting matched so that

00:07:37:06 00:07:39:14
you have a good understanding

00:07:39:17 00:07:40:24
of these potential threats

00:07:41:21 00:07:43:11
to your ethics test.

00:07:43:19 00:07:45:04
So you're saying I want to do things

00:07:45:06 00:07:47:06
well, then there are

00:07:47:07 00:07:49:15
aspects within your design

00:07:49:16 00:07:50:22
of your workflow for your AI

00:07:51:00 00:07:52:06
application that can give

00:07:52:25 00:07:54:14
rise to some complications.

00:07:55:02 00:07:57:02
The first step then is to map those,

00:07:57:03 00:07:58:20
identify what they are, and then we

00:07:58:24 00:08:01:01
fall into some fairly robust

00:08:01:13 00:08:02:17
managerial aspects where

00:08:03:11 00:08:05:12
we say, OK, what are those

00:08:05:13 00:08:06:24
mitigation strategies that I can

00:08:07:00 00:08:08:16
bring to bear to account for this

00:08:08:17 00:08:10:03
risk exposure that I have?

00:08:10:21 00:08:12:19
How do I minimise my risk exposure

00:08:12:20 00:08:14:04
and this are ethical risks,

00:08:14:05 00:08:15:19
potential risks that you can avoid

00:08:16:02 00:08:17:09
doing incorrect things and

00:08:17:24 00:08:19:13
then institutionalising them via AI

00:08:20:09 00:08:21:17
or just like embedding them

00:08:22:09 00:08:23:22
into code, as it were, and just

00:08:23:23 00:08:25:09
perpetuating those errors.

00:08:25:14 00:08:27:04
So this is an early mapping,

00:08:27:15 00:08:29:18
understanding how it fits within

00:08:29:19 00:08:31:16
what you're trying to apply and

00:08:31:17 00:08:33:03
then minimise and mitigate and

00:08:33:12 00:08:35:08
just live with that notion

00:08:35:09 00:08:36:10
and that idea of saying that I'm

00:08:36:11 00:08:37:25
going to revisit and keep coming

00:08:38:06 00:08:39:09
back to this process to

00:08:40:02 00:08:41:13
saying, do I have any threats to

00:08:41:15 00:08:42:12
recognise new things?

00:08:42:13 00:08:43:22
Am I changing anything in my

00:08:43:23 00:08:45:18
workflow, being able to

00:08:45:19 00:08:47:08
actually start delivering against

00:08:47:15 00:08:49:08
that goal, which is what is the most

00:08:49:09 00:08:50:23
responsible or what is a proper

00:08:51:03 00:08:52:19
ethical application of AI

00:08:53:03 00:08:54:03
for my business?

00:08:54:24 00:08:56:18
And so, Felipe, this

00:08:57:20 00:08:59:05
this research that that we're

00:08:59:06 00:09:00:23
talking is going to be released next

00:09:00:24 00:09:02:10
week. And if anyone is wanting

00:09:02:21 00:09:04:10
to find out more, they can visit

00:09:04:16 00:09:05:16
OxfordFutureOfMarketing.com.

00:09:07:05 00:09:09:09
So what I'm hearing from you is,

00:09:09:16 00:09:11:03
is in some sense saying, look, it's

00:09:11:04 00:09:13:00
not about changing the way that

00:09:13:01 00:09:14:03
you do everything.

00:09:14:04 00:09:15:25
It's about sort of bringing a

00:09:15:26 00:09:17:14
set of responsible, ethical,

00:09:17:15 00:09:19:22
accountable notions around AI and

00:09:19:23 00:09:21:22
data and analytics usage into

00:09:22:04 00:09:23:21
existing workflows, which I think is

00:09:24:15 00:09:25:19
is quite appealing.

00:09:25:20 00:09:27:03
Right. It's not about overhauling

00:09:27:04 00:09:28:18
everything we do for this new way

00:09:28:24 00:09:30:18
of doing things, but rather finding,

00:09:30:24 00:09:32:20
again, some kind of happy medium,

00:09:33:02 00:09:34:22
you know, am I getting it right in

00:09:34:23 00:09:36:13
that characterisation the way that

00:09:36:14 00:09:37:14
you're thinking?

00:09:37:17 00:09:39:11
Yeah, I mean, I think it's important

00:09:39:15 00:09:41:03
for managers everywhere, really

00:09:41:12 00:09:43:06
to have that sense of terra firma

00:09:43:07 00:09:44:16
essentially saying, like, this is

00:09:44:17 00:09:45:16
familiar ground.

00:09:46:01 00:09:47:04
This is something that we do in

00:09:47:06 00:09:48:20
business. There's a lot of

00:09:49:01 00:09:50:22
new news with AI; there's a lot

00:09:50:23 00:09:51:19
of hype.

00:09:51:20 00:09:53:16
But from the business sense

00:09:53:17 00:09:55:17
and for the business case use of it,

00:09:55:18 00:09:57:00
it's very useful to go back

00:09:57:13 00:09:58:22
to some of those basic ideas

00:09:59:15 00:10:01:14
and say, like, actually, I might

00:10:01:15 00:10:03:02
have a new business model, I might

00:10:03:03 00:10:05:00
have new capabilities, but actually

00:10:05:01 00:10:06:16
the art and science of management

00:10:07:06 00:10:08:20
behind it, it's still relatively

00:10:09:10 00:10:11:02
stable that I can have appropriate

00:10:11:03 00:10:13:04
controls over these different

00:10:13:05 00:10:14:24
components. It's a new machine,

00:10:15:04 00:10:16:14
but actually how I manage the

00:10:16:15 00:10:18:01
machinery doesn't have a whole

00:10:18:14 00:10:20:04
lot of new moving parts to it.

00:10:20:12 00:10:22:12
We have some concerns

00:10:22:13 00:10:23:20
that are specific to AI and

00:10:24:09 00:10:26:03
some uniqueness that we describe.

00:10:26:16 00:10:28:00
And we were able to leverage

00:10:28:21 00:10:30:10
a lot of pre-existing, like -

00:10:30:18 00:10:32:05
one of the most exciting parts

00:10:32:15 00:10:34:01
for me for this project was

00:10:34:19 00:10:36:09
the way that we arrived at these

00:10:36:10 00:10:38:07
recommendations was by actually

00:10:38:08 00:10:40:11
going to existing body of knowledge,

00:10:40:19 00:10:42:06
everything that companies have

00:10:42:09 00:10:44:12
mentioned on their stances,

00:10:44:13 00:10:45:23
their guidances, governments,

00:10:46:13 00:10:47:13
NGOs.

00:10:47:23 00:10:49:06
We took all of that body of

00:10:49:18 00:10:51:02
knowledge, as well as the academic

00:10:51:03 00:10:52:16
literature on responsibility and

00:10:52:22 00:10:54:04
ethics, and AI,

00:10:54:19 00:10:56:09
put it all together and use our own

00:10:56:10 00:10:57:21
AI machinery on it to organise

00:10:58:15 00:11:00:03
it and give it the shape to put

00:11:00:11 00:11:01:18
things in perspective.

00:11:03:01 00:11:03:24
So I think I think what you're

00:11:04:00 00:11:05:10
talking about here is quite
appealing,

00:11:05:22 00:11:07:20
I guess, process for

00:11:07:21 00:11:09:10
the businesses of all kinds of

00:11:09:11 00:11:10:15
shapes and sizes

00:11:11:10 00:11:13:03
to use. And I like that kind of

00:11:13:09 00:11:14:18
you know, that same use of that

00:11:14:19 00:11:16:04
businesses need that terra firma,

00:11:16:05 00:11:17:13
which is business, and I want to

00:11:17:14 00:11:18:13
come back to that.

00:11:18:14 00:11:20:15
But it's also suggesting

00:11:21:00 00:11:22:17
that businesses themselves need to

00:11:22:23 00:11:24:12
be taking responsibility

00:11:25:03 00:11:26:17
for responsible AI.

00:11:27:20 00:11:29:10
I want to hear what you all watching

00:11:29:19 00:11:30:18
think about this, so we're going to

00:11:30:19 00:11:31:21
do a poll and I'm going

00:11:32:15 00:11:34:10
to invite you to to

00:11:34:11 00:11:35:06
do an online poll.

00:11:35:07 00:11:37:01
And we'll come back to the results

00:11:37:09 00:11:38:16
a little bit later on in the

00:11:38:17 00:11:39:13
programme.

00:11:39:14 00:11:41:14
But the question is, who should

00:11:41:15 00:11:43:03
be most responsible

00:11:43:25 00:11:46:08
for ensuring that applications of AI

00:11:46:20 00:11:48:14
in the private and public sectors

00:11:48:15 00:11:50:03
are appropriate, ethical

00:11:50:11 00:11:51:11
and responsible?

00:11:51:18 00:11:53:04
So is it government?

00:11:53:05 00:11:54:21
Is it the tech companies and so on

00:11:54:24 00:11:56:02
and so on, whether it's

00:11:56:22 00:11:59:01
industry bodies or the businesses

00:11:59:02 00:12:00:19
themselves, maybe intergovernmental

00:12:01:02 00:12:02:11
organisations like the United

00:12:02:12 00:12:03:11
Nations or

00:12:04:11 00:12:05:11
individuals?

00:12:05:15 00:12:06:20
But Felipe, one other thing I want

00:12:06:21 00:12:08:20
to talk about with you is

00:12:08:21 00:12:10:10
you mentioned sort of I guess these

00:12:10:11 00:12:12:09
principles or pillars of thinking

00:12:12:10 00:12:14:01
about what what, I guess, a set of

00:12:14:03 00:12:15:18
guidelines or policies would be.

00:12:16:05 00:12:17:15
There's a term that you mentioned

00:12:17:16 00:12:19:05
that I really want to hear a little

00:12:19:06 00:12:20:13
bit more on, which is human

00:12:20:18 00:12:21:18
centricity.

00:12:22:09 00:12:23:17
What what do you mean by that?

00:12:24:17 00:12:25:17
So that one is

00:12:26:15 00:12:28:17
kind of one of the most important

00:12:28:18 00:12:31:06
layers of the ethics considerations,

00:12:31:13 00:12:32:23
which probably shouldn't surprise

00:12:32:24 00:12:34:05
anybody in the room, right.

00:12:34:11 00:12:35:17
We're talking about ethics

00:12:36:07 00:12:37:17
and human considerations, the

00:12:37:18 00:12:39:10
consequences that impacts on

00:12:39:11 00:12:41:01
individual human individuals here.

00:12:41:03 00:12:43:05
So here what we're talking about is

00:12:43:15 00:12:45:22
a combination of actually

00:12:46:01 00:12:47:23
achievable means and

00:12:47:24 00:12:49:10
goals within the AI to deliver

00:12:49:21 00:12:51:12
on human benefits.

00:12:51:18 00:12:53:15
So one core thing inside of that

00:12:53:16 00:12:55:07
is this idea of beneficence, which

00:12:55:16 00:12:57:04
is you're going to generate

00:12:57:11 00:12:59:14
something good out of this process

00:12:59:15 00:13:00:14
for people.

00:13:00:25 00:13:02:18
It can be very broad or it can be a

00:13:02:22 00:13:04:13
good business outcome that comes out

00:13:04:14 00:13:05:13
of it.

00:13:05:17 00:13:07:18
A transparency which is often

00:13:07:19 00:13:09:11
discussed in terms of

00:13:10:01 00:13:11:21
how deploying trustworthy AI is on

00:13:12:02 00:13:13:15
the idea to have an intelligible,

00:13:14:12 00:13:15:22
understandable system.

00:13:16:15 00:13:17:24
All of these are components that

00:13:18:00 00:13:19:01
make it centred around

00:13:19:24 00:13:21:14
the human, essentially - it's almost

00:13:21:15 00:13:22:21
like using the word to define itself

00:13:22:22 00:13:25:06
- is the more

00:13:25:10 00:13:27:12
that a human can interact and

00:13:27:17 00:13:29:01
the more that the system then

00:13:29:06 00:13:31:03
appreciates that there's a human

00:13:31:05 00:13:33:01
that is going to bear a consequence

00:13:33:08 00:13:35:11
of our automated decision

00:13:35:12 00:13:36:11
making, then the more

00:13:37:07 00:13:38:15
stable you are into being a

00:13:38:16 00:13:40:17
responsible business and the more

00:13:40:18 00:13:42:04
grounded you are in the fact that

00:13:42:05 00:13:43:15
there is going to be a human cost

00:13:43:16 00:13:44:24
associated with some of the

00:13:45:09 00:13:46:18
decisions that we make inside of our

00:13:46:19 00:13:47:19
businesses.

00:13:48:01 00:13:49:06
All right, thank you so, Felipe.

00:13:50:02 00:13:51:15
Sit tight because we'll come back to

00:13:51:16 00:13:52:24
you in the Q&A a little bit

00:13:53:11 00:13:54:11
later.

00:13:54:14 00:13:55:14
So thanks, Doctor Felipe Thomaz,
Associate

00:13:56:10 00:13:57:06
Professor of Marketing. And I want

00:13:57:07 00:13:59:05
to now bring in our two

00:13:59:06 00:14:00:21
panellists as I introduced before

00:14:01:09 00:14:03:09
Dr. Yasmeen Ahmad, who is

00:14:03:10 00:14:05:04
VP of Strategy at Teradata, which

00:14:05:05 00:14:07:04
is an enterprise technology

00:14:07:05 00:14:08:22
company, and Dr. Natalia Eframova,

00:14:09:18 00:14:11:01
who is a computer scientist

00:14:11:23 00:14:13:01
and is a

00:14:13:20 00:14:15:24
Research Fellow in Marketing

00:14:16:02 00:14:17:14
and AI here at the Said Business

00:14:17:15 00:14:19:17
School, and Natalia is also heavily

00:14:19:18 00:14:21:02
involved in the research work

00:14:21:13 00:14:22:23
that Felipe was just telling us

00:14:22:24 00:14:23:24
about.

00:14:24:04 00:14:25:15
So welcome to both of you.

00:14:26:00 00:14:27:14
What I want to do and we'll come to

00:14:27:15 00:14:28:19
the poll results a little bit later

00:14:28:20 00:14:29:25
as well. But but first, I

00:14:30:16 00:14:32:07
want to actually come to

00:14:33:01 00:14:35:01
a question that actually

00:14:35:02 00:14:36:18
has popped up from some of our

00:14:37:13 00:14:39:01
A.I. Diploma - Executive AI

00:14:39:16 00:14:40:18
Diploma students,

00:14:41:12 00:14:43:10
which is actually around how

00:14:43:11 00:14:44:17
well a business is doing

00:14:45:10 00:14:46:04
at the moment.

00:14:46:05 00:14:47:25
So Andrew, Claire

00:14:48:04 00:14:50:01
and a few others have asked this

00:14:50:02 00:14:51:15
question from that group.

00:14:51:22 00:14:53:07
You know, if we sort of think about

00:14:53:08 00:14:54:08
responsible AI, where's

00:14:55:04 00:14:57:00
kind of the starting point at the

00:14:57:01 00:14:58:05
moment with businesses?

00:14:58:21 00:15:00:05
Are businesses already pretty

00:15:00:06 00:15:01:08
responsible with this, or is there

00:15:01:09 00:15:02:08
room for improvement?

00:15:02:20 00:15:04:09
Yasmeen, let me let me go to you

00:15:04:10 00:15:05:09
first.

00:15:06:11 00:15:07:23
Thanks, Andrew. That is a great

00:15:07:25 00:15:08:24
question.

00:15:09:00 00:15:10:18
So how are businesses thinking about

00:15:10:19 00:15:12:07
this? And I think it's worth

00:15:12:08 00:15:13:13
reflecting on

00:15:14:09 00:15:15:23
how do businesses think about AI?

00:15:16:16 00:15:18:01
And if I just rewind back in

00:15:18:13 00:15:20:11
my career, I was

00:15:20:12 00:15:22:05
looking after data science analytics

00:15:22:19 00:15:24:13
teams who were working with our

00:15:24:14 00:15:26:08
clients, some of the largest

00:15:26:09 00:15:28:00
companies in the world, and in

00:15:28:24 00:15:30:20
new AI data science

00:15:30:21 00:15:32:13
techniques, a few years back

00:15:32:19 00:15:33:23
at that point, AI,

00:15:35:03 00:15:36:10
some of the newer machine learning

00:15:36:11 00:15:37:13
techniques, etc.

00:15:37:17 00:15:39:22
were isolated to these

00:15:39:23 00:15:41:07
data science functions, these

00:15:41:13 00:15:43:07
centres of excellence that have been

00:15:43:09 00:15:44:25
set up, whether that was in the bank

00:15:45:00 00:15:46:07
or the retail or the telco

00:15:47:03 00:15:48:03
company.

00:15:48:08 00:15:49:08
And so it was self

00:15:50:09 00:15:51:22
constrained in some ways because it

00:15:51:23 00:15:53:20
was one group that was developing

00:15:54:00 00:15:54:24
these algorithms.

00:15:55:00 00:15:55:24
It was one stop shop

00:15:56:24 00:15:58:08
for looking at how are those

00:15:58:19 00:16:00:09
algorithms being developed, where

00:16:00:14 00:16:02:00
what kind of use cases are they

00:16:02:01 00:16:03:01
being applied to?

00:16:03:02 00:16:04:11
Where in the business are we

00:16:04:17 00:16:06:05
leveraging those algorithms?

00:16:06:18 00:16:08:23
You fast forward to today and

00:16:08:24 00:16:10:06
we really see AI being much

00:16:10:24 00:16:12:04
more pervasive across the

00:16:12:07 00:16:13:07
organisation.

00:16:13:19 00:16:15:18
Those use cases are no longer

00:16:15:23 00:16:17:24
limited to one group that's

00:16:18:00 00:16:19:18
doing the development work or in one

00:16:20:02 00:16:21:02
business function.

00:16:21:12 00:16:23:09
We see whether it's

00:16:23:10 00:16:25:13
supply chain, customer experience,

00:16:25:14 00:16:27:22
operations, fraud departments,

00:16:28:08 00:16:29:08
they're all leveraging AI

00:16:30:11 00:16:32:03
analytics techniques to

00:16:32:07 00:16:33:25
improve their operations, to create

00:16:34:05 00:16:35:05
differentiation.

00:16:35:21 00:16:37:12
That makes it really challenging for

00:16:37:13 00:16:39:09
really large organisations

00:16:39:16 00:16:40:21
to now be able to control

00:16:41:24 00:16:44:06
how that AI is applied, to control

00:16:44:13 00:16:45:17
the decisions that AI is

00:16:46:13 00:16:47:13
making.

00:16:47:17 00:16:49:01
And coming back to what Felipe

00:16:49:15 00:16:51:04
said there, the human centricity

00:16:52:02 00:16:53:13
piece, even when the decision

00:16:54:00 00:16:56:03
is automated, is there a human

00:16:56:09 00:16:58:06
who is ultimately responsible for

00:16:58:07 00:16:59:21
the decision that's being actioned?

00:17:00:08 00:17:01:18
So I think many organisations

00:17:02:09 00:17:03:17
have had to move to looking

00:17:04:09 00:17:05:11
at general guidelines,

00:17:06:05 00:17:08:07
putting in frameworks to support

00:17:08:14 00:17:09:14
application of AI,

00:17:10:11 00:17:12:07
some of those organisations because

00:17:12:08 00:17:13:13
they've seen the backlash

00:17:14:04 00:17:15:22
of when they get it wrong and

00:17:16:10 00:17:17:24
consumers don't like it

00:17:19:07 00:17:20:07
- AI - being applied in certain

00:17:20:08 00:17:21:10
circumstances.

00:17:21:20 00:17:22:20
And so we're seeing

00:17:23:23 00:17:24:23
that AI is

00:17:25:19 00:17:27:19
becoming the responsibility of

00:17:27:20 00:17:29:14
roles like the Chief Data Officer

00:17:29:16 00:17:32:02
to think about not just

00:17:32:09 00:17:34:01
enabling the organisation with the

00:17:34:04 00:17:36:03
tools, but also how do we

00:17:36:04 00:17:37:22
start to govern, how

00:17:38:07 00:17:39:20
AI is used and applied across the

00:17:39:21 00:17:41:20
business and really pervasively

00:17:41:21 00:17:42:21
across the business.

00:17:43:11 00:17:44:18
So Natalia, I want to I want to

00:17:44:19 00:17:46:04
bring you in here and

00:17:46:14 00:17:47:23
kind of take the question in a

00:17:47:24 00:17:49:16
slightly different way: Do we

00:17:49:19 00:17:50:14
actually need this?

00:17:50:15 00:17:51:25
I mean, I know you and others have

00:17:51:26 00:17:53:11
worked hard on thinking about these

00:17:53:12 00:17:55:07
guidelines, but but, you

00:17:55:08 00:17:56:13
know, why is there a need

00:17:57:04 00:17:59:02
for you know, it sort of seems like

00:17:59:03 00:18:00:16
we we have to keep on convincing

00:18:00:17 00:18:02:03
businesses they need to they need to

00:18:02:04 00:18:03:13
actually do these things because

00:18:03:14 00:18:05:10
they are not perhaps innate.

00:18:05:24 00:18:06:24
So why why is

00:18:07:20 00:18:09:03
this a problem that we need to

00:18:09:04 00:18:10:03
solve?

00:18:12:10 00:18:13:20
Thank you for the question, Andrew.

00:18:13:21 00:18:15:14
And it's a really, really good one,

00:18:15:21 00:18:16:24
an important one.

00:18:17:05 00:18:19:02
And this is something not all

00:18:19:03 00:18:21:01
businesses ask themselves, are we

00:18:21:02 00:18:22:15
using the data correctly?

00:18:22:22 00:18:24:01
Are we implementing A.I.

00:18:24:12 00:18:25:12
correctly?

00:18:25:18 00:18:26:21
Because AI function is,

00:18:27:23 00:18:29:21
in many cases, not the central

00:18:29:22 00:18:30:18
to the business.

00:18:30:19 00:18:31:23
It it has a support role

00:18:32:17 00:18:33:17
in the organisation.

00:18:34:03 00:18:35:21
And in this case, businesses do not

00:18:35:25 00:18:37:20
pay that much attention

00:18:38:02 00:18:39:23
to what is happening there, how

00:18:39:24 00:18:41:14
they should curate their data, how

00:18:41:15 00:18:42:25
the should care, take care of

00:18:43:10 00:18:45:02
their data, how they

00:18:45:07 00:18:47:01
use it in operations.

00:18:47:02 00:18:48:09
This is all very important

00:18:49:08 00:18:51:03
and AI is

00:18:51:07 00:18:52:03
something new.

00:18:52:04 00:18:53:16
This is not something we use in

00:18:53:18 00:18:56:00
practise for very, very long and

00:18:56:13 00:18:58:09
best practises are simply

00:18:58:10 00:18:59:21
not there.

00:19:00:03 00:19:02:02
So every business when

00:19:02:22 00:19:04:20
coming to the question

00:19:04:21 00:19:06:14
'is my AI ethical?'

00:19:06:24 00:19:08:21
has to decide it for

00:19:08:22 00:19:10:06
for itself.

00:19:10:14 00:19:12:05
And it's not always the case that

00:19:12:12 00:19:14:07
they have resources to do that

00:19:14:08 00:19:15:17
or they have education to do

00:19:16:05 00:19:17:23
that, or simply

00:19:18:04 00:19:19:07
they cannot find proper

00:19:19:25 00:19:20:25
guidelines.

00:19:21:10 00:19:22:17
So in practise,

00:19:23:07 00:19:24:14
AI, is ethical as business

00:19:25:12 00:19:27:20
decides it should be ethical

00:19:27:24 00:19:29:22
for now then.

00:19:29:24 00:19:30:24
Do you think it varies, Natalia,

00:19:31:17 00:19:32:23
sort of around the world?

00:19:33:03 00:19:34:03
Are some parts

00:19:35:06 00:19:36:20
of the world thinking more about

00:19:36:23 00:19:38:13
this than than others or thinking

00:19:38:24 00:19:40:04
differently about this to

00:19:40:25 00:19:41:22
others?

00:19:41:23 00:19:43:15
And I'll go with you, Natalia,

00:19:43:16 00:19:44:18
first, but I also want Yasmeen's

00:19:45:03 00:19:46:02
perspective on this. Given that

00:19:46:03 00:19:47:18
you're both both you know,

00:19:48:03 00:19:49:13
you have perspectives on different

00:19:49:14 00:19:50:13
parts of the world.

00:19:50:15 00:19:52:05
So, Natalia, what's your thoughts?

00:19:53:21 00:19:55:23
Well, of course, it's not the same.

00:19:56:07 00:19:58:01
It differs a lot

00:19:58:08 00:20:00:02
across geographies.

00:20:01:17 00:20:03:09
I think that from

00:20:03:14 00:20:05:04
really my personal experience, AI

00:20:05:09 00:20:07:07
ethics is mostly

00:20:07:08 00:20:08:15
developed in more developed

00:20:08:25 00:20:10:23
countries with more

00:20:10:24 00:20:12:04
technical businesses, and

00:20:12:22 00:20:14:04
it's far less developed in

00:20:14:22 00:20:17:04
smaller economies as just

00:20:17:05 00:20:18:12
historically there are less

00:20:19:05 00:20:20:12
resources and

00:20:21:12 00:20:23:06
less opportunities for businesses to

00:20:23:10 00:20:25:12
look at these problems.

00:20:26:05 00:20:28:04
So in many cases,

00:20:28:16 00:20:30:07
AI has to comply with regulations,

00:20:30:18 00:20:32:08
but in some countries is completely

00:20:32:09 00:20:33:05
unregulated.

00:20:33:06 00:20:34:17
And so it's more difficult for

00:20:34:23 00:20:36:10
businesses to regulate themselves,

00:20:36:13 00:20:37:13
of course.

00:20:38:10 00:20:40:08
Yasmeen, from your experience

00:20:41:01 00:20:42:24
working with clients, customers

00:20:43:00 00:20:44:14
in lots of different parts of the

00:20:44:15 00:20:45:09
world.

00:20:45:10 00:20:47:04
What's your take on on this?

00:20:48:03 00:20:50:03
I would have to agree with Natalia,

00:20:50:04 00:20:51:14
I don't think it's consistent

00:20:52:04 00:20:53:04
across the globe.

00:20:53:11 00:20:55:07
In fact, I think it has a lot to

00:20:55:08 00:20:57:00
do also with societies and cultures

00:20:57:17 00:20:59:15
and what's acceptable, what's

00:20:59:16 00:21:01:02
not. And often

00:21:01:11 00:21:02:23
for a society and culture, what's

00:21:02:24 00:21:04:22
acceptable or what consumers,

00:21:04:23 00:21:07:14
individuals, citizens

00:21:07:21 00:21:09:14
are willing to accept is

00:21:09:18 00:21:11:03
reflected in law.

00:21:11:10 00:21:13:11
So there's typically

00:21:13:12 00:21:14:18
more regulation in Europe.

00:21:15:00 00:21:16:13
And so a lot of regulation will

00:21:16:14 00:21:17:13
often drive companies

00:21:18:10 00:21:19:21
to take those steps

00:21:20:16 00:21:22:02
even before the regulation is

00:21:22:03 00:21:23:13
written, because there's that

00:21:23:15 00:21:25:00
feeling of responsibility and

00:21:25:13 00:21:27:07
it will be regulated.

00:21:27:17 00:21:29:14
Whereas having lived in Europe and

00:21:29:15 00:21:30:17
now living in the US, I

00:21:31:11 00:21:33:00
see the difference there between

00:21:33:03 00:21:34:03
those geographies.

00:21:34:04 00:21:35:12
In the US there's typically

00:21:36:01 00:21:37:08
less regulation.

00:21:37:17 00:21:39:15
In some areas there's kind

00:21:39:16 00:21:41:02
of more innovation pushing the

00:21:41:04 00:21:42:19
boundaries and then at some point

00:21:42:20 00:21:44:03
some regulation will come in.

00:21:44:04 00:21:45:17
But it's not consistent across the

00:21:45:18 00:21:46:14
US.

00:21:46:15 00:21:48:14
And then you say, see, in

00:21:48:15 00:21:50:01
Asia-Pacific and Australia, it

00:21:50:14 00:21:52:02
really again varies by country,

00:21:52:07 00:21:54:01
varies by geography,

00:21:54:09 00:21:56:11
and it is linked to that the legal

00:21:56:12 00:21:58:05
regulatory system of what's accepted

00:21:58:19 00:22:00:13
and how our business is governed

00:22:00:19 00:22:02:08
and crucially, what our citizens

00:22:02:24 00:22:04:10
and consumers expecting.

00:22:05:09 00:22:06:18
And actually, Yasmeen, what about I

00:22:06:19 00:22:08:03
mean, we talk about geography, but

00:22:08:04 00:22:09:17
what about different industries?

00:22:10:12 00:22:11:20
That's a great lens to look

00:22:12:13 00:22:14:08
at it, Andrew, and I would

00:22:14:14 00:22:16:10
even take it a step

00:22:16:11 00:22:18:01
deeper than the industry level.

00:22:18:02 00:22:19:07
It's it's the use case in

00:22:19:22 00:22:21:15
the application level, because even

00:22:21:24 00:22:23:20
within an industry, there are

00:22:23:21 00:22:25:06
some business use cases, some

00:22:25:22 00:22:27:07
business functions, departments

00:22:27:08 00:22:28:07
where

00:22:29:03 00:22:30:17
AI is more freely leveraged

00:22:30:24 00:22:32:25
versus other areas in the business.

00:22:32:26 00:22:34:15
And I think it comes down to

00:22:34:21 00:22:36:05
what's the business outcome.

00:22:36:16 00:22:38:22
So how much risk

00:22:38:23 00:22:40:06
are we willing to accept with that

00:22:40:07 00:22:42:01
business outcome if the AI

00:22:42:04 00:22:43:15
gets it wrong or if the AI is

00:22:43:22 00:22:44:22
biased?

00:22:44:23 00:22:46:02
So say we have AI applied

00:22:46:23 00:22:48:16
to our supply chains,

00:22:50:01 00:22:52:02
if the supply chain is not

00:22:52:03 00:22:53:10
quite efficient it might have a

00:22:53:11 00:22:54:24
negative impact on the business, but

00:22:54:25 00:22:56:17
nobody is going to to

00:22:56:22 00:22:58:13
splash that across the front pages

00:22:58:14 00:22:59:25
as being a biased algorithm or

00:23:00:07 00:23:01:07
unethical.

00:23:01:09 00:23:03:03
However, any time you're dealing

00:23:03:04 00:23:05:14
with citizens or customer experience

00:23:05:15 00:23:07:12
or consumer applications of A.I.,

00:23:07:19 00:23:09:11
it takes on another lens.

00:23:09:20 00:23:10:20
So I think naturally

00:23:11:17 00:23:13:14
some industries have more

00:23:13:21 00:23:16:09
consumer citizen

00:23:16:11 00:23:18:13
say it's health care or it's banks

00:23:18:14 00:23:20:19
or retailers have a lot of consumer

00:23:20:20 00:23:22:02
and citizen data and they're

00:23:22:03 00:23:23:10
leveraging AI on that data.

00:23:23:17 00:23:25:19
And so there is more scrutiny there.

00:23:26:01 00:23:27:06
Whereas if you go to the

00:23:27:07 00:23:28:24
manufacturing industries,

00:23:29:14 00:23:31:03
maybe less so because again, the

00:23:31:11 00:23:33:14
application of the AI outcome,

00:23:34:03 00:23:35:09
there's maybe less ethical

00:23:35:21 00:23:37:06
considerations on that

00:23:37:19 00:23:38:19
outcome.

00:23:39:10 00:23:40:14
So it's a good point sort of how

00:23:40:15 00:23:42:01
close you are, I guess, to to the

00:23:42:02 00:23:43:11
humans, which is essentially

00:23:43:23 00:23:45:21
back to Felipe's principle of human

00:23:45:22 00:23:47:24
centricity, I guess.

00:23:48:23 00:23:50:15
Natalia, I want to come back to you

00:23:50:16 00:23:52:05
with a different question, though.

00:23:52:12 00:23:53:21
You know, how do we get

00:23:54:13 00:23:56:13
businesses to really pay

00:23:56:14 00:23:57:21
attention and mean this is sort of

00:23:57:22 00:23:59:10
comes to a question that's that's

00:23:59:11 00:24:01:07
come up already from some of our

00:24:01:21 00:24:03:04
executive diploma students.

00:24:03:10 00:24:04:25
But if I'm a,

00:24:05:09 00:24:07:02
you know, a manager in some

00:24:07:03 00:24:09:02
organisation, how how

00:24:09:03 00:24:11:05
do you convince me that

00:24:11:14 00:24:12:09
being honest is just kind of the

00:24:12:10 00:24:13:13
right thing to do that?

00:24:13:16 00:24:15:05
I really need to to think deeply

00:24:15:11 00:24:16:12
about this, and I'm better in the

00:24:16:13 00:24:18:01
way that we're doing things

00:24:18:11 00:24:20:08
when it's yet another thing that

00:24:20:09 00:24:21:16
I need to, in some sense, comply

00:24:21:17 00:24:23:20
with. Practically speaking,

00:24:24:04 00:24:25:07
how do we get the right

00:24:26:04 00:24:27:09
people in

00:24:28:02 00:24:29:10
organisations to be thinking about

00:24:29:11 00:24:30:17
this, in I guess an action

00:24:31:07 00:24:33:02
way as opposed to a

00:24:34:05 00:24:36:04
sort of a token gesture kind of way?

00:24:38:06 00:24:39:07
That's a great question.

00:24:39:08 00:24:40:08
Thank you, Andrew.

00:24:40:12 00:24:41:12
And

00:24:42:09 00:24:44:01
if we look even deeper,

00:24:44:09 00:24:46:11
we need to ask ourselves, how

00:24:46:12 00:24:47:18
can any individual working

00:24:48:11 00:24:50:24
with AI impact

00:24:50:25 00:24:51:25
the outcome and be ethical?

00:24:52:09 00:24:53:22
Because when

00:24:54:06 00:24:56:12
it comes to AI implementation,

00:24:56:13 00:24:57:24
it's not only about managerial

00:24:58:14 00:25:00:20
decisions in many cases

00:25:00:21 00:25:03:01
when it comes to manager,

00:25:03:06 00:25:05:12
there is a lot of steps done,

00:25:06:19 00:25:09:02
for example, data collection or

00:25:09:17 00:25:12:00
data cleaning or just

00:25:12:01 00:25:14:01
dissemination some information about

00:25:14:02 00:25:15:15
how AI is operating.

00:25:15:21 00:25:18:01
So I would say it is important

00:25:18:02 00:25:19:01
to every business

00:25:19:23 00:25:21:21
role to think about

00:25:21:22 00:25:23:09
what is that we're doing?

00:25:23:11 00:25:25:11
And it's important to understand,

00:25:25:14 00:25:27:13
at least on very high level, what

00:25:27:14 00:25:29:10
is AI doing, which

00:25:29:14 00:25:31:07
which with which you're working.

00:25:31:12 00:25:33:08
In many cases, managers

00:25:33:09 00:25:35:06
and other

00:25:35:07 00:25:37:09
industry pros don't really

00:25:37:10 00:25:39:03
know very well what's happening, and

00:25:39:12 00:25:41:16
it is up to higher management

00:25:41:19 00:25:43:09
to educate to the

00:25:44:22 00:25:46:15
level people understand

00:25:46:19 00:25:48:04
what are the consequences and

00:25:48:17 00:25:50:18
not only short term like tomorrow

00:25:50:19 00:25:52:11
consequences, but more longitudinal

00:25:53:05 00:25:54:05
consequences.

00:25:54:12 00:25:56:06
If I use this data on

00:25:56:13 00:25:59:02
what will be impact on the society

00:25:59:03 00:26:00:18
or on my clients

00:26:01:04 00:26:03:02
next year, a two and 10

00:26:03:03 00:26:04:03
years, unfortunately,

00:26:05:04 00:26:06:06
there are not many use

00:26:07:05 00:26:08:15
cases now.

00:26:08:16 00:26:11:02
And in general, management

00:26:11:03 00:26:13:01
is not aware what

00:26:13:02 00:26:14:03
can go wrong.

00:26:14:14 00:26:15:14
But it's probably

00:26:16:20 00:26:18:15
our role as

00:26:18:24 00:26:20:15
educators and

00:26:20:21 00:26:22:20
institutions to build more.

00:26:22:21 00:26:24:18
This case is to say, OK, that

00:26:25:00 00:26:26:07
there are so many things that

00:26:27:03 00:26:28:18
can go wrong and we need to think

00:26:29:04 00:26:30:10
about them now.

00:26:30:17 00:26:31:22
And the role of education

00:26:32:23 00:26:34:16
here is super important.

00:26:35:19 00:26:37:05
So the extent to which the

00:26:37:17 00:26:39:12
consequences of the

00:26:39:13 00:26:41:11
almost proximity to potential

00:26:42:01 00:26:43:13
consequences for human beings,

00:26:43:24 00:26:46:02
you know, your customers, society

00:26:46:19 00:26:48:17
and citizens, you know,

00:26:48:18 00:26:50:10
that suggests, you know, how

00:26:50:24 00:26:52:22
in some sense, seriously,

00:26:53:11 00:26:54:18
this needs to be thought through.

00:26:54:19 00:26:55:25
And then I guess that's Yasmeen's

00:26:56:07 00:26:58:05
point. Then, Natalia, your point is,

00:26:58:06 00:26:59:01
well, actually, it's everyone's

00:26:59:02 00:27:00:01
responsibility.

00:27:00:04 00:27:01:17
This is not something that just

00:27:02:01 00:27:03:24
has to exist in sort of the

00:27:04:02 00:27:05:18
technical parts of an organisation

00:27:05:19 00:27:07:08
with the computer scientists or data

00:27:07:09 00:27:08:18
scientists and engineers who

00:27:09:06 00:27:10:05
have to certainly think about it

00:27:10:06 00:27:11:06
from certain dimensions and govern

00:27:11:07 00:27:12:08
it in certain ways.

00:27:12:14 00:27:13:24
It's not just for, sort of, the

00:27:14:00 00:27:15:21
middle and upper management to sort

00:27:15:22 00:27:18:02
of say, hey, we need to impose

00:27:18:03 00:27:19:14
these rules and these regulations,

00:27:19:15 00:27:20:14
these frameworks.

00:27:20:17 00:27:22:07
It's for everyone, everyone who's

00:27:22:13 00:27:24:11
thinking about how data may be used,

00:27:25:03 00:27:26:17
how algorithms may be used to to

00:27:26:24 00:27:28:13
inform or make decisions or make

00:27:28:23 00:27:30:23
recommendations and so on and so

00:27:30:24 00:27:32:13
forth, which I think is is a very,

00:27:32:14 00:27:33:18
very important practical

00:27:34:04 00:27:35:04
perspectives.

00:27:35:18 00:27:37:09
It's sort of a top down and bottom

00:27:37:10 00:27:40:02
up technical and non-technical

00:27:40:08 00:27:41:19
set of things to think about.

00:27:42:00 00:27:43:13
The other point I note, Natalia, is

00:27:43:14 00:27:45:05
to never forget about the law,

00:27:45:10 00:27:46:22
what I call the law of unintended

00:27:46:23 00:27:47:23
consequences.

00:27:48:04 00:27:49:09
You know, we may see that through

00:27:49:10 00:27:50:21
the lens of uncertainty, but we may

00:27:50:22 00:27:52:08
also see that as a

00:27:53:05 00:27:55:11
almost a proposition or a challenge

00:27:55:18 00:27:57:06
to people, an organisation to say,

00:27:57:07 00:27:59:02
OK, well, OK, no, it's not

00:27:59:03 00:28:00:20
just about what's obvious that could

00:28:00:21 00:28:02:03
go wrong here. And therefore, how do

00:28:02:04 00:28:03:14
we prevent that. But what what are

00:28:03:19 00:28:05:15
kind of the less obvious things and

00:28:05:16 00:28:06:21
maybe to think a little bit more

00:28:07:17 00:28:09:11
out of the box on on those

00:28:09:12 00:28:11:08
unintended consequences so

00:28:11:09 00:28:12:07
you can have everything on your

00:28:12:08 00:28:13:18
radar. But at least expanding

00:28:14:06 00:28:16:07
that set of possibilities may indeed

00:28:16:18 00:28:18:09
be another way to practically help

00:28:18:10 00:28:20:10
thinking about implementing

00:28:20:18 00:28:22:19
responsible, ethical AI.

00:28:23:12 00:28:24:16
So I think that's all, I got more

00:28:24:17 00:28:26:06
questions for you, but I actually

00:28:26:10 00:28:28:02
want to come to a question that was

00:28:28:11 00:28:29:19
posed by one of our diploma

00:28:30:11 00:28:31:23
students, Joanna, who

00:28:32:08 00:28:33:19
is in the US at the moment.

00:28:34:01 00:28:35:14
And it's sort of to this point,

00:28:35:22 00:28:37:08
I think, about everyone being

00:28:37:11 00:28:38:21
responsible. But her question is,

00:28:38:22 00:28:40:00
well, who needs to be in

00:28:40:18 00:28:42:02
the room? How to how do you get the

00:28:42:03 00:28:43:17
right people or the right personas

00:28:44:13 00:28:46:03
essentially in the room to

00:28:46:17 00:28:48:11
to express their concerns, to think

00:28:48:12 00:28:49:19
about perhaps these unintended

00:28:49:20 00:28:51:14
consequences within an organisation?

00:28:52:02 00:28:53:11
So if it's not just the technical

00:28:53:12 00:28:55:08
people, if it's not just,

00:28:55:18 00:28:57:07
say, middle and upper management,

00:28:57:15 00:28:58:15
who should basically

00:28:59:16 00:29:00:22
have a voice in governing

00:29:01:20 00:29:03:02
this within organisations?

00:29:03:07 00:29:04:17
I think it's it's a really important

00:29:05:01 00:29:06:08
question in terms of how we

00:29:06:15 00:29:08:07
implement it. Yasmeen, what do you

00:29:08:10 00:29:09:10
think?

00:29:10:01 00:29:11:17
That's that's a good question, and

00:29:12:03 00:29:13:21
it's I think it's

00:29:14:05 00:29:15:09
back to what you were saying,

00:29:15:10 00:29:16:23
Andrew, it's really important that

00:29:16:24 00:29:18:14
it's not just the analysts or the

00:29:18:18 00:29:20:13
data scientists who are thinking

00:29:20:14 00:29:22:05
about the implications of

00:29:23:07 00:29:25:15
the ethics or the fairness

00:29:25:25 00:29:27:18
of algorithms being applied.

00:29:27:24 00:29:29:09
In fact, it's more important

00:29:29:25 00:29:31:17
than ever to have business leaders,

00:29:32:01 00:29:33:23
business people

00:29:33:24 00:29:35:05
in the room to discuss the

00:29:35:10 00:29:36:24
implications of analytics and

00:29:37:00 00:29:38:22
algorithms and being applied,

00:29:39:05 00:29:40:10
because typically it's those

00:29:40:11 00:29:42:01
business leaders and those business

00:29:42:02 00:29:43:21
people who really understand

00:29:44:02 00:29:45:05
the application and how

00:29:46:10 00:29:48:09
that application will go out

00:29:48:10 00:29:49:07
in the world.

00:29:49:08 00:29:50:11
And so what we found it

00:29:51:04 00:29:53:11
helpful to do is focus

00:29:53:16 00:29:55:13
your key stakeholders on

00:29:55:20 00:29:57:00
specifics of what is that

00:29:57:18 00:29:59:08
use case that you're trying to drive

00:29:59:09 00:30:00:13
and what's the outcome.

00:30:00:22 00:30:02:17
And when thinking about

00:30:03:12 00:30:04:15
the the the responsible

00:30:05:19 00:30:07:19
use of algorithms, it's

00:30:07:20 00:30:09:02
useful to think about what if the

00:30:09:03 00:30:10:09
algorithm got it wrong.

00:30:11:06 00:30:12:16
And in analytical terms, when

00:30:13:05 00:30:15:04
we're using algorithms, we often

00:30:15:05 00:30:17:00
talk about false positives or false

00:30:17:01 00:30:18:01
negatives.

00:30:18:02 00:30:19:16
What's the chance that if we predict

00:30:19:17 00:30:20:16
something to be true,

00:30:21:21 00:30:22:21
it's actually not true?

00:30:23:02 00:30:25:03
It's false positive or a false

00:30:25:04 00:30:27:06
negative, If we say it's not true,

00:30:27:09 00:30:28:21
but it does actually happen to be

00:30:28:22 00:30:29:21
true.

00:30:29:23 00:30:31:14
What's - with this

00:30:31:24 00:30:33:16
this application of the algorithm,

00:30:33:20 00:30:35:02
what's acceptable to us as

00:30:35:18 00:30:37:11
a business? And the use case often

00:30:37:12 00:30:38:25
drives that acceptability.

00:30:39:10 00:30:41:06
So think about in health care.

00:30:41:12 00:30:43:09
We've seen use cases around

00:30:43:10 00:30:44:21
A.I. and algorithms being used

00:30:45:10 00:30:48:01
and applied to mammograms

00:30:48:02 00:30:49:08
or CT scans to look at are

00:30:49:24 00:30:52:12
there abnormal or abnormalities

00:30:52:19 00:30:53:21
with that image.

00:30:54:06 00:30:56:09
And so actually to get

00:30:56:10 00:30:57:09
a

00:30:58:08 00:30:59:24
false negative where the algorithm

00:31:00:01 00:31:01:05
said there's nothing wrong, but

00:31:01:06 00:31:03:01
there actually is in a health

00:31:03:02 00:31:04:11
care situation that might not be

00:31:04:12 00:31:06:02
acceptable, you need a high level

00:31:06:11 00:31:08:01
of accuracy for an algorithm.

00:31:08:12 00:31:10:07
So at that point, if

00:31:10:08 00:31:11:19
you're able to, as a business,

00:31:12:01 00:31:13:08
express what the algorithm

00:31:14:10 00:31:16:09
can run, but there's a level

00:31:16:10 00:31:17:20
of accuracy that's acceptable

00:31:18:10 00:31:20:00
to the business problem that then

00:31:20:10 00:31:22:07
gives you your data science through

00:31:22:08 00:31:23:15
analytics teams a direction

00:31:24:07 00:31:25:07
to say, OK, we

00:31:26:06 00:31:27:21
can go into testing algorithms in

00:31:27:24 00:31:29:15
this area, but until we don't get

00:31:29:24 00:31:31:14
to the level of specificity

00:31:31:20 00:31:33:06
and the business teams can can

00:31:33:19 00:31:35:09
help to define what that level of

00:31:35:10 00:31:36:22
specificity is,

00:31:37:06 00:31:39:02
that algorithm is not

00:31:39:03 00:31:40:25
acceptable for for

00:31:41:16 00:31:43:13
production or real real

00:31:43:14 00:31:44:14
world use.

00:31:44:22 00:31:45:23
And so I describe that

00:31:46:19 00:31:48:05
scenario because I think it's a

00:31:48:06 00:31:49:06
useful way.

00:31:50:04 00:31:51:22
I liked how Felipe, I got a preview

00:31:51:24 00:31:53:13
of the paper that

00:31:54:00 00:31:55:14
the team have been working on at

00:31:55:16 00:31:56:24
Oxford with the ICC

00:31:57:13 00:31:59:09
and that Felipe described

00:31:59:10 00:32:00:09
it earlier.

00:32:00:11 00:32:02:12
You need a way of making

00:32:02:13 00:32:04:15
these big conversations about ethics

00:32:04:16 00:32:06:07
and fairness, etc., more tangible,

00:32:07:01 00:32:07:25
more real world.

00:32:07:26 00:32:09:15
How can you apply them in business?

00:32:10:13 00:32:11:21
Getting into a little bit more of

00:32:11:22 00:32:13:12
the details around the algorithm on

00:32:13:13 00:32:15:11
what level of accuracy is acceptable

00:32:15:17 00:32:17:15
helps to frame the conversation

00:32:17:16 00:32:20:03
in the implementation.

00:32:20:04 00:32:21:10
How can we move forward?

00:32:21:11 00:32:23:02
What's the level of acceptability?

00:32:24:11 00:32:25:09
Thanks Yasmeen. I think that's

00:32:25:10 00:32:27:02
really, really helpful way to think

00:32:27:05 00:32:29:00
about these errors.

00:32:29:09 00:32:30:17
In statistics we would call them

00:32:30:18 00:32:32:00
type one and type two errors with

00:32:32:01 00:32:33:07
sort of false positives and false

00:32:33:08 00:32:34:18
negatives. And and what might

00:32:35:03 00:32:36:21
the consequences of those be if the

00:32:36:24 00:32:38:21
algorithm is wrong, for whatever

00:32:38:22 00:32:39:21
reason, whether it's the wrong

00:32:39:22 00:32:41:13
algorithm or the data is not quite

00:32:41:14 00:32:43:03
right or there's some bias

00:32:43:10 00:32:45:07
for one way or one, one reason or

00:32:45:08 00:32:46:02
another.

00:32:46:03 00:32:48:01
But then in practical

00:32:48:02 00:32:49:12
terms, like, well, if it gets it

00:32:49:13 00:32:51:05
wrong, what does that mean and what

00:32:51:08 00:32:52:08
is our tolerance or

00:32:53:04 00:32:54:15
required level of precision?

00:32:54:16 00:32:55:10
If you want to think about

00:32:55:11 00:32:57:03
differently, that brings it down to

00:32:57:10 00:32:58:20
sort of conversations that you

00:32:58:21 00:32:59:17
actually have.

00:32:59:18 00:33:01:14
And back to Joanna's question,

00:33:01:15 00:33:03:04
I think the people in the room need

00:33:03:05 00:33:04:15
to be the people who can really talk

00:33:04:16 00:33:06:07
about what those real life

00:33:06:08 00:33:07:22
consequences would would actually

00:33:08:14 00:33:10:04
be. What would those errors

00:33:10:10 00:33:12:18
mean to the customers, to

00:33:12:19 00:33:14:05
other people who might be affected,

00:33:14:06 00:33:15:14
maybe to your employees, to

00:33:16:01 00:33:17:21
regulators, to whoever else might be

00:33:17:22 00:33:18:23
a relevant stakeholder.

00:33:19:04 00:33:20:15
So I guess the point is, have that

00:33:20:16 00:33:22:13
diversity of opinion and

00:33:22:18 00:33:24:09
that multistakeholder perspective,

00:33:24:13 00:33:26:13
you know, quote unquote in the room

00:33:27:03 00:33:28:08
to think about these things.

00:33:28:09 00:33:29:11
But the question actually is, what

00:33:29:12 00:33:30:22
do you how do you frame the

00:33:30:23 00:33:31:17
questions?

00:33:31:18 00:33:33:10
How do you get them to think about

00:33:33:11 00:33:35:06
these issues? And I think that, sort

00:33:35:07 00:33:36:23
of false positives, false negatives,

00:33:37:07 00:33:38:25
you know, is a really useful way of

00:33:39:00 00:33:40:13
thinking. So we're about at the

00:33:40:14 00:33:41:22
midpoint of our broadcast, I just

00:33:41:23 00:33:43:05
wanted to welcome anyone who's

00:33:43:06 00:33:44:05
joined us.

00:33:44:09 00:33:45:14
You know, since the beginning,

00:33:46:04 00:33:47:11
you're obviously watching Leadership

00:33:47:12 00:33:48:16
in Extraordinary Times here at the

00:33:48:17 00:33:49:24
Said Business School at the

00:33:50:00 00:33:51:17
University of Oxford, I'm Professor

00:33:52:01 00:33:53:01
Andrew Stephen, the L'Oreal

00:33:53:08 00:33:54:20
Professor of Marketing and the

00:33:54:24 00:33:56:11
Research Dean. And I have been

00:33:56:12 00:33:58:06
talking with Dr. Yasmeen Ahmad

00:33:58:13 00:34:00:17
from Teradata and Dr Natalia

00:34:01:19 00:34:03:03
Eframova from here at the Said

00:34:03:04 00:34:04:03
Business School.

00:34:04:08 00:34:05:12
And we're talking about how
businesses

00:34:05:13 00:34:06:24
can use AI responsibly.

00:34:07:09 00:34:09:01
I want to now go to have a look at

00:34:09:02 00:34:10:02
our poll and see what

00:34:11:02 00:34:12:25
you all thought about

00:34:12:26 00:34:14:10
who should be responsible for

00:34:14:11 00:34:15:11
responsible AI.

00:34:15:13 00:34:16:21
So if we can take a look at

00:34:17:09 00:34:18:22
the results and then I'm going to

00:34:18:23 00:34:19:22
see what both

00:34:21:16 00:34:22:12
Natalia and Yasmeen think about this.

00:34:22:13 00:34:24:15
So there's no clear

00:34:24:16 00:34:25:24
frontrunner, I suppose, here.

00:34:26:00 00:34:27:18
But the government is is on top,

00:34:28:01 00:34:29:14
followed by the global

00:34:29:15 00:34:31:09
intergovernmental organisations, the

00:34:31:10 00:34:33:00
tech companies, industry bodies

00:34:33:07 00:34:35:09
and individuals last.

00:34:35:10 00:34:37:06
But quite a spread

00:34:37:07 00:34:39:07
here. So to be honest,

00:34:39:08 00:34:40:12
that's not exactly what I

00:34:41:04 00:34:42:04
expected to see.

00:34:42:14 00:34:43:17
What do you think, Natalia?

00:34:43:18 00:34:45:03
What's your reaction to

00:34:45:20 00:34:46:20
our poll results?

00:34:48:03 00:34:49:18
Well, that's very interesting.

00:34:49:19 00:34:51:04
I also didn't expect this

00:34:51:16 00:34:53:17
and I would say, well,

00:34:53:18 00:34:55:06
that would be amazing if a

00:34:55:07 00:34:56:18
government was able to control

00:34:57:08 00:34:58:08
it and provide

00:34:59:09 00:35:01:03
us all guidelines what we should be

00:35:01:04 00:35:02:04
doing.

00:35:02:17 00:35:04:09
Unfortunately, for now, it's not the

00:35:04:10 00:35:06:02
case. We do have some

00:35:07:04 00:35:09:18
recommendations from policymakers,

00:35:10:19 00:35:12:00
and I think that's amazing

00:35:12:17 00:35:14:05
that we are moving towards

00:35:14:15 00:35:15:17
these same direction.

00:35:15:23 00:35:18:12
But unfortunately, currently,

00:35:19:10 00:35:21:05
probably the only regulations

00:35:21:06 00:35:23:16
like hard regulations that exist

00:35:24:10 00:35:26:10
inside the companies inside big

00:35:26:18 00:35:28:19
technical industry

00:35:28:20 00:35:29:21
organisations.

00:35:31:25 00:35:33:00
Intergovernmental and

00:35:34:01 00:35:35:18
international bodies

00:35:36:17 00:35:38:11
do catch up slowly

00:35:38:12 00:35:40:08
and they do

00:35:41:03 00:35:42:13
publish their own guidelines

00:35:42:24 00:35:45:04
and sets of recommendations.

00:35:45:23 00:35:47:21
What I believe is that

00:35:47:22 00:35:48:23
inputs from all levels

00:35:50:00 00:35:51:02
are very important and

00:35:52:01 00:35:53:20
why, from a technical perspective, I

00:35:53:24 00:35:55:06
believe that it's difficult for

00:35:55:07 00:35:57:17
governments to do is

00:35:58:12 00:36:00:16
they really don't have

00:36:00:17 00:36:01:16
a tool for now to

00:36:02:12 00:36:04:09
check what is

00:36:04:14 00:36:05:14
AI doing.

00:36:05:24 00:36:07:03
So no one outside of the

00:36:07:22 00:36:09:14
organisation can

00:36:09:25 00:36:11:23
check what the algorithm

00:36:11:24 00:36:12:24
is doing. And even if

00:36:13:23 00:36:15:13
someone wants to check the code,

00:36:15:20 00:36:17:21
the problem can be not in the code,

00:36:18:03 00:36:20:00
but in the data or

00:36:20:10 00:36:22:04
elsewhere in the operations or

00:36:22:05 00:36:23:05
production line.

00:36:23:15 00:36:26:05
So it's very, very complex problem.

00:36:26:12 00:36:28:15
And until we

00:36:28:16 00:36:30:17
develop this centralised

00:36:30:24 00:36:32:06
understanding, centralised

00:36:32:09 00:36:33:14
guidelines, how AI should

00:36:34:07 00:36:35:12
be developed, the problem

00:36:36:06 00:36:37:18
that the responsibilities would be

00:36:37:19 00:36:39:17
largely on the companies themselves

00:36:39:25 00:36:41:10
because they know their business is

00:36:41:11 00:36:42:10
the best.

00:36:42:23 00:36:44:02
Yasmeen, what do you think?

00:36:45:17 00:36:47:14
I have to agree with Natalia there.

00:36:47:15 00:36:48:23
It's such a complex problem

00:36:49:20 00:36:52:03
to try and find a regulation

00:36:52:04 00:36:53:21
or a framework that would cover

00:36:54:04 00:36:55:12
the amount of innovation that is

00:36:55:13 00:36:56:14
happening in the space at the

00:36:56:15 00:36:57:13
moment.

00:36:57:14 00:36:59:00
It's very hard for a government to

00:36:59:01 00:37:00:22
regulate. And having said that,

00:37:00:23 00:37:02:17
regulation often happens

00:37:02:18 00:37:03:17
after the fact.

00:37:03:18 00:37:05:08
Regulation looks at how

00:37:05:09 00:37:06:16
technologies,

00:37:07:13 00:37:09:16
how digital

00:37:09:22 00:37:11:23
is being leveraged and then looks

00:37:11:24 00:37:13:11
to regulate uses,

00:37:13:19 00:37:15:19
so I think companies do have a

00:37:15:20 00:37:17:16
real responsibility to make

00:37:17:17 00:37:20:07
sure that they are responsible

00:37:20:08 00:37:21:23
and accountable for the algorithms

00:37:22:10 00:37:23:19
that they're developing and how they

00:37:23:20 00:37:24:19
are applying them.

00:37:24:25 00:37:27:12
And as I think about organisations,

00:37:27:13 00:37:29:11
I also I don't

00:37:29:12 00:37:31:17
think it's about just the strict

00:37:31:18 00:37:33:07
frameworks or guidelines.

00:37:33:08 00:37:34:20
I think all large organisations

00:37:35:09 00:37:37:12
will have a ethics framework.

00:37:38:08 00:37:40:04
It was mentioned earlier, often

00:37:40:05 00:37:41:21
the foundations have been there in

00:37:41:24 00:37:43:07
the organisation, they need to be

00:37:43:08 00:37:44:23
evolved. The ethics frameworks need

00:37:44:24 00:37:46:23
to be evolved to take into

00:37:46:24 00:37:48:18
account new risks or new factors

00:37:48:19 00:37:50:19
that have come up because of A.I.

00:37:50:20 00:37:51:20
and digital.

00:37:52:01 00:37:53:07
So, yes, there's existing

00:37:53:08 00:37:54:18
frameworks. They need to evolve.

00:37:54:22 00:37:56:09
But we also need to look at company

00:37:56:10 00:37:58:12
cultures and how a company

00:37:58:13 00:38:00:14
leads through this evolution

00:38:00:15 00:38:02:04
and the change because AI digital

00:38:02:20 00:38:04:18
analytics is now impacting

00:38:04:23 00:38:06:12
all aspects of our lives, all

00:38:06:13 00:38:08:05
aspects of organisations.

00:38:08:14 00:38:10:16
And so some of the

00:38:10:17 00:38:12:07
ethics in the fair and

00:38:12:12 00:38:13:14
equal and unbiased use

00:38:14:08 00:38:15:24
of algorithms needs to be embedded

00:38:16:06 00:38:18:04
in company culture and needs to

00:38:18:06 00:38:19:11
be embedded in how people

00:38:20:09 00:38:21:17
build these tools, how they

00:38:22:07 00:38:24:03
apply the algorithms, how they

00:38:24:11 00:38:25:20
analyse data.

00:38:26:07 00:38:28:04
To Natalia's point, it's so complex,

00:38:28:05 00:38:29:11
there's so many touchpoint, there's

00:38:29:12 00:38:30:23
so many people that are involved in

00:38:30:24 00:38:32:15
the process, it needs to be part of

00:38:32:16 00:38:34:17
the the DNA of the organisation

00:38:34:18 00:38:36:20
to want to have

00:38:37:02 00:38:38:22
ethical use of these new

00:38:39:04 00:38:40:17
technologies, new tools at their

00:38:40:18 00:38:41:17
disposal.

00:38:42:03 00:38:44:04
So there's a question

00:38:44:05 00:38:45:24
that's come up from from Richard in

00:38:46:00 00:38:47:12
my home country of Australia, in

00:38:47:13 00:38:49:01
fact, who was asking about,

00:38:49:12 00:38:50:16
you know, this point about

00:38:50:17 00:38:52:05
innovation that both of you talked

00:38:52:06 00:38:53:21
about. So if we regulate,

00:38:54:16 00:38:56:12
I guess, too much, the worry is

00:38:56:13 00:38:57:22
that will sort of box in

00:38:58:08 00:38:59:23
the opportunities that might

00:38:59:24 00:39:01:18
therefore stifle

00:39:01:24 00:39:02:19
innovation.

00:39:02:20 00:39:04:17
So I guess it feels like

00:39:04:18 00:39:06:05
it's this balance of, yes, we need

00:39:06:06 00:39:07:20
to have some, I guess, guidance from

00:39:08:05 00:39:09:20
governments and intergovernmental

00:39:09:21 00:39:11:22
organisations, but obviously can't

00:39:11:23 00:39:12:16
cover everything.

00:39:12:17 00:39:14:15
And nor, nor, probably should they.

00:39:15:05 00:39:16:16
And I guess these regulations

00:39:16:23 00:39:18:23
because of of the need

00:39:18:24 00:39:20:07
for further innovation and the fact

00:39:20:08 00:39:22:01
that that is happening anyway,

00:39:22:10 00:39:23:22
we need to sort of have these as

00:39:23:23 00:39:25:18
living regulations in some

00:39:25:19 00:39:27:19
sense or guidelines around that.

00:39:28:01 00:39:30:01
And I think to me, my take on

00:39:30:05 00:39:31:15
the poll result is actually

00:39:32:01 00:39:32:24
it's sort of everyone's

00:39:33:00 00:39:34:08
responsibility in one way or

00:39:34:09 00:39:35:11
another. So I think, as I was

00:39:35:12 00:39:36:21
saying, we need multiple voices in

00:39:36:22 00:39:37:23
the room to think about those

00:39:37:24 00:39:39:24
consequences and unintended

00:39:40:00 00:39:41:23
ones and and the false

00:39:41:24 00:39:43:07
positives and false negatives.

00:39:43:15 00:39:44:23
I guess the point is that that

00:39:44:24 00:39:46:11
higher level, we also need

00:39:46:19 00:39:48:07
this to not only be

00:39:48:15 00:39:50:07
something that governments do, not

00:39:50:08 00:39:52:02
only something that

00:39:52:07 00:39:53:24
intergovernmental organisations do,

00:39:54:00 00:39:56:04
but indeed something that companies

00:39:56:05 00:39:57:11
do, groups of companies or

00:39:57:23 00:39:58:24
organisations,

00:39:59:20 00:40:01:12
the tech companies and so on and so

00:40:01:13 00:40:03:09
on. So I think there's there's

00:40:03:10 00:40:05:08
a need for this to be,

00:40:05:15 00:40:06:17
I guess, governed

00:40:07:25 00:40:09:02
with with a small G in

00:40:10:00 00:40:11:17
a in a very collective way, which,

00:40:11:24 00:40:12:24
of course, it's going to be fraught

00:40:12:25 00:40:14:15
with difficulties given all those

00:40:14:21 00:40:15:19
different stakeholders.

00:40:15:20 00:40:17:17
But I like what I'm hearing is that

00:40:17:18 00:40:19:10
that seems to be a ways forward.

00:40:19:20 00:40:21:06
I want to bring back Felipe

00:40:22:00 00:40:23:24
now and bring him

00:40:24:00 00:40:25:09
back into the conversation,

00:40:26:07 00:40:27:18
because I think we've just got quite

00:40:27:19 00:40:29:03
a few questions now that have been

00:40:29:04 00:40:30:03
popping up.

00:40:30:10 00:40:31:17
Thank you, to those of you who've

00:40:31:18 00:40:33:04
asked questions that I think we can

00:40:33:13 00:40:34:25
we can start to go through.

00:40:35:20 00:40:37:00
So, Felipe, welcome back.

00:40:37:04 00:40:38:15
Just quickly, before we go on to

00:40:38:22 00:40:39:21
some more questions, what was your

00:40:39:22 00:40:41:09
take on those poll results?

00:40:42:07 00:40:43:20
Very similar to everybody else's.

00:40:44:07 00:40:45:21
I think I have very similar read to

00:40:45:22 00:40:47:23
you in that it does seem like

00:40:47:24 00:40:49:20
it is very much like it's a joint

00:40:49:21 00:40:51:18
responsibility that everybody

00:40:52:00 00:40:53:15
has to carry some of the burden.

00:40:53:24 00:40:55:20
The other bit that I think

00:40:55:21 00:40:57:12
going to that first point that we

00:40:57:13 00:40:58:17
have, you know, governments have

00:40:58:18 00:41:00:12
been the most responsible, even

00:41:00:21 00:41:02:04
by a narrow margin, I guess

00:41:02:21 00:41:04:09
a bit of a word of caution or

00:41:04:10 00:41:05:09
concern here from

00:41:06:07 00:41:08:08
somebody that hasn't lived forever

00:41:08:09 00:41:09:13
in the developed world.

00:41:09:24 00:41:11:07
Not everybody's government is

00:41:11:08 00:41:12:03
fantastic.

00:41:12:04 00:41:13:17
Not everybody's government has your

00:41:13:18 00:41:14:18
best interests in mind.

00:41:15:05 00:41:17:08
So just saying necessarily that

00:41:17:18 00:41:19:03
whose government is going to decide

00:41:19:04 00:41:20:11
what is correct to do?

00:41:20:18 00:41:22:06
I think that's something to worry

00:41:22:07 00:41:23:11
about. That's something that I put

00:41:23:12 00:41:25:05
out there to say, you know, yes,

00:41:25:06 00:41:26:08
regulation matters very

00:41:27:01 00:41:28:02
much to your point, that

00:41:28:24 00:41:30:10
there is a concern against the

00:41:30:19 00:41:32:10
attention against innovation.

00:41:32:21 00:41:34:13
But some companies are going to be

00:41:34:19 00:41:36:24
restricted exclusively to their

00:41:37:05 00:41:38:17
regional domains and they will

00:41:39:03 00:41:40:20
have just one set of legislation to

00:41:40:21 00:41:41:16
worry about.

00:41:41:17 00:41:43:24
A number of companies are going to

00:41:44:02 00:41:46:06
exist across national boundaries.

00:41:46:15 00:41:48:17
And then you run

00:41:48:18 00:41:50:06
into issues where you're picking and

00:41:50:07 00:41:51:06
choosing to the bare

00:41:52:05 00:41:54:03
minimum of legal requirement rather

00:41:54:04 00:41:55:24
than doing the ethical thing or

00:41:56:00 00:41:57:12
doing the appropriate thing for

00:41:57:13 00:41:58:17
yourself, for your company, et

00:41:58:18 00:42:00:04
cetera. By just forcing others

00:42:00:15 00:42:01:14
to say, all right, what's the bare

00:42:01:15 00:42:02:18
minimum I have to do?

00:42:03:01 00:42:04:06
And that's what you do when you

00:42:04:07 00:42:06:03
exploit everybody that's kind of

00:42:06:04 00:42:07:14
moving away from ethics and just

00:42:07:21 00:42:09:02
saying, hey, just let let

00:42:09:19 00:42:11:06
me make as much money as I can as

00:42:11:07 00:42:12:14
quickly as I can until somebody

00:42:12:15 00:42:13:14
catches on.

00:42:14:13 00:42:15:22
So good point, you know, not all

00:42:15:23 00:42:16:22
governments are

00:42:17:19 00:42:19:13
necessarily the right governments, I

00:42:19:15 00:42:20:20
suppose you could say here.

00:42:21:00 00:42:22:13
And then just to reinforce your

00:42:22:19 00:42:24:07
point about large organisations, I

00:42:24:08 00:42:26:05
think as we've seen actually

00:42:26:06 00:42:27:23
with other areas of responsibility,

00:42:28:09 00:42:30:01
such as environment and

00:42:30:20 00:42:32:04
social responsibility, I think about

00:42:32:05 00:42:33:04
ESG.

00:42:33:05 00:42:35:03
We see large multinationals

00:42:35:11 00:42:37:12
actually having

00:42:37:13 00:42:39:08
a big impact in different

00:42:39:09 00:42:40:21
countries just because of their

00:42:40:22 00:42:42:04
global reach. So I think there is a

00:42:42:11 00:42:44:16
an important role of business,

00:42:45:05 00:42:47:02
particularly businesses with

00:42:47:13 00:42:48:17
an international footing

00:42:49:20 00:42:51:18
to sort of lead by example

00:42:52:05 00:42:53:15
in a lot of this and of course,

00:42:53:16 00:42:54:12
collaborate with those other

00:42:54:13 00:42:56:05
stakeholders that we know we talked

00:42:56:06 00:42:57:14
about. So but speaking about

00:42:57:21 00:42:59:09
government and there's a question

00:42:59:18 00:43:01:22
from from Andrew, who's one of our

00:43:02:07 00:43:04:09
executive students here, who does

00:43:04:10 00:43:06:06
work for a government and

00:43:06:07 00:43:07:04
is asking, should there be a

00:43:07:05 00:43:09:02
hierarchy of government concern,

00:43:09:15 00:43:11:00
starting with regulation

00:43:11:11 00:43:13:16
for physical risk to the individual?

00:43:13:17 00:43:14:20
So in other words, I guess more

00:43:14:21 00:43:16:17
broadly, maybe government's thinking

00:43:16:18 00:43:18:03
about sort of that type of harm

00:43:18:13 00:43:19:18
organisation's think about other

00:43:19:19 00:43:20:25
types of harm, perhaps.

00:43:21:19 00:43:23:06
I don't know. Let's start with

00:43:23:15 00:43:25:09
Yasmeen and then I want to want to

00:43:25:10 00:43:26:08
hear what the others also think

00:43:26:09 00:43:27:05
about this, because I think it's a

00:43:27:06 00:43:28:04
really important, practical

00:43:28:05 00:43:29:04
question.

00:43:29:12 00:43:30:05
Yasmeen?

00:43:30:06 00:43:31:24
That's a great question,

00:43:32:03 00:43:34:06
because the implications

00:43:34:07 00:43:35:13
of how algorithms are used

00:43:36:02 00:43:38:03
is a multidimensional

00:43:38:12 00:43:40:11
challenge of there's various

00:43:40:12 00:43:41:22
types of risks that

00:43:42:07 00:43:43:13
can be created.

00:43:43:20 00:43:45:16
So as I think about even our

00:43:45:17 00:43:47:18
organisation and our risk framework,

00:43:47:19 00:43:49:18
there are a whole diverse set

00:43:49:19 00:43:50:19
of our categories of

00:43:51:16 00:43:53:14
risks that we consider to our

00:43:53:15 00:43:55:21
business and for a government.

00:43:55:22 00:43:57:19
I don't see why it wouldn't be the

00:43:57:20 00:43:59:04
same with AI.

00:43:59:12 00:44:01:12
So as I think about

00:44:02:00 00:44:03:09
different types of risk, the one

00:44:03:10 00:44:05:06
that popped into my mind, we did

00:44:05:07 00:44:07:09
a use case with a retailer around

00:44:07:17 00:44:08:20
wastage and we know how

00:44:09:16 00:44:11:17
much sustainability is important

00:44:11:18 00:44:12:14
right now.

00:44:12:15 00:44:13:25
And in this use case, we were

00:44:14:01 00:44:15:19
actually supporting the retailer

00:44:15:24 00:44:17:07
through AI to reduce waste,

00:44:18:01 00:44:20:06
to reduce how much grocery

00:44:20:07 00:44:21:20
products were thrown away at the end

00:44:21:21 00:44:22:18
of the day.

00:44:22:19 00:44:24:07
But equally on

00:44:24:16 00:44:26:09
vice versa, A.I.

00:44:26:12 00:44:27:24
has has also almost accelerated

00:44:29:04 00:44:31:00
or amplified the fast

00:44:31:04 00:44:32:10
fashion business and other

00:44:32:17 00:44:34:15
businesses that are creating a ton

00:44:34:16 00:44:35:15
of waste.

00:44:35:16 00:44:37:18
And so you can begin to think about

00:44:37:25 00:44:39:22
AI algorithms and the different

00:44:39:23 00:44:41:07
types of risk or implications

00:44:42:14 00:44:44:01
they have to societies.

00:44:44:08 00:44:46:04
There is the sustainability in

00:44:46:05 00:44:47:18
terms of green and environmental

00:44:48:01 00:44:49:06
impacts. There is impacts

00:44:50:02 00:44:52:05
to humans and peoples and jobs

00:44:52:06 00:44:53:05
and careers.

00:44:53:06 00:44:55:22
There's impact to diversity,

00:44:56:03 00:44:58:03
equality, inclusion.

00:44:58:15 00:45:00:10
So if I was to to

00:45:00:11 00:45:01:15
think about how to put that

00:45:01:16 00:45:03:05
framework together for government, I

00:45:03:06 00:45:04:08
would be thinking about those

00:45:04:09 00:45:05:09
different categories.

00:45:05:19 00:45:07:18
And in terms of prioritisation,

00:45:07:19 00:45:09:04
there's definitely some categories

00:45:09:05 00:45:10:19
that you may prioritise over and

00:45:11:02 00:45:13:09
above others, those that have impact

00:45:13:10 00:45:15:02
to human life, for example.

00:45:15:20 00:45:17:11
But equally, I think all of those

00:45:17:15 00:45:19:07
categories are important and

00:45:19:08 00:45:21:11
potentially require specialists

00:45:21:12 00:45:23:02
or experts or again, business

00:45:23:03 00:45:24:18
experts who understand those areas

00:45:25:03 00:45:26:03
and are able to fully

00:45:26:24 00:45:28:18
think through the implications

00:45:28:23 00:45:30:22
of A.I., which may

00:45:30:23 00:45:32:06
not be apparent initially to

00:45:32:21 00:45:34:14
you as a developer or

00:45:34:21 00:45:36:06
a builder of the tools or the

00:45:36:08 00:45:37:25
technology platforms, because this

00:45:38:04 00:45:40:04
is just such a new

00:45:40:05 00:45:42:08
application of analytics

00:45:42:09 00:45:43:17
and new areas.

00:45:45:08 00:45:46:15
So another question and this one I'm

00:45:46:16 00:45:48:08
going to direct at

00:45:48:12 00:45:49:08
you, Felipe.

00:45:49:09 00:45:51:02
This comes from Dennis who's here in

00:45:51:03 00:45:52:03
the U.K.

00:45:52:05 00:45:53:22
and he's asking or suggesting,

00:45:54:03 00:45:55:19
shouldn't all innovation be governed

00:45:55:20 00:45:57:06
by the purpose of the company,

00:45:57:22 00:45:59:15
hopefully being mindful to human

00:45:59:16 00:46:00:24
beings, then the planet and

00:46:01:14 00:46:03:21
then finally profit in that order?

00:46:04:09 00:46:05:09
What do you think?

00:46:05:24 00:46:08:02
I mean, I do like the word hopefully

00:46:08:03 00:46:08:24
there, right.

00:46:08:25 00:46:10:16
All discussions about ethics, none

00:46:11:03 00:46:12:03
of it matters

00:46:13:02 00:46:14:22
until there's a trade off being

00:46:14:23 00:46:16:09
made. So everybody agrees that

00:46:16:21 00:46:18:21
we should protect privacy until

00:46:18:22 00:46:21:00
it's 10 percent of your sales

00:46:21:06 00:46:22:25
that goes away if you take that

00:46:22:26 00:46:24:13
action, then suddenly everybody

00:46:24:24 00:46:26:24
in the boardroom just goes, maybe

00:46:27:00 00:46:28:18
we think about this some more.

00:46:29:06 00:46:30:23
So it's never a problem until it

00:46:30:24 00:46:32:05
touches kind of that money

00:46:32:10 00:46:33:10
component.

00:46:33:12 00:46:34:17
And that's that's a large

00:46:35:13 00:46:36:24
part of this initiative in this

00:46:37:00 00:46:38:07
project and the research in the

00:46:38:08 00:46:39:22
guidances is

00:46:40:03 00:46:41:10
I think ultimately we want

00:46:41:13 00:46:43:07
individuals to self-determine it and

00:46:43:08 00:46:44:11
say, yeah, your business

00:46:45:11 00:46:47:05
is going to decide what is the best

00:46:47:06 00:46:48:13
innovation and it will make

00:46:49:02 00:46:50:14
the right, good choices.

00:46:51:07 00:46:53:07
Frameworks like these are to help

00:46:53:16 00:46:55:06
people make the right choice

00:46:55:11 00:46:56:18
when the time comes.

00:46:57:11 00:46:58:17
If you're making a choice when

00:46:58:18 00:46:59:22
there's no consequence for your

00:46:59:23 00:47:01:04
business, if you're not giving

00:47:01:05 00:47:02:05
something up, then

00:47:03:02 00:47:04:08
you're not really facing a choice,

00:47:04:10 00:47:05:08
you're just doing the right thing

00:47:05:09 00:47:07:16
kind of with no cost.

00:47:07:23 00:47:09:07
I'm always more curious when it's

00:47:09:12 00:47:11:03
somebody's mortgage payment, if you

00:47:11:04 00:47:13:06
can't pay rent, if you lose your job

00:47:13:07 00:47:14:13
because you did the right thing,

00:47:14:14 00:47:15:24
that's your ethical quandary.

00:47:16:12 00:47:17:21
That's that's the thorn in the

00:47:17:22 00:47:18:17
question.

00:47:18:18 00:47:20:05
And that's when it matters having

00:47:20:06 00:47:21:19
this grounding of I know what my

00:47:22:00 00:47:23:14
company stands for, I know I'm going

00:47:23:15 00:47:24:17
to be backed for doing the right

00:47:24:18 00:47:26:15
thing, even if it's going to cost

00:47:26:16 00:47:27:16
us revenue and not

00:47:28:14 00:47:29:20
just because it was legally

00:47:29:21 00:47:31:18
required, that this is what

00:47:31:19 00:47:33:04
keeps me out of jail, because it's

00:47:33:05 00:47:34:19
the right thing to do in what you

00:47:34:20 00:47:36:21
said for people, for environment,

00:47:37:07 00:47:38:07
for et cetera.

00:47:39:07 00:47:41:10
So, again, comes back to that,

00:47:41:11 00:47:42:17
you know, it can't just be sort of

00:47:42:18 00:47:44:10
one set of rules imposed

00:47:44:17 00:47:46:08
from on high, whoever that happens

00:47:46:09 00:47:47:14
to be, whether it's government or

00:47:47:22 00:47:49:25
the organisation or some mix,

00:47:50:19 00:47:51:22
you need that bottom up.

00:47:51:23 00:47:52:18
Well, that's sort of you know,

00:47:52:19 00:47:54:08
everyone has to think about this,

00:47:54:15 00:47:56:09
but that's the tension as well,

00:47:56:14 00:47:58:02
because once you make it personal,

00:47:58:03 00:47:59:13
then, you know, people are

00:47:59:14 00:48:01:16
different. And so so that's that's

00:48:01:17 00:48:02:17
why this is messy

00:48:03:14 00:48:05:04
and complex. But that's why we do

00:48:05:11 00:48:06:23
need to be talking about these

00:48:07:00 00:48:08:23
issues and finding ways to take

00:48:08:24 00:48:09:23
action on them.

00:48:10:06 00:48:11:09
I'm going to go to another question

00:48:11:10 00:48:12:21
now. This one's for you, Natalia.

00:48:13:03 00:48:14:25
Is there a sense that algorithmic

00:48:14:26 00:48:16:12
tools and models are held to a

00:48:16:19 00:48:18:15
higher standard than

00:48:18:16 00:48:20:07
what would be applied to humans in

00:48:20:08 00:48:21:09
similar situations?

00:48:21:10 00:48:22:22
A bit of a human versus machine

00:48:23:17 00:48:25:17
type of type of consideration here.

00:48:26:10 00:48:28:11
It's an interesting question,

00:48:28:12 00:48:30:11
and I would say it depends

00:48:30:19 00:48:32:22
on what kind of algorithm

00:48:32:23 00:48:33:24
we are talking about.

00:48:35:08 00:48:36:08
So definitely some

00:48:37:05 00:48:39:05
algorithms that are performing

00:48:39:19 00:48:41:22
much, much better than humans.

00:48:41:23 00:48:43:15
Of course, they had

00:48:43:20 00:48:46:02
to be scrutinised more

00:48:46:03 00:48:48:01
because they just

00:48:48:08 00:48:49:22
work on a very, very different level

00:48:49:23 00:48:52:02
of accuracy. For some algorithms,

00:48:52:05 00:48:53:13
it's not not really, really

00:48:54:06 00:48:55:21
that important, like, think about

00:48:56:01 00:48:57:03
automatic irrigation systems.

00:48:57:20 00:48:59:18
Well, is it that important that

00:48:59:19 00:49:02:01
it gives the precise

00:49:02:02 00:49:04:05
amount of water plus two

00:49:04:06 00:49:05:06
minus three millilitres?

00:49:05:17 00:49:06:17
Probably not.

00:49:06:22 00:49:08:07
So it really, really matters

00:49:08:21 00:49:10:01
what kind of application we're

00:49:10:02 00:49:11:01
talking about.

00:49:11:08 00:49:12:22
Going to back to Yasmeen's point

00:49:13:10 00:49:15:07
about medical applications.

00:49:15:15 00:49:17:01
If we think about

00:49:17:16 00:49:19:11
human health and

00:49:20:13 00:49:22:08
X-ray analysis results,

00:49:23:11 00:49:24:20
is there like

00:49:25:12 00:49:26:23
a single chance that it can go

00:49:27:10 00:49:28:23
wrong? No, if

00:49:29:11 00:49:30:23
something can go wrong, that if

00:49:31:10 00:49:32:24
the algorithm can potentially be

00:49:33:07 00:49:35:04
incorrect, even this when

00:49:35:05 00:49:37:02
this chance is tiny, should

00:49:37:03 00:49:38:07
we go for this algorithm?

00:49:38:11 00:49:40:02
No. It really

00:49:41:09 00:49:43:17
depends on this large hierarchy

00:49:43:18 00:49:45:19
of risks and which

00:49:46:06 00:49:48:05
at the top is the human as

00:49:48:06 00:49:49:06
a human being and his

00:49:50:08 00:49:51:09
physical and financial

00:49:52:13 00:49:54:10
and other assets and

00:49:54:11 00:49:56:10
then going back down to

00:49:56:20 00:49:58:13
take danger or risk that

00:49:58:17 00:49:59:24
technology imposes.

00:50:00:04 00:50:02:07
So I would say it's, again,

00:50:02:08 00:50:03:09
a complex problem.

00:50:03:10 00:50:05:07
All problems in this discussion seem

00:50:05:08 00:50:06:19
to be very complex,

00:50:07:05 00:50:09:09
but it's really impossible to tell.

00:50:09:10 00:50:10:10
But we

00:50:11:05 00:50:12:20
also need to think

00:50:14:13 00:50:16:06
beyond the problem itself.

00:50:16:07 00:50:18:02
So what people often

00:50:18:03 00:50:19:23
not consider is

00:50:21:07 00:50:23:08
what are the bigger risk?

00:50:23:09 00:50:25:09
For example, when we talk about

00:50:25:24 00:50:27:23
large data centres who

00:50:28:06 00:50:30:03
drown their servers and thus they

00:50:30:04 00:50:31:12
save electricity on cooling

00:50:32:08 00:50:33:11
of the data centres,

00:50:34:07 00:50:35:07
do they always think

00:50:36:04 00:50:38:09
about ocean

00:50:38:10 00:50:40:07
and warming the water and how

00:50:40:08 00:50:42:12
it impacts wildlife

00:50:42:13 00:50:43:12
in the long term?

00:50:43:16 00:50:44:17
So this is incredibly

00:50:45:13 00:50:46:13
complex.

00:50:47:11 00:50:49:06
I would say a system of questions

00:50:49:10 00:50:51:01
that we need to ask ourselves.

00:50:51:11 00:50:52:20
And I think it comes to sort of

00:50:52:21 00:50:54:03
this, I guess the undertone

00:50:54:16 00:50:56:04
there is thinking about the
fallibility

00:50:56:05 00:50:58:06
of of humans versus

00:50:58:07 00:50:59:06
the algorithms and

00:51:00:03 00:51:01:22
where potential bias might come in.

00:51:01:23 00:51:03:13
And I think we've got time for one

00:51:03:14 00:51:05:00
more question. And I want to want to

00:51:05:02 00:51:06:02
pose this one to you, Yasmeen,

00:51:06:10 00:51:07:18
because I think it's related to

00:51:07:19 00:51:08:18
that. And this comes from Walid,

00:51:08:23 00:51:10:10
who's in Saudi Arabia, and he's

00:51:10:18 00:51:12:17
talking about essentially the use

00:51:12:18 00:51:14:22
of AI in judicial decisions

00:51:15:02 00:51:17:08
or implementing laws,

00:51:17:14 00:51:19:12
for example, in the context of, you

00:51:19:13 00:51:20:23
know, getting rid of corruption

00:51:21:19 00:51:23:04
where maybe there is corruption.

00:51:23:12 00:51:25:07
Let the presumably

00:51:25:11 00:51:26:25
non corrupt algorithm

00:51:27:07 00:51:28:12
make these decisions.

00:51:29:02 00:51:30:21
How do you feel about that and those

00:51:30:22 00:51:33:02
sorts of approaches

00:51:33:03 00:51:34:16
of replacing the fallible humans

00:51:35:02 00:51:36:23
with the less fallible machine?

00:51:37:20 00:51:39:05
That's a very interesting question.

00:51:39:06 00:51:40:14
And it does link on from the

00:51:40:19 00:51:41:22
previous question.

00:51:42:12 00:51:44:05
In fact, there have been algorithms

00:51:45:01 00:51:46:13
specifically in the US judicial

00:51:47:08 00:51:48:16
system that have been used,

00:51:49:09 00:51:50:13
for example,

00:51:51:10 00:51:52:20
to predict the likelihood of

00:51:53:02 00:51:55:05
reoffending for people

00:51:55:06 00:51:56:17
who have been convicted of a crime.

00:51:57:06 00:51:59:04
And some of those algorithms,

00:51:59:05 00:52:00:21
the predictions have been

00:52:01:03 00:52:02:13
used as part of the judicial

00:52:03:04 00:52:04:24
decision making process of do

00:52:05:00 00:52:06:02
you give somebody bail?

00:52:06:13 00:52:07:14
Are various decisions

00:52:08:20 00:52:10:12
that might be linked to that person

00:52:10:22 00:52:12:18
in their life and circumstances.

00:52:14:04 00:52:15:12
And actually, I can't quote

00:52:16:03 00:52:17:19
the exact paper here, but happy to

00:52:18:01 00:52:19:09
share offline some of those

00:52:19:22 00:52:21:06
algorithms have been shown to be

00:52:21:07 00:52:22:18
biased because again, when you

00:52:23:02 00:52:24:17
look at the judicial system,

00:52:25:11 00:52:27:07
you might end up - as

00:52:27:08 00:52:28:21
you train algorithms, you're

00:52:28:22 00:52:30:15
potentially training them, and

00:52:30:16 00:52:32:04
they're learning biases from the

00:52:32:12 00:52:33:20
world that we live in today.

00:52:34:01 00:52:36:15
And so they are reinforcing

00:52:36:16 00:52:37:23
those biases or potentially

00:52:38:08 00:52:40:05
amplifying those biases that

00:52:40:06 00:52:41:22
they're a person of a certain race

00:52:42:11 00:52:44:20
or colour or a background

00:52:44:21 00:52:46:00
level of education is now

00:52:46:20 00:52:48:13
more likely to reoffend.

00:52:49:03 00:52:51:13
And that gets very dangerous.

00:52:51:14 00:52:53:16
And actually, it's a great example

00:52:53:17 00:52:55:13
of this whole ethical question.

00:52:56:02 00:52:57:18
We want to give every human person

00:52:58:08 00:52:59:13
a fair chance in life.

00:53:00:01 00:53:02:05
But if the algorithm has learnt from

00:53:02:06 00:53:03:09
real world data that if

00:53:04:10 00:53:06:03
you're of a certain background,

00:53:06:11 00:53:08:09
you live in a specific zip

00:53:08:10 00:53:09:21
code and you have a

00:53:10:05 00:53:11:17
certain level of education, you're

00:53:11:18 00:53:12:18
likely to carry a

00:53:13:15 00:53:15:16
commit an offence, that's

00:53:15:17 00:53:17:13
dangerous. We're no longer giving an

00:53:17:14 00:53:19:03
individual a fair chance at life

00:53:19:04 00:53:20:14
where we're stereotyping them

00:53:21:03 00:53:23:14
based on circumstances

00:53:23:15 00:53:25:06
of their upbringing, of their life

00:53:25:07 00:53:26:06
or where they've lived.

00:53:26:13 00:53:28:09
And so this is where it gets very

00:53:28:10 00:53:30:14
grey, as Natalia was mentioning.

00:53:30:19 00:53:32:14
How do we think about

00:53:32:15 00:53:34:21
these applications,

00:53:35:04 00:53:36:18
even though you can do it, should

00:53:36:19 00:53:37:18
you do it?

00:53:37:19 00:53:39:10
And so I think,

00:53:39:21 00:53:41:11
can you do it? Absolutely.

00:53:41:12 00:53:42:22
You can train algorithms.

00:53:42:23 00:53:44:07
Do you want to apply it?

00:53:44:08 00:53:45:22
Do we think it's fair or do we think

00:53:45:23 00:53:48:03
it's ethical is another question.

00:53:48:04 00:53:49:18
And it's important to consider that

00:53:49:19 00:53:51:02
question before just taking

00:53:51:20 00:53:53:09
an algorithm based on its output

00:53:53:21 00:53:55:13
and applying that to society.

00:53:55:21 00:53:57:13
Because, worst case scenario, you

00:53:57:14 00:53:58:20
now create a flywheel or a

00:53:59:08 00:54:01:04
perpetuating situation where

00:54:01:10 00:54:02:20
a certain part of society is

00:54:03:11 00:54:05:02
disadvantaged and the algorithm is

00:54:05:06 00:54:06:14
reinforcing that and people

00:54:07:04 00:54:08:17
can't escape that cycle.

00:54:09:04 00:54:11:01
So I think as a you know,

00:54:11:05 00:54:12:13
just looking at that history of

00:54:12:14 00:54:13:14
where the algorithms

00:54:14:13 00:54:15:23
have been applied in the judicial

00:54:15:24 00:54:17:21
system, we have to be careful

00:54:17:22 00:54:19:13
about how they're applied and what

00:54:19:14 00:54:21:13
consequences that has for people

00:54:22:14 00:54:23:23
I was going to say, it's back to

00:54:23:24 00:54:25:13
that law of unintended consequences.

00:54:25:14 00:54:26:15
But we know that those shouldn't be

00:54:27:02 00:54:28:13
unintended consequences now.

00:54:28:14 00:54:30:15
And I guess really to me that

00:54:30:16 00:54:32:18
this reminds us that, yes,

00:54:32:19 00:54:34:16
humans might be flawed in making

00:54:34:17 00:54:35:20
certain types of decisions and

00:54:35:21 00:54:37:14
maybe, you know, we might put some

00:54:37:15 00:54:39:11
hope in an AI system

00:54:39:12 00:54:40:19
or a set of algorithms, but

00:54:41:11 00:54:42:18
they're going to not be perfect

00:54:42:19 00:54:44:07
either. And really, I think it's

00:54:44:11 00:54:46:22
it's the the bringing together,

00:54:47:06 00:54:48:15
well, what I always like to call

00:54:48:16 00:54:50:09
augmented intelligence of the humans

00:54:50:18 00:54:52:15
and the machines where we

00:54:52:16 00:54:53:22
probably have a better chance of

00:54:53:23 00:54:55:12
actually doing things and reducing

00:54:55:13 00:54:57:09
those those types of errors

00:54:57:10 00:54:58:23
that Yasmeen introduced

00:55:00:04 00:55:01:08
earlier in the programme.

00:55:02:04 00:55:03:16
So unfortunately, we're out of time.

00:55:03:17 00:55:04:23
We could keep talking about this,

00:55:04:24 00:55:07:03
but the clock is

00:55:07:10 00:55:08:21
is unforgiving.

00:55:09:01 00:55:10:11
So I just want to thank Yasmeen,

00:55:10:23 00:55:12:20
Natalia and Felipe for

00:55:12:21 00:55:14:06
spending some time with all of us

00:55:14:07 00:55:15:25
today to talk about these very

00:55:15:26 00:55:16:25
complex issues.

00:55:17:13 00:55:19:05
And I'm sure this is not the first

00:55:19:07 00:55:20:07
time we've talked about these in

00:55:20:08 00:55:21:12
Leadership in Extraordinary Times

00:55:21:19 00:55:23:24
and certainly won't be the last.

00:55:26:19 00:55:28:06
My thanks to Professor Andrew

00:55:28:07 00:55:30:00
Stephen, Dr. Yasmeen Ahmad,

00:55:30:21 00:55:31:21
Dr. Natalia

00:55:32:19 00:55:33:20
Eframova and Dr. Felipe Thomaz.

00:55:35:06 00:55:37:01
My name is Peter Tufano and

00:55:37:02 00:55:38:05
you've been listening to Leadership

00:55:38:06 00:55:39:20
in Extraordinary Times, a podcast

00:55:40:12 00:55:41:23
from Oxford University's Said

00:55:42:00 00:55:43:00
Business School.

00:55:43:18 00:55:45:08
If you've enjoyed this episode, give

00:55:45:09 00:55:47:03
us a rating and a review and

00:55:47:04 00:55:48:19
subscribe to future episodes

00:55:48:23 00:55:50:14
wherever you get your podcasts.

00:55:51:10 00:55:52:25
In the next episode, I'll be

00:55:52:26 00:55:54:12
chatting to Gillian Tett,

00:55:55:02 00:55:56:13
the US editor at large for the

00:55:56:15 00:55:57:22
Financial Times, about how

00:55:58:14 00:56:00:10
anthropology, the other

00:56:00:12 00:56:01:17
AI, can explain business

00:56:02:10 00:56:03:10
and life.

00:56:03:24 00:56:05:13
If you'd like more information about

00:56:05:14 00:56:07:05
this episode and the Leadership in

00:56:07:14 00:56:09:12
Extraordinary Times series, please

00:56:09:13 00:56:10:12
visit OxfordAnswers.org.

00:56:12:06 00:56:14:02
Until next time, thanks

00:56:14:03 00:56:14:15
for listening.

<end subtitles>