McCargow director of AI and PWC I met up with rod at PWC office in London and we

chatted about modeling unintended consequences, AI ethics audits, working

with dubious companies and intentions, what we should be teaching our children

and future careers a recipe for AI future mitigating job displacement and

other AI for good topics. If you like this podcast then check out the other

episodes at machine-ethics.net or you can contact us at hello@machine-ethics.net

you'll find us at Twitter, Instagram and YouTube, to support the

podcast go to patreon.com/machineethics extended interviews reviews and

more my thoughts on the episodes and AI topics and news of the month, thanks

again to Rob and hope you enjoy

Hi Rob thanks for joining me on the podcast

thank you having week could you introduce yourself and what you do

absolutely so I'm the director of artificial intelligence at PwC in the UK

so our team is basically and tasked with applying the technology across the

breadth of our organization on both internal projects but also working with

that clients across all industry sectors on solving some of their hardest

business problems as well using different forms of AI and a lot of the

other things I'm involved with also involve working with governments around

the world on the impacts on national strategy and policy and as part of our

assets on the advisory board of the All Party Parliamentary Group on on AI

amongst other appointments yes so it's been said that you're the nicest man in

AI how'd you feel about that depends who said it well it's just first I've heard

so on that point Robin to you kind of what is AI when you're talking about AI

what are you talking about more specifically well I think it depends on

the the audience that we're dealing with at the time and if we're working with

clients across different corporate functions across HR for example and

compliance maybe we don't necessarily get into a deep deepest of technological

descriptions but for me I think it's a high level to differentiate from

technology of old technology that falls into the AI domain of technologies that

can sense think act and through an iterative feedback loop learn and they

clearly sit as interesting bedfellows along more mature technologies such as

robotic process automation for example but we tried to focus across the breadth

of the main AI technologies but if I'm being candid the very first thing I do

in any of these things is state that I keep the job title to get me into the

room but the first thing I do is say they are doesn't really exist we have

this assembly of really interesting technologies on the pinnate from machine

learning and deep learning to natural language processing and generation and

other techniques that make up this AI family yeah so so the AI of

kind of science fiction doesn't exist necessarily but you've got this kind of

suite of things which go under that banner at the moment indeed yeah

great and we were talking briefly before but kind of how does PwC fit in with

this how they're talking about a I and and what they're doing in anyway I guess

as well well I think where right now is we've seen amazing breakthroughs of the

technology in in consumer use cases in use cases of fascinating utility but not

necessarily a huge amount of consequence on people's lives

so there's fantastic things being served up through iron maps or movie

recommendation engines and all sorts of ecommerce types of applications I think

where we're starting to see businesses in for example heavily regulated

industries healthcare financial services banking insurance et cetera criminal

justice starting to wrestle with this technology realizing that this has a

profound impact on their business they have to get moving on starting to

embrace and adopt but by doing so this opens up this whole cupboard of new

risks which I'm sure we'll get into over the course of the conversation today so

for us I think because we're already working with just about every

organization across the across the land it's some capacity are the auditing or

advising them in some capacity we're often on-site there is the trusted

advisor to help debunk some of the myths ology provide the right level of comfort

and confidence around the tech and allow them to get started and start moving a

pace with the innovation offered by AI so I see you a lot at

these sorts of conversations that you mentioned the kind of what happens when

you have these sort of technologies in those places in healthcare in the

justice system oh there's something like top-level I mean obviously this machine

at these forecasts and we took up a lot about this sort of things is there some

of the things which you're keen on like things that you are interested in

talking about in terms of those sorts of ethical issues yeah I mean I think be

more led by the you know the explosion in these events that I get the privilege

to go and speak at and and I think judging by the Q&A after them the two

areas that seem to elicit by some comms severable distance the most interest and

the most inquiry first of all I think is around the impact on the workforce

through automation through human machine interaction and through education skills

and future proofing of careers yeah I think that's one big category that

always creates huge amount of interest and then anything else that falls into

that AI ethics bucket is again of significant interest and and that's for

me is is a fascinating area and as we start seeing this started to scale in in

these use cases of significant consequence this brings a whole level of

interest across the breadth of different corporate functions to make sure that

people are fully conversant with the implications on their business yes II

think it's really important that those business leaders are appreciate the

technology and they might not have a low level understanding but if they're going

to apply it then they better well know what they're applying yeah this this is

now an absolute necessity rather than a nice-to-have this is a fundamental

prerequisite for a four up for a executive C suite member of a board for

example because this these specific use cases will more often not rear their

head in their departments and I think the one that I found very interesting to

look at and make sure that we're clear focused on are some of the the HR

applications you and I monitor this the press and the media quite a bit around

keeping up to date and what's happening and the ones that seem to constantly

rear their heads like social media where we first met I think are are those sort

of ones around recruitment for human performance type of monitoring systems

and as a consequence you know people like HR directors absolutely have to get

to grips with this technology and quickly yeah

or and there's some stories there of how that's been negative or like done not

necessarily really badly but like in a you know kind of good and evil sort of

way but like dubiously something that maybe we don't want to promote

thing and there's the Amazon example comes to mind about having you know

promoting men in their CV machine learning tactics and things of that and

because of past bias data than this little thing so it's really about

getting around that sort of Missy misunderstanding

maybe not the malicious use but like a stupidity in these high stakes arenas

but yeah I'm a big believer that the substantial majority of people were

configuring and deploying these systems are coming in with the best of

intentions yes with with good values more often not but and we we I think we

see where they don't always work out well they don't always leads to the best

outcomes they often can lead to amplification of bias and discrimination

for example and typically affecting vulnerable groups of they called us or

all customers it is often because that there's not been the right mixture of

people in the room to provide the right level of challenge and I think on the

one hand we very well aware that there's a substantial issue around the

homogeneity of the workforce you know it is very well noted that it's extremely

white a male which there's a lot of great corporate ishutin cluding some of

ours to try to address that imbalance but I think I'm also looking at the

Disciplinary homogeneity as well and it's absolutely critical to out there

are people in the room to give that level of challenge and and that go/no-go

power of veto and for example when wherever configuring a specific use case

in our team will always make sure we have the right subject matter experts in

the room and and even beyond that in fact I've been talking about ethics and

AI in business for quite a number of years now and I thought about it

actually take some action on this and that we've just hired our first AI

ethicist to the team months ago cool who we we know with the level of rigor we

face as a organization and scrutiny we know that we're confident that we meet

the high standards around data security and privacy around regulatory compliance

around around you know the whole issues of risk and quality

but specifically with regards to ethics that need to now think through not just

secondary but tertiary unintended consequences it's now critical that

giving people that power and freedom to explore investigates and model and

challenge I think it is something really valuable now and and that's you know

giving us that interesting new type of job they'll be talking about the jobs of

the future what we've just created you know anyone for our organization so

prove it's gonna happen yeah that's great they what's that kind

of remit is it kind of like future rising or is it more like philosophy or

people yeah this twenty word Lander in in the real life world of what's

expected yeah which is actually that's what I mean yeah London in the kind of

reality of what's happening on the ground I mean really we have the

opportunity to meet some of these for a meet the world-class academics and

philosophers and ethicists working on this and you know it's fascinating I've

learned a lot in recent years but if you sitting there in business making random

decisions to drive profit or to reduce cost or future prove the organization

there's maybe not the same man you ship of deliberation that happens and if you

think about ethics specifically with of course in this explosion of publication

of new ethical principles in the last two years in particular I think at the

last count we'd we'd come across the in excess of 70 if you add together the big

tech companies the World Economic Forum the I Triple E the baking principles it

having all these together yeah you've got a lot of material out there and we

have reality of businesses going to be able to read all of those and discern

which one is most appropriate for their particular geography and setting yeah

and and acts accordingly and so what we what we have is we actually have read

all these whole team on it and went in with a fine-tooth comb and built

effectively a traceability matrix so what we can now be able to say to

clients through what we call our responsible AI approved

is okay we feel that with for the right governance in place around the project

its had the right approach in terms of identification and the biasing of

datacenters prior to training we're confident that it's appropriately

scrutinized from a security and privacy perspective and for this particular use

case it's got the appropriate level of interpret ability and explain ability

now moving beyond that we can say it's got this relatively clean bill of health

with the caveat to give you the confidence to move forward now there's a

conscious decision to make as a leadership team running these projects

to say what do we optimize this solution for is it to maximize profit performance

is there a trade-off to be made that allows you to drive even

further transparency into the system and you want to then optimize for fairness

and the fairness today is fascinating I think even more fascinating than ethics

there was a piece of work enough maybe you still could share the the link so

you can share with you regionally we did a piece on this and we found that

fairness is something that's constantly raised in all of these ethical principle

documents yeah and the very high level ones are all very laudable and they stay

on ethics so AI should be benevolent it should be good for Humanity it should be

transparent it should be fair and inequitable eccentrics yeah but are you

get to fairness there's in excess of 20 mathematical definitions of fairness so

if you then just take that at all who's it fair to yeah you can't be fair

universally to every single person in society

yeah therefore do you have to define who it's fair to yes so when you get into

those sort of conversations you can really make sure that the projects are

proceeding with that level of rigor of conversation certainty buy-in and a

conscious choice around what the project is is optimized to do yeah and there are

sort of things that the our own ethicists will

the ability to shape and starless conversations for our own projects and

clients that we work with the fairness thing is really interesting I think that

comes that's part of the ethical conversation because what you know like

the morality of an individual like what is fairness to you does that actually

warrior she told me about when we say fairness is a really interesting point

if if a a company came to you and went oh this is so great but we actually

trying to optimize for you know the the outcome the the monetary return and

actually maybe the fairness is not on a high or an agenda

well the sorts of conversations do you have there yes quite difficult I mean I

think I don't think I don't think we've had them like that yet but I think it's

an interesting question to raise alphabetically as as the market becomes

more sophisticated I think there's two things there I think that there there

may well get to the point where there's certain use cases and applications that

it's simply not appropriate to go near the certain industries that might be

more difficult than others to work with but I think we also have to respect the

fact that to attract the very best talent the best talent want to be

applying their skills to the you know the the appropriate use cases and with

that in mind giving people the the right to to not have to time partake in

certain projects is something I think that's getting slightly comfortable with

yes oh and we've seen this haven't we in that last a year with how breaks of

employee activism which i think is something which your organization's need

to take into account around so what we want to be aligned with doing yeah and

it's I mean for me it's look it's laudable that people who are taking

these sorts of actions in the face of things which go against their principles

they're you know internal I spoke to one of those people not so long ago actually

who Jack Paulson I don't know yeah of young he quit Google because of the the

arm stuff and one was the one of the first people to do so and then there was