Latest / The Retail Podcast / Responsible AI in Retail

Transcript

Responsible AI in Retail · The Retail Podcast

• 0:10
  Hello, and welcome to the Retail Podcast.
• 0:12
  Now today we're going to do a bit of a deep dive in an area
• 0:15
  that has had a lot of press coverage.
• 0:18
  But translating press coverage into the reality for retailers
• 0:21
  sometimes is half the trick. I'm joined by Rob Mckendrick
• 0:25
  from REO Blue. Rob, why don't you tell us a
• 0:28
  little bit about yourself, your background, a little bit about
• 0:32
  what Oreo Blue is, and then we'll come back to the
• 0:34
  conversation around AI. Sure.
• 0:37
  Great. And thanks for having me on.
• 0:39
  Yeah, So I'm Robin Kendrick. I work for Harry Blue.
• 0:42
  We're a consultancy in data and AI and my role is field Chief
• 0:46
  data officer. So I help our customers in terms
• 0:50
  of understanding data strategy, how they're using data and how
• 0:54
  to, you know, do things safely and in a positive way.
• 0:57
  A lot of that comes from a background of working at the
• 0:59
  Co-op for six years. We're at a variety of of data
• 1:02
  roles in ethics, data governance and engineering.
• 1:06
  So yeah, so really excited to be doing this with you.
• 1:09
  Oh fantastic. I know Aria Blue released a
• 1:12
  report on Responsible AI. For those who've not had a
• 1:15
  chance to read it or unaware about, you know, what it means,
• 1:19
  What is responsible AI and why? Why are you guys focusing on it?
• 1:24
  Yeah, thanks. So I mean, put really simply, it
• 1:26
  is what it what it says, it's about using AI in a responsible
• 1:30
  way, which doesn't just mean from an ethical point of view.
• 1:35
  It can also mean from a business point of view, environmental
• 1:38
  point of view. Obviously it's become really
• 1:41
  talked about as you said in the press because of generative AI
• 1:45
  and chat bots. But it also applies to things
• 1:48
  such as machine learning algorithms that you might use
• 1:50
  for optimization or pricing. And I think that, I mean, you've
• 1:55
  only got to look recently right at dynamic pricing, which is a
• 1:58
  great algorithm for Uber, you know, where you've got an
• 2:01
  increase in demand means an increase in price, which means
• 2:04
  an increase in supply because more drivers comes that area.
• 2:07
  But obviously say, let's say concert ticketing, it's not
• 2:11
  doesn't work so well. So it looks a bit like you're
• 2:14
  using, well, it looks like a bit like being opportunistic and
• 2:17
  greedy. Everybody's blaming is on an
• 2:19
  algorithm and saying, well, it was the algorithm that said,
• 2:22
  let's let's double or triple the price of these seats.
• 2:24
  So I think I think it's really important from that point of
• 2:27
  view of understanding how you're using some of these techniques
• 2:31
  and being mindful in the way you do that.
• 2:33
  It's interesting, I mean, because I've seen grocers do
• 2:36
  this with users that are on applications and even at
• 2:40
  conferences, I've heard people say, you know, if you're
• 2:43
  offering everyone 10% discount, why you're offering your loyal
• 2:47
  customers who would spend the money anyway, another 10%,
• 2:52
  right? And so that was like, OK, I
• 2:54
  don't know how it's business. And so there's this sort of line
• 2:59
  between business operations and, I don't know, profit margins and
• 3:05
  then what's the right thing to do or you know, what your
• 3:08
  customers would actually want you to do.
• 3:10
  I think that's the way to look at it.
• 3:11
  What surprised when you looked at the report?
• 3:13
  I, I think firstly, a lot of, a lot of people are waiting for
• 3:16
  the government to say something about AI specifically and
• 3:21
  feeling like until that happens, they don't need to worry about
• 3:24
  it. But you know, CMA have announced
• 3:26
  a, a review of what happened with Ticketmaster.
• 3:30
  If you, if you use a fuse data badly, use personal data badly,
• 3:33
  you can fall, fall foul of the ICO.
• 3:36
  So, so you know you can fall foul of GDPR, right?
• 3:40
  So that that's a really easy thing to do.
• 3:43
  As you say, if you use pricing, you can get into a situation of
• 3:46
  accidentally introducing bias where you might give discounts
• 3:50
  to one particular set of people versus another based upon
• 3:55
  something like gender or race that that might be a byproduct
• 3:59
  of your products or the way you intend to do the pricing.
• 4:02
  But you know, it's all those kind of things that people can
• 4:07
  get in trouble for, can get prosecuted for, but it doesn't
• 4:12
  rely on a new law or regulation coming out.
• 4:15
  It's just part of the existing regulations.
• 4:18
  And I think so the first thing is people are kind of waiting
• 4:21
  for the government to say this is what you can do and before
• 4:25
  they make a change. And I think you really need to
• 4:26
  make that change now, I think. I think secondly, a lot of
• 4:30
  people think about the bad consequences of using AI as
• 4:35
  being something just in terms of brand reputation or say
• 4:39
  regulation. But a badly trained model might
• 4:43
  introduce pricing reductions which are unsustainable for your
• 4:47
  business. It could predict, you know, you
• 4:50
  could have an AI algorithm in in supply chain, which might over
• 4:54
  purchase something because it's got some bad data or the models
• 4:58
  badly trained. So there's potential downside
• 5:02
  financial there. And also, you know, isn't it's
• 5:05
  just not cheap, right? So a large language model, which
• 5:10
  might give you a few percent uplift on a chat bot might cost
• 5:15
  you five times as much as the alternative approach.
• 5:18
  So all of those things aren't just about, you know,
• 5:20
  regulation. They're also about the bottom
• 5:22
  line to, to grocers and retailers.
• 5:25
  And the third thing I think is that people assume that it might
• 5:29
  slow them down. And I, I always use the analogy
• 5:32
  of if you've got a, you know, racing car with a really good
• 5:35
  brakes and steering and go a lot faster because you've got the
• 5:38
  safety built in. So I, I think a lot of people
• 5:40
  worry that if they implement something around being mindful
• 5:44
  and responsible with AI that, that that might slow them down,
• 5:47
  slow their innovation. But I think it really opens up
• 5:50
  the ability to do things and do things purposefully and, and do
• 5:54
  them quickly and well. Who would you normally be
• 5:56
  presenting this to? Who in the business should care
• 5:59
  about this? Yeah.
• 6:00
  So I think that's a really interesting thing.
• 6:03
  So I think there are a lot of people that care a little bit
• 6:05
  about it. If you talk to say head of
• 6:07
  compliance, they're they've got to worry about AI.
• 6:11
  Head of data protection would be worried about it.
• 6:13
  Data scientists, they're kind of worried about it.
• 6:16
  So you've got quite a lot of people who have a concern but
• 6:19
  don't feel they're empowered to do something about it or don't
• 6:21
  feel it's necessarily their scope.
• 6:23
  So when we talk to people about implementing something around
• 6:27
  responsible AI, we say what you've got to do is you're going
• 6:31
  to take this top down. You've got to talk to the
• 6:33
  executive when you've got to set some AI principles that really
• 6:36
  set the tone of of how you're going to use AI for an
• 6:40
  organisation. So, you know, go into the
• 6:43
  C-Suite, talk about, talk about some of these opportunities,
• 6:47
  potential downsides and, you know, make sure it's up to them
• 6:53
  to show the responsibility and accountability within their
• 6:55
  organisation. So put in place some principles,
• 6:59
  put in place some cheques and measures and and then maybe
• 7:03
  enable those heads of compliance, heads of data
• 7:06
  science to do the right thing in their organisation.
• 7:09
  But you know, it really needs to come down from the top.
• 7:12
  Have you seen that in action like in do retailers do that or
• 7:15
  grocers? Are they effectively doing that?
• 7:18
  Are they trying to to do? That the the research we
• 7:21
  conducted said there were certainly very few are talking
• 7:24
  about having done that, right. Yeah.
• 7:26
  So it, it was literally sort of, you know, one or two out of 40
• 7:30
  retailers that we looked at in the UK have said something about
• 7:34
  that as a Co-op where I used to work.
• 7:38
  We, we worked quite a lot on that, but it was part of our
• 7:41
  whole environmental social purpose of the Co-op to do
• 7:45
  things responsibly. So we consider it partly a good
• 7:48
  thing to do. And the, the retailers we're
• 7:51
  working with at the moment, it's putting this together as a bit
• 7:56
  of a new thing for them. So it's something that they
• 8:00
  might have been doing, but it's something that they kind of do
• 8:03
  it in pocket. So they, they haven't really got
• 8:05
  their hands around it. Last year, Global Data Science
• 8:09
  from Unilever talked about the fact they've been looking at at
• 8:11
  this worldwide and they they'd assessed 40 of their AI
• 8:16
  algorithms for being responsible huge amount.
• 8:19
  They had 400 to go. That AI was so prevalent in
• 8:22
  their organisation that, you know, it was literally the tip
• 8:24
  of the iceberg that they'd looked at.
• 8:26
  But if I'm not a top three retailer in the UK, right, I
• 8:31
  probably won't have that depth of field or that I'm relying on
• 8:36
  other people to help me. So where do I start?
• 8:39
  Maybe is the better question. Yeah, I, I think so.
• 8:43
  So when you start from, from that, you know, let's assume
• 8:47
  you've done the C-Suite conversation and you said, OK,
• 8:49
  we've got, we've got in place our, we're going to place our
• 8:53
  principles. We want to do something about
• 8:54
  it. I think creating a creating your
• 9:00
  understanding, your list of all the applications you might
• 9:04
  already have or systems, you know, because you might be, you
• 9:07
  might have, you know, be using Microsoft and you go, oh,
• 9:10
  switched on Co pilot. That's cool.
• 9:13
  You know, you've got AI there. Most of the ERP systems and
• 9:18
  finance systems are claiming they've got some kind of AI
• 9:20
  capability. So you've got something there
• 9:22
  that's, you know, already in place.
• 9:23
  So understanding that to begin with and saying, what are we
• 9:26
  using? And then looking at them and
• 9:31
  from a risk perspective and say, what are the potential things
• 9:34
  that could go wrong with some of these things?
• 9:37
  What would we, what do we need to put in place to investigate
• 9:41
  them more fully? And then you can start to work
• 9:44
  back if you've got, if you've got a system that's, you know,
• 9:47
  where we've built a, we've built a brand new system to give
• 9:52
  offers to customers and it's based on AI.
• 9:56
  Do you know if the data going into that model is correct?
• 10:00
  And what kind of ways can the data be incorrect that would
• 10:03
  that would cause that model to go badly wrong?
• 10:06
  And then you can sort of kind of trace that back.
• 10:08
  And and as you say, not everybody has the money of a
• 10:13
  Unilever. So what things can you do?
• 10:15
  You can do things like use the algorithm, but make sure you've
• 10:18
  got some testing around it as the algorithm rolls out.
• 10:21
  You know, we used AI solutions to work out that the Co-op what,
• 10:26
  what bread to bake and what cookies to break bake.
• 10:28
  But there was a Baker who looked at that and said, well, that's,
• 10:31
  that's clearly on because I'm not going to do exactly that,
• 10:34
  but I'm, I'm mindful. So you call that human in the
• 10:37
  loop in, you know, in the same way that, you know, in, in
• 10:41
  banking, you would hope that your application for a credit
• 10:46
  card might have somebody looking at it at some point.
• 10:48
  You know, if it's on the margin, you might go, OK, is there
• 10:52
  somebody looking at that? And in the same way you you want
• 10:54
  to see if there are any high risk algorithms, whether you've
• 10:57
  got a human in the loop who can, you can look at those things.
• 11:00
  Is that what term? Is that a Rob term?
• 11:02
  I love that term. That's a.
• 11:03
  That's a. That's AAI term.
• 11:07
  In the, in the, in the, in the know they know, right?
• 11:10
  Yeah, yeah, yeah, exactly. So it's like, yeah, but it but
• 11:13
  it's a good one because it's like unlike a lot of the other
• 11:16
  techno jargon that that actually tells you what it is right there
• 11:19
  in that making such a. Futuristic term.
• 11:22
  I can imagine that term. It's the first time I'm hearing
• 11:25
  it, so I can imagine that term, you know, was there a human in
• 11:27
  the loop? I was like, Oh yeah, right.
• 11:29
  We, we didn't have a human in that loop because it was just
• 11:31
  all process, which I guess then leads into I, I think what
• 11:35
  you're talking about and what the report was, is about your AI
• 11:37
  framework. So what, what are the pillars to
• 11:40
  the framework? Was that what you were just
• 11:42
  covering there? So yeah, that that was that was
• 11:44
  around the whole AI safety pillar.
• 11:48
  So, So what we, what we've done is we've built it up from the
• 11:51
  core of data governance and architecture.
• 11:54
  So do you understand your data? Do you understand how it got
• 11:59
  into the system? Do you understand how it was
• 12:02
  collected? So that's the first, first item.
• 12:04
  You think of it like expanding out, right?
• 12:06
  So that's, that's the first, that's the first part.
• 12:08
  The second part is looking at safety, which is where the human
• 12:11
  loop comes in and understanding how the system is built, what
• 12:16
  tools you're using and having those cheques and balances.
• 12:19
  You've got AI safety as the next thing.
• 12:22
  Building out from that, you can start to look at biosynthetics.
• 12:25
  So, you know, just because something is doable, should we
• 12:29
  do it which can which, you know, as I mentioned at the beginning,
• 12:33
  that could be something around it's going to cost quite a lot
• 12:37
  to build this and maintain it. Is is that cost worth the worth
• 12:41
  the benefit at the end of the day, just because it's like
• 12:43
  cool, funky to have that AI, should we do it?
• 12:46
  But also from a, you know, customer experience, as you
• 12:50
  said, if everybody gets, if everybody starts getting dynamic
• 12:53
  pricing and you know, some of the customers feel like they're
• 12:56
  being, you know, being miss sold or something that that's,
• 13:02
  that's, that's a bad thing. You can do that, but you know,
• 13:04
  it might be legal to do it as well.
• 13:06
  It's just that case of is it good from a customer safety and
• 13:08
  ethical point of view? And then then the last pillar
• 13:12
  around that, the top thing is, is there somebody that's at the
• 13:16
  top of the organisation from a governance point of view saying,
• 13:20
  do we have a strategy? Do we have overall control of
• 13:24
  this? Have we enabled people with the
• 13:26
  right roles and responsibilities to look at this?
• 13:30
  And it's those four areas, the data, the safety, the ethics and
• 13:34
  the governance and the components we've put within
• 13:38
  those that really says OK, and now I've got everything I need
• 13:41
  to be confident I'm being responsible with AI.
• 13:44
  Rob, thank you so much for taking us through.
• 13:47
  You know what is responsible AI, some of the outcomes good and
• 13:52
  bad, where in the business we need to to look at it.
• 13:55
  I guess I'm curious in terms of, you know, where can people get
• 13:59
  hold of the report and what next, what should they be doing?
• 14:03
  Reports on our website, arioblue.com names on that
• 14:10
  report contains and some of the other things on the website
• 14:12
  contain the framework. So people can look at that and
• 14:16
  understand a little bit more about the framework and, and
• 14:19
  they can make a, an assessment for themselves on how, how well
• 14:23
  they've considered some of those things in the framework around
• 14:26
  around data around safety, ethics and governance.
• 14:29
  We also have some tools that help us so we can quite quickly
• 14:32
  help people with looking at, at that maturity and, and then help
• 14:35
  them plan out what to do next. So start with the report.
• 14:37
  There's some really great, great ideas in there about how to, how
• 14:40
  to get going. And as I say, the the framework
• 14:42
  which people can use and and have a think about themselves.
• 14:45
  I get I get this is just a personal curiosity.
• 14:48
  How long do these projects take? How long would you be?
• 14:50
  Like is this an expensive thing to be doing?
• 14:53
  I I think that that there are two things.
• 14:56
  One is the way we run that kind of maturity settlement, it's
• 14:59
  really quick. It's a few days of interviews in
• 15:02
  somebody's office or you know, virtually more and more so, so
• 15:06
  that that's not a long process. I think that if you are an
• 15:11
  organisation who has started to think about these things and
• 15:15
  have made a mistake, then the regulator's going to be a lot
• 15:18
  more lenient on you than if you haven't started at all.
• 15:22
  So the process of getting going is really important and then you
• 15:26
  can take it at your own speed. As you know time funding allows.
• 15:33
  You'll probably find that a lot of a lot of places in your
• 15:35
  organisation have started thinking about these things and
• 15:38
  maybe just needs writing down the things they're already
• 15:40
  doing. So you might not be in a bad
• 15:42
  situation as you as you sort of to begin with.
• 15:45
  So having a look at that, understanding the framework and
• 15:48
  then applying it and thinking about what gaps you might have
• 15:52
  would be, you know, really first, great first step and
• 15:55
  something that will help you on the journey and help you keep
• 15:58
  you on the right side of the regulators as well.
• 16:01
  That's brilliant and definitely regulation is coming.
• 16:04
  And so I guess it's better to be on the right side of it before
• 16:07
  it's even there exactly because it's a lot harder to unpick some
• 16:11
  of these spaghetti systems that you see.
• 16:14
  Rob, thank you so much for taking the time to to talk to us
• 16:18
  and we'll speak to you soon. Brilliant.
• 16:20
  Thanks a lot. Take care.