Latest / The Road to Accountable AI / Nuala O'Connor: Frontline Human-Machine Interface at Walmart

Transcript

Nuala O'Connor: Frontline Human-Machine Interface at Walmart · The Road to Accountable AI

0:00
You're welcome.
0:01
It's so great to talk to you.
0:03
It's so great to be here, Kevin.
0:04
Thanks for having me.
0:06
You've been involved for a long time at a number of different organizations, public
0:10
sector, private sector, nonprofit, on issues around internet policy, privacy,
0:16
and so forth.
0:18
How is it different or is it different in your perspective what's happening now with
0:24
AI?
0:25
I think that I just heard a ding, so I'm going to have to start over again to
0:33
silence my notifications.
0:36
Okay.
0:38
You know, Kevin, I think there are a lot of conversations right now about whether
0:41
this is truly different and kind of the really monumental change in how we live
0:46
our lives.
0:47
I'm a little bit of a skeptic about that in that I guess perhaps because you and I
0:51
have been working in this area for so long, we've seen
0:54
the trains of large language models and automated decision making and just
1:01
software that runs really fast in the background.
1:04
Why I think this feels really different to people is that, first of all, this has
1:09
come into the wild.
1:10
I mean, it's come into the public use very quickly and that it seems to be alive.
1:17
People seem to feel like this is making decisions about them or it's talking back
1:21
to them.
1:22
And although we know cognitively it's not alive, having a converse with you, having
1:29
an inanimate object suddenly talk your language feels different than previous
1:34
iterations of technology.
1:37
Why I think it's important that we scrutinize it is, and I've said this
1:41
before, it's not only the data that we're putting into it, it's the decisions it's
1:45
making about us in the world.
1:48
What about the development of AI governance and policy?
1:52
How have you seen the regulatory developments in this area around AI
1:55
relative to what we saw a decade ago or two decades ago with the internet and
1:59
privacy?
2:00
I think because it's moved so quickly from not known to known in the public, the
2:06
policymaking world is trying to catch up as usual to technology and doing it
2:10
quickly.
2:11
What I've also seen is that good companies, I think, are ready to apply
2:17
governance principles to new technologies because this is not the first one.
2:20
We have internet, we have...
2:22
social media, we have lots of policies around other uses of technologies by our
2:27
own, what we call employees, our associates.
2:30
And so we really, we in particular, my company, we're ready.
2:36
We had already thought about and started to build and started to benchmark with
2:41
other companies.
2:42
What does good governance look like?
2:44
for these tools.
2:46
And of course, you know, just say AI, it's almost like saying sidewalks, right?
2:49
It encompasses so much, right?
2:52
But to be general, kind of looking at emerging technologies, biometrics, AI, and
2:57
a whole bunch of other de novo uses of technology in the real world, proximate to
3:03
human beings, we did set about looking at what the tech companies were offering,
3:08
looking at what other companies were recommending, looking at what trade
3:11
associations and think tanks were saying about,
3:15
governance and the possibilities of governance.
3:17
I also sometimes like to call it trying to govern the ungovernable, right?
3:20
Because it's so opaque in so many ways, but it's also, we shouldn't engage in that
3:26
kind of tech exceptionalism that we are so special and so new and so different that
3:30
we shouldn't play by some reasonable rules or that it's just too hard to figure out.
3:36
If it's too hard to figure out, we shouldn't be using it, right?
3:38
So what we did was look at ordinary processes that companies put into place.
3:44
what the company's principles are, and then how we apply those as people.
3:50
And it's usually our engineering, our data scientists, our tech folks who want to
3:55
deploy new tools that will affect human beings.
4:00
We pull those out of the work streams and say, if this tool is going to be making
4:04
consequential decisions about a human being, their rights, their knowledge,
4:09
their abilities to move forward, in most particular,
4:12
their employment and advancement, that tool is going to require additional close
4:19
scrutiny by technologists and by lawyers and policy people before it goes live.
4:24
And then at waypoints, you know, six months a year, and then perhaps annually
4:29
after that, because as you know, these tools drift, they change, that, you know,
4:33
it is a constant governance model.
4:35
And we're still building towards automation in our governance and
4:39
monitoring of tools of all kinds.
4:42
And, but again, I don't think just because it's new and funky and cool, it should
4:46
escape some kind of scrutiny.
4:48
In fact, maybe, maybe more so because of that.
4:52
So you issued a Walmart Responsible AI pledge in October of 2023.
4:57
And first of all, I'm curious, it's interesting that you call it a pledge.
5:00
It's not just a set of principles.
5:03
Tell me a little bit about the process of developing that.
5:07
And like all of our kind of internal policies, statements, pledges, it's really
5:12
a multifunctional endeavor.
5:15
I don't think the lawyers should sit in a room and write things.
5:18
I don't think the technologists should sit in a room and write things.
5:21
The comms people shouldn't be working by themselves.
5:23
It was really a kind of collaborative approach that a number of our teams came
5:27
to my team and said, we'd like to say something publicly about the guardrails
5:32
and the guidelines and really the values that we live by as a company.
5:38
And I said, really, who's going to listen to that?
5:40
Again, somewhat the skeptic.
5:42
But in all seriousness, the pledge is really to our customers and associates,
5:45
people who we are engaging with to both be transparent and accountable in our uses.
5:52
And that sort of really reflects one of the things I say to my team all the time.
5:57
When we are introducing new technologies into the environment of human beings,
6:03
whether it's online or in the stores, in the real world,
6:06
in offices and in warehouses anywhere our associates work, we have a duty to explain
6:13
because in many cases it may be the first time someone's come into contact with
6:18
this.
6:18
And so I always say to our team, inform, educate, entertain.
6:23
And let me tell you what I mean by that.
6:24
Inform, obviously notice that, you know, that something is happening in the
6:28
environment that is different.
6:29
It's particularly in the stores, which were, you know, 20 years ago, you could
6:32
walk in, you could walk out, you could pay cash, you would be unobserved to any
6:35
system in the world.
6:36
That's changing, obviously, with your cell phones, with electric cash registers, with
6:41
cameras in the stores for security and safety.
6:44
So I think we have a duty to inform.
6:46
We are meeting people where they are in their comfort level with technology, in
6:50
their educational level, in their language abilities.
6:53
And we have to speak to all of those people.
6:56
So that's inform.
6:57
Educate.
6:58
Tell them a little more.
7:00
Tell them a little more about what the technology does, how it works.
7:04
When I grew up in New York, there was a company called Sims.
7:07
I actually had a summer job there one summer and their phrase was, an educated
7:11
consumer is our best customer.
7:13
I believe the same thing about technology and the technology and responsible
7:17
technology use of our company.
7:18
If they know the responsible choices we're making about these technologies, it will
7:23
engender trust and mutual respect and hopefully really benefit the ongoing
7:28
customer store relationship.
7:30
And the last part is entertain.
7:32
And honestly,
7:33
This is fun.
7:34
You and I were talking a little while ago, you know, I, in all of the years we've
7:38
been working, we, I get to learn something new almost every day.
7:41
And, you know, certainly every year there's a new, you know, or every few
7:44
years there's some monumental change in the technology.
7:47
That's a joy.
7:48
It's really a joy to be a part of that and to learn and to be surrounded by people
7:54
who are just so, so smart and dedicated and both want to do the right thing and
7:58
also want to use the latest technology to do whatever it is their job.
8:02
requires.
8:04
So and it should be fun.
8:05
We had a long time ago, we had these sweepers in some of the stores.
8:09
And if you go into Sam's Club, you might still see them.
8:12
And they forget the brand name, but they were about the height of a, you know, I
8:18
don't know, 10 or 12 year old boy.
8:19
I'm thinking about my son is probably, you know, a few years ago.
8:22
And but almost almost eye level.
8:24
And they had cameras.
8:25
And the purpose of the camera was actually to scan the shelves to say what's out of
8:30
stock so that
8:31
people in the back store room would go, okay, on aisle seven, we need to get more
8:36
cereal.
8:38
But a reasonable person would think, was that camera looking at me?
8:42
What's that camera?
8:43
What's it recording?
8:44
What's it gonna do with that information?
8:46
So I said to the guys, first of all, I thought we should have put vests on them
8:49
and made them, anthropomorphize them and also have disclosures right there on the
8:54
machine that says, I'm not looking at you.
8:56
In fact, we had blurred out any facial.
8:58
features of any human being to just be looking at boxes and QR codes and the
9:05
like.
9:06
But it can be fun.
9:07
I have no problem with anthropomorphizing the technology if it helps people
9:11
understand and feel comfortable with it.
9:16
me just stop a second.
9:17
Do you want to see if you can shut off that other thing that's dinging?
9:22
If it's too much of a pain, it's not a big deal.
9:24
This is not Hollywood, but yeah.
9:26
I don't know what that is.
9:28
okay.
9:30
how do I shut off the notifications?
9:32
Let's take a look on my
9:37
Okay, I think it's gone.
9:39
I think it's all off now.
9:42
Alright, thank you.
9:43
Yeah, it was not a big deal, but...
9:44
No, no, we don't want it in the background.
9:47
Okay.
9:47
Back at it.
9:48
Okay.
9:50
Hope that works.
9:51
No, OK, we'll see.
9:52
All right.
9:58
That's a good.
10:07
first.
10:09
no, no, you're okay.
10:10
Yeah, it's, I mean, whatever this, this tool Riverside is recording each of us
10:13
individually and uploading it.
10:14
So it should be fine.
10:15
Even if the.
10:18
Okay.
10:22
Yeah, it's interesting that one of the values that is in that pledge is customer
10:26
centricity, which is not typically what people see in terms of AI.
10:30
You've got things like fairness and so forth.
10:33
But why is customer centricity for Walmart a value in terms of responsible AI?
10:39
I think so many times companies roll out technologies that they say are for the
10:44
customer, but are really for the company or really have some other purpose.
10:48
And in this case, when we say customer centricity, it really goes to what I was
10:52
saying before.
10:53
Does the customer understand?
10:55
Do they feel comfortable with it?
10:57
Is it a use case that enhances their experience?
11:00
That's, I think, some of the values I think about when I think about customer
11:03
centricity.
11:04
But ultimately, it's about does this deployment...
11:08
serve the humans with which it is proximate.
11:13
And if it doesn't, why are we doing it?
11:15
And if those customers or associates don't understand it, then we have failed in our
11:20
duty of care, our duty to inform, educate, and explain.
11:25
So I think it really goes to, we have recognized that people will be impacted
11:30
even just having it in their environment, having it nearby, and that...
11:35
There may be different levels of comfort with it and that we have a responsibility
11:40
to those, to all of our customers, regardless of their comfort.
11:44
What else do you think is distinctive about the way you approach these AI
11:50
issues, given that Walmart is a retailer that touches literally hundreds of
11:54
millions of customers directly as opposed to enterprise service providers, for
12:00
example?
12:00
Well, again, it's because it's a frontline human machine interface, right?
12:05
We get to live in the real world and physical world as well as the internet and
12:09
see how people feel comfortable with these tools.
12:13
It's funny that we were focused on customers because I would say our most
12:16
effective rollouts initially were in tools that support our associates in what
12:23
they're doing.
12:23
You know, I give the example and I don't think it's a real one, maybe a real one
12:28
now.
12:29
when I was doing a Sam's Club tour and they let me decorate cupcakes, which I
12:33
think is hilarious.
12:34
Now, anyone who's had a young child knows when you're ordering a birthday cake, you
12:38
want to make sure you get the red one or the blue one or the right number of roses
12:42
or the right number of decorations.
12:44
And so there are guidebooks that tell associates, this is how many roses you put
12:49
on an eight by 11 sheet cake and this is what it should look like for a new
12:53
associate or even for someone experienced who hasn't done this particular task
12:57
before.
12:58
That means they run to the back room, they have to flip through the thing, they have
13:01
to page through and say, you know, what is, and then read and then decorate and
13:04
then read.
13:05
Just one example of all the many thousands of things associates do in a day.
13:09
With AI, there is a potential and I believe it's deployed already in some
13:14
places in the country to simply say, and we have a tool called Ask Sam, we have
13:20
another tool, we have tools named a number of things that are kind of in the Walmart
13:23
legend and language.
13:25
And say, you know, Sam, how do I decorate an eight and a half by 11 cheesecake for a
13:30
young, you know, for a boy, for a girl, for, you know, in a fire truck, whatever.
13:34
And that will save time, saves effort.
13:38
They can say, you know, repeat that.
13:40
It will really be an efficiency builder for an associate.
13:43
And again, that's a very, I think, fun and upbeat example, but there are lots of, you
13:48
know, paper based or knowledge worker examples as well that improve efficiency.
13:53
One of the things we have put online are a number of our employee manuals, our
14:00
benefits manuals, basic information.
14:02
When you've got millions of associates, many of them are going to be asking the
14:06
same question, right?
14:06
So saves time, makes sure the information is accurate, every associate is getting
14:10
the same information.
14:12
That's a good example of where regular old AI is a good enough tool.
14:17
do not need generative AI because you actually don't want it learning.
14:20
You don't want it learning and changing the content.
14:21
You want that content to be static and accurate and correct for everybody.
14:25
That's, I think, one of the other points I would make is people have gotten really
14:28
excited about generative AI.
14:30
And it's a fascinating tool.
14:31
And there's lots of good and bad about it.
14:33
But you don't necessarily, it's not necessarily the right tool for every job
14:37
that you might think of.
14:38
And so fit for purpose, to me, is a really big question.
14:43
you know, and accuracy and truth of the information to me, those are some of the
14:48
biggest worries I have about gender AI.
14:52
How do you structure the work on responsible AI in your organization or
14:58
your organization relative to the rest of the company?
15:00
Thanks for asking.
15:01
So digital citizenship is, as we define it, the responsible use of technology and
15:07
data, period, full stop.
15:09
That's the mission.
15:10
We have now got the privilege of leading a team of several hundred people.
15:14
And that includes lawyers, technologists, data scientists, policy people.
15:19
It's basically half kind of legal and half compliance and operational.
15:23
And so we went about setting up a team.
15:25
I really think that the woman who has to lead that team, when we created it,
15:30
had one of the best jobs in the company because her first year was doing all the
15:35
research and the benchmarking and finding all the best practices and then standing
15:38
up the team for us.
15:41
And again, it is the people who created the original policy went right up to the
15:45
leadership of the company to say, these are the rules for rolling out an AI tool
15:50
that has an impact on a human being.
15:53
And then created a kind of a gate in our rollout processes to
15:58
to try to explain it without too much jargon, that says, is this an AI tool?
16:04
Does it impact humans?
16:05
If so, you've got to be reviewed and answer some questions and engage with the
16:11
Digital Values Team.
16:13
That's the name of the team.
16:14
And again, that team includes, it is led right now by a data scientist, and they
16:19
will kind of probe and scrutinize the algorithm itself, the purpose for the
16:24
tool, who's rolling out, who is the responsible kind of business owner of it.
16:28
and then schedule, not just do the initial scrutiny, but schedule kind of follow up
16:33
monitoring and oversight.
16:35
So kind of all the tools of a traditional compliance program, but applied to this
16:40
new technology.
16:41
And so far, so good.
16:43
And we've worked with quite a few teams.
16:45
I don't have an exact number of how many tools they've scrutinized, but the team is
16:49
busy and it's growing.
16:50
So I think the metric is, you know, as we do more in this area, they have more work
16:55
to do.
16:56
But the biggest, I think the biggest challenge and the biggest opportunity is
16:59
really one of education, of getting out and making sure that all of our
17:04
technologists understand what's expected of them and how we expect them to think
17:09
about the tools that they're deploying from the beginning.
17:13
So, you know, if we educate everybody, actually it will cut down on the work for
17:16
this team, but all to the good because there's plenty of other work to do.
17:21
And I really want...
17:23
each of our associates to think of themselves as a responsible user of the
17:27
technology and data.
17:30
This is always a challenge with compliance though.
17:32
How do you convince them that this is valuable and it's not going to slow down
17:36
what they're trying to do?
17:37
Such a good question.
17:38
And it's funny because I don't say I bristle at those questions.
17:42
It's always been a perennial, like is privacy getting in the way of productivity
17:47
or of profits or whatever.
17:50
We really do believe in customer trust and associate trust and putting people first.
17:56
So I really posited as an issue of trust and responsibility and that that engenders
18:03
a better relationship with.
18:06
the company and that we are in a competitive environment where this is a
18:12
benefit.
18:12
It's considered a good thing, I think, by our customers.
18:15
We see more and more educated customers about technology and asking the questions
18:19
about how it's being used.
18:21
So I really don't try.
18:23
I don't think of it as as antithetical, but rather as kind of enhancing our
18:28
relationship with our customers.
18:30
And we do try to be lean.
18:31
Obviously, we try to be really efficient and not
18:35
every AI tool that's just deciding the example I always use and it is a real one
18:39
is, are the bananas on the conveyor belt right?
18:43
We've got AI that looks at fruit and vegetables and meat and things and says,
18:48
based on what it has learned over the last how many years and what a good thing,
18:52
right?
18:52
We want good quality food to be eaten and safe and that sort of thing.
18:58
That we do not consider, it's impactful to humans obviously indirectly, but we're not
19:02
making judgments about humans.
19:03
They do not need to go through the digital.
19:05
value scrutiny.
19:07
But I think frankly, in my experience, most people want to do the right thing.
19:11
They're not always sure how their job, you know, affects that right thing and the
19:15
service to the customer and associate.
19:17
So it's a big company.
19:19
And so education and outreach across multiple languages, multiple parts of the
19:23
world, multiple job descriptions, that to me is an exciting educational challenge.
19:30
Yeah, I mean, it sounds like you've got a screen that is similar to what many of the
19:36
governments are coming up with in terms of there's high -risk AI or whatever it's
19:41
described at.
19:41
This is what we really need to worry about.
19:43
Then there's all kinds of other uses of AI that maybe don't need the same scrutiny.
19:47
I hope people do look at it as a spectrum that not everything deserves or merits or
19:55
needs the same level of scrutiny, but some of it very much does.
19:58
And in the corporate world, the impactful decision -making that we would have would
20:04
obviously be in hiring, advancement, people's rights and responsibilities as an
20:09
employee.
20:10
In the customer world,
20:14
it's actually been even more fun because it's things like our AI chat bots and
20:18
putting together different fashion pieces on the website and saying, you bought a
20:22
red sweater, maybe this blue skirt goes with it.
20:24
This is fun.
20:27
Obviously, we want to be careful.
20:28
We don't want to get it right.
20:29
But it is not a consequential decision, depending on where you're going that day,
20:34
whether the skirt matches the sweater.
20:37
But again, of a different ilk than financial or career or other things.
20:45
How does the team stay up to date with the fast changing nature of technology in this
20:49
area?
20:49
That is such a great question.
20:51
And it is, I think, a constant part of everyone's job to be reading, to be
20:57
engaged.
20:58
We are members of quite a few, not only professional trade associations in retail,
21:03
in law, and that sort of thing.
21:04
I do find the IAPP, the International Association of Privacy Professionals, to
21:09
be increasingly aware.
21:11
And they just today put out a big report on AI governance.
21:14
and we're members of something called the Data and Trust Alliance, which was out
21:18
front very early on.
21:19
We were one of the founding members in looking at cross -industry guidance.
21:24
So it's not just one area of industry, not just one size of institution, but how do
21:30
we make governance real for all companies?
21:33
And in fact, we, as well as many small and medium enterprises, sometimes are
21:38
procuring things from third parties, from vendors.
21:40
And...
21:42
just like every other new technology, we see a flourishing of vendors coming in and
21:46
saying, this is the key thing, this is the perfect tool for everything, solution for
21:53
everything you've ever needed.
21:54
And many people don't know how to scrutinize that, right?
21:57
And we've had salespeople in our office saying, we can't really disclose the
22:03
algorithm and we can't really explain how it works.
22:06
And we're like, well, come back when you can, because that's not gonna be good
22:09
enough for our customers or associates.
22:11
So.
22:11
So rules around how to engage with vendors and how to scrutinize those tools as well
22:16
as the ones you're building internally I think are really important and that's a
22:19
role that DTA has played.
22:23
Do you have benchmarks or goals or a way to assess whether you're being successful
22:27
in these efforts?
22:29
I think benchmarking, especially when it's in the policy world, is hard.
22:33
But certainly some of our benchmarking is how many tools have gone through, how many
22:37
secondary reviews we've done.
22:41
Customer complaints are also a metric if things are making the wrong decision or
22:45
the right decision.
22:48
And so that's what we're looking at right now.
22:51
But if you've got any ideas, I'm all ears.
22:54
Well, no, it's an interesting challenge because on the one hand, there's value in
23:00
coordination and standardization.
23:03
And on the other hand, what makes sense for an organization at the scale of
23:07
Walmart is not going to be at the right thing for a startup or for a company
23:12
that's in a completely different industry.
23:14
And so that's why I'm curious about how different organizations think about and
23:18
see progress in those areas.
23:19
Yeah, that's such a good point.
23:20
I think there's some core values that probably cut across in terms of
23:24
transparency and kind of fit for purpose or, you know, scrutinizing high risk tools
23:29
more than others.
23:31
But we recognize also the place of privilege that I sit in and being able to
23:35
stand up an entire team to think about and then go do and create a whole new
23:38
compliance program around this.
23:40
And so well aware.
23:42
And that's why DTA set about creating kind of products and workbooks and things for
23:47
smaller medium enterprises that.
23:49
They're kind of off the shelf guidance as well.
23:52
Are there any major gaps that you see either in terms of what resources or tools
23:57
are available to you or in any other area?
24:01
I think we can always do more.
24:03
Again, back to the comment about governing the ungovernable, we can all do more to
24:08
really think about how to automate and really probe the tools in a consistent
24:14
fashion.
24:16
And like we were saying before, there's obviously a lot of concern around bias and
24:20
about accuracy.
24:23
But how do you prove that it is doing the thing that it's supposed to do and that
24:28
it's not doing the thing it's not supposed to do?
24:31
There's not a lot of standardization in that yet.
24:34
And so standard setting bodies, where are you?
24:37
I'm sure someone's working on a framework for us to all follow at some point.
24:42
And I know you and I were going to talk about legislation and government
24:47
interaction.
24:49
I get very concerned when I think government actors are saying, we have to
24:54
regulate this one thing.
24:56
Again, as we were saying, AI is, you know.
24:58
it's like the air or the, you know, the sidewalks.
25:01
It's just a tool and a platform for doing something else.
25:05
And I'm usually the one saying, is there a law that prevents this already?
25:09
You know, is there something on the books that simply hasn't been enforced in this
25:13
area or hasn't been applied to tools like this yet?
25:17
If it's really, really that new, you know, certainly it may need some level of
25:23
regulation or legislation.
25:25
But I'm usually the skeptic saying,
25:28
We may have rules about this already, about being unfair to people based on race
25:34
or gender.
25:35
And let's look at what's on the books already.
25:40
Yeah, I mean, that's a big challenge that we're seeing a lot where I talk to lots of
25:45
people who say, well, companies can't be trusted and their incentives are not
25:50
aligned.
25:50
And we know this technology is so new.
25:53
So how do you convince people that the resources and the mechanisms are there to
25:58
address the problems that arise?
26:00
I certainly hear the same things and I certainly know that big companies are just
26:06
whatever, big companies are made out of people, right?
26:09
And so...
26:11
It's not been my experience, not here and not in other big and small places I've
26:15
worked that people want to do the wrong thing.
26:17
In fact, I think our incentives are entirely aligned because if we lose the
26:21
trust of our customers, if we lose the trust of regulators in the countries in
26:25
which we do business, then the consequences are much greater than just
26:30
failing in one program or product.
26:32
It's a...
26:32
you know, an enterprise wide impact.
26:34
So I, again, I'm skeptical.
26:36
I understand it's good to have a healthy skepticism of large institutions in
26:41
general.
26:41
And I, and I respect that.
26:42
And that's why part of back to why we came out with the AI pledge is like, that is
26:46
something that we feel is worth talking about that we are committed to walking our
26:50
values forward in the digital world, in this tool, in this setting and in others.
26:55
And if people feel like we're not doing it.
26:58
They should let me know, they should let our CEO know, and they do occasionally,
27:03
including some of our own associates, which I actually love.
27:05
I love it when people raise their hand and say, this doesn't feel quite right to me.
27:09
And I've got a letter once from a pharmacist in Michigan, and he said, I
27:13
don't like that policy.
27:15
And it actually started a whole review from the ground up of how we were dealing
27:19
with a particular issue.
27:21
So I think we try to be very responsive.
27:24
Something I've tried to talk about within the company is,
27:27
and all of the good governance we've put in on these tools and others, mistakes may
27:33
still happen.
27:34
That's the really big challenge I find, especially for a company that prides
27:37
itself on its reputation and doing things right.
27:41
It's not an environment where we'd like to fail fast and move on.
27:45
We really do kind of bet our reputation on a lot of things.
27:49
And what I try to encourage my team and others in leadership is to realize that...
27:55
If a mistake happens, we're going to have to work hard and fast to fix it.
27:59
But I think people will give us grace if we are transparent about it.
28:03
If we say, this is what happened.
28:05
Here's what we're doing to fix this incident.
28:09
And here's what we're putting in place to have it not happen again.
28:12
So especially in kind of generative AI, when it moves and talks and it moves fast,
28:18
we need to be ready to both be humble and be nimble in that area.
28:25
Yeah, I mean, one of the challenges with generative AI is what you said touches so
28:28
many people that now it's not just the data scientists who are building things.
28:35
What does trust really mean to you in that context?
28:38
I think it's a really important and valuable point that this is all about
28:42
creating and promoting trust.
28:44
But what does that actually boil down to?
28:46
On the website, it means we've correctly described a product.
28:52
Say you've got a child who's allergic to peanut butter, you better be sure that the
28:57
product you've bought does not have peanut butter in it.
28:59
That's really important to us, and that was important to us before there was
29:02
generative AI.
29:03
But now if you have generative AI creating product descriptions, you've got to make
29:07
sure they're right.
29:08
And as we know, generative AI is not always right.
29:12
So in some tools we're just about to put out in the...
29:17
We used outside consultants to red team, to probe it, to try to crash it, and
29:24
things still do go wrong, but I think putting in the effort before it goes live
29:30
to really try to stress test it and then to stay, again, really vigilant about how
29:36
the AI might change is essential.
29:39
I feel, again, a deep sense of responsibility, not just because it's the
29:42
catchphrase of our team, but I think...
29:46
all kind of purveyors of technology should be thinking about the impact it's having
29:50
on the world and on the people it comes into contact with.
29:54
And going forward, are there either developments that you're looking to or
29:58
anything that you see coming in the next few years that are going to be
30:01
particularly important in this area?
30:04
Well, I can tell you the things I'm the most worried about, unfortunately, it's
30:08
not the way to start to answer that question.
30:10
But the things that I worry about are the appearance of accuracy and truth when it
30:18
is not there.
30:19
It's too cute to say fake news, but just simply the authoritative answers that
30:25
often are provided by these tools that are simply incorrect.
30:30
Do all of our citizens in this country or elsewhere?
30:34
have the ability to be good consumers of this kind of media savvy or media, you
30:40
know, digital kind of citizenship in the true sense of can we consume and analyze
30:48
information for its content and for its provenance?
30:53
And do we have the tools to help us do that?
30:57
I saw a great study years ago about kind of government issued propaganda and
31:04
Americans were more likely to click on something that was, it kind of seemed to
31:10
be true, but it was a little bit off than many parts of Eastern Europe.
31:15
And I thought, wow, that kind of shocks me.
31:17
You know, we have free press and we have all this stuff.
31:19
And it was because they had lived with propaganda, perhaps in more recent times
31:23
than we had.
31:25
And I thought, well, how do we get to, you know, a place where we have really
31:29
empowered?
31:30
Internet users of all ages, shapes, sizes, everything to scrutinize the information
31:37
or perhaps have penalties for people who put out things knowingly that are
31:42
incorrect.
31:43
That's to me one of the hardest questions for the Internet community, for the world,
31:48
and one of the consequences of gendered AI that worries me the most.
31:54
The other one is our kids using it to write term papers.
31:57
But that's me.
31:59
The exciting thing to me is always, not just the general utility for human beings,
32:05
but I have a son who's on the spectrum.
32:08
We have lots of associates who are disabled or in some way we have many
32:11
customers.
32:13
We've just started, you may have seen quiet hours in the stores where every
32:17
morning for two hours, everything in the sound wise and the lights are lowered.
32:22
And it's been a real, it's been a joy to watch that evolve and people really get
32:27
excited about it.
32:28
I always call it kind of the ancillary effects as well.
32:33
People of all kinds who may not fit in its traditional kind of disability community
32:39
have said, some people just like it quiet.
32:41
My husband says I would much rather shop when it's quiet.
32:44
And whatever.
32:46
And so the thing I'm the most excited about, not just about AI and gender AI,
32:50
but technology is the ability to assist humans who are physically, cognitively
32:57
otherwise.
32:58
diverse or challenged in some way.
33:01
And Gen .AI, especially in voice enabled, is just a boon for many, many people.
33:06
And I'm so excited about that.
33:08
What that means for our workforce, enabling more people to join our
33:12
workforce, generally join the workforce, and enabling customers to find the
33:18
information they want and the goods and services they want.
33:21
I think that's one of the best and highest uses of technology.
33:25
No question.
33:25
No, it's all tremendously exciting and it's sort of the paradoxes in and of this
33:30
technology that what makes it exciting is also what in some ways makes it scary.
33:34
I mean, you talked about sort of two sides of citizenship in both the corporate and
33:39
the public governmental context where this technology can be so great at
33:43
personalizing, but that can also be abused.
33:46
So true, so true.
33:47
And transparency is necessary but not sufficient answer.
33:52
Absolutely.
33:53
Nuala, really great to talk to you.
33:54
Thanks so much for sharing your experiences and what you're doing.
33:57
Delighted to be here.
33:57
Thanks for having me.