Latest / The Road to Accountable AI / Olivia Gambelin: Ethical AI as Good Business Practice

Transcript

Olivia Gambelin: Ethical AI as Good Business Practice · The Road to Accountable AI

0:02
Olivia, welcome to The Road to Accountable AI.
0:05
Thank you so much for having me here, Kevin.
0:08
What first attracted you to the AI ethics field?
0:13
I would say it was probably a couple different factors, but at the core of it,
0:17
I'm just really fascinated by ethics and morality.
0:19
Everyone in life has different lens that they see the world through, they try and
0:23
interpret the world through.
0:25
For me, for whatever reason, ethics has been that lens that I'm trying to
0:29
understand how people interact with each other, what we do in life, what drives us.
0:33
So at the root of it all, it was really my love of ethics.
0:36
And then AI just provides such an interesting, almost mirror back towards
0:41
us.
0:41
So...
0:42
perfect place for someone interested in ethics to really sink their teeth into.
0:47
Yeah, that's interesting.
0:48
So talk more about what it is about AI that you think makes it particularly
0:51
interesting for someone who focuses on it.
0:54
Well, I would say one of the main reasons that AI is such an interesting, almost
0:58
like playground, I want to say, when it comes to ethics, is the fact that we are,
1:03
in a way, trying to mimic or emulate our own ethics and morality, our own human
1:08
behaviors.
1:09
So when it comes to the kind of decision-making that we're putting into,
1:12
say, neural nets and so on, that is at certain levels.
1:16
And yes, there are deviations, but it is an echo of our own humanity and our own
1:21
human nature.
1:22
And so what it's doing for us is it's highlighting these really fascinating
1:27
ethical challenges really that we as humans have been dealing with for
1:30
thousands and thousands of years.
1:32
And it's bringing it into like a, like a new lens.
1:35
We're seeing it from a new angle where we're almost removed from the equation and
1:40
we can see how our actions play out and our decisions play out and then be able to
1:44
go from there and say, well, how do we actually reverse engineer this down to the
1:48
roots, which is us as people and understand what
1:51
How can we make better decisions?
1:52
How can we, let's say, fine tune our understanding of ethics and morality?
2:00
So if there are these hard ethical challenges that we've dealt with for a
2:03
long time, and these scenarios where the right answer is not necessarily obvious,
2:09
how does that translate to a company saying, okay, I'm gonna deploy this system
2:13
and I've got to make a choice about how I design it?
2:16
How can that lead you to actually give in constructive advice?
2:20
Yes, that actually ends up as one of, I would say one of the first mental blockers
2:25
to companies engaging with ethics because they can come at it at this standpoint,
2:29
this perspective of, well, if there are so many challenges and so many, well, it
2:34
depends kind of answers, how can we actually create concrete action and
2:40
practice to be able to apply to our solutions?
2:43
And so what I like to point out here is it's the
2:47
It is the practice of going through and figuring out, well, where could this go
2:51
wrong?
2:52
That we then get to start to pinpoint, well, here, it can go wrong here.
2:56
And we can actually put in safeguards for if and when that does happen.
3:01
It also enables us really to truly predict in what ways that we need to innovate,
3:08
honestly, in AI to be able to counterbalance some of those ethical
3:12
challenges.
3:13
or to understand where some of those shortcomings are going to lie, where AI,
3:16
for example, shouldn't be applied, and instead we still need the human
3:19
decision-making factor in place.
3:21
So it's less, ethics is really, when it comes to practicing in AI, it's less about
3:27
having these concrete black and white decisions, and it's more being able to
3:32
predict, very, very forward-facing, but that predictive problem-solving and being
3:38
able to retroactively...
3:43
fix some holes that have been there for a while.
3:46
Yeah.
3:47
So when do companies typically contact you?
3:50
Is it, they're starting down the path of deploying an AI system or they're trying
3:54
to have a broad framework or something has gone wrong or they're worried something in
3:59
particular is going to go wrong.
4:01
So AI ethics and responsible AI is a really rich field.
4:03
And there's a lot of us that have different niches.
4:06
My niche specifically is at the very beginning.
4:09
So there are a lot of ethicists that will work on the development cycles.
4:12
There'll be hand in hand with engineers.
4:14
I sit more with executive and leadership teams and sometimes even design teams
4:19
because my focus is on how do we develop a strategy, a responsible AI strategy for
4:23
putting good practices in place from the start, that's going to enable ethical
4:28
decision-making and enable best practices in AI.
4:31
But then also on that design side, I'm really sitting with the teams and the
4:36
strategy of, well, how do we actually, I have my specialty of ethics by design.
4:41
I find it fascinating.
4:42
So how do we actually design for ethical values from the very start?
4:46
This is me.
4:47
So I tend to work with companies that are more proactive, that are looking to use
4:52
the responsible and ethics lens, really, as a competitive edge.
4:56
but that also understand that this is long-term where the market's going to go.
4:59
So they're getting out ahead of their competitors.
5:02
I know, for example, though, I have a lot of ethicists and fellow colleagues that
5:06
are brought in retrospectively when something's gone wrong and they need to
5:09
fix it.
5:10
But I tend to sit on where do we start and answering that question of what am I
5:15
missing?
5:17
What's the first thing that a company needs to do?
5:20
Woof.
5:21
Well, as a classic ethicist, I'm going to say it depends, because it always depends
5:26
on the company.
5:27
But from my perspective, one of the first things a company needs to do is take stock
5:31
of what they already have in place.
5:33
So I will actually, I'm about to launch, and I guess when this podcast is out, this
5:39
will already be out.
5:40
But it's something called the Values Canvas.
5:42
And it is, if you've ever seen the business model canvas, it's basically a
5:47
sheet of paper, or one document.
5:49
And it gives you this map of what am I missing?
5:52
What do I need to see?
5:53
What do I actually need in place?
5:55
And this values canvas, it goes over the people, process, and technology layers
5:59
that go into developing AI and using AI responsibly.
6:04
So from my perspective, and I'm a little bit biased here, that's actually a great
6:08
place for companies to start because what it does is it empowers them.
6:11
They're seeing, oh, this isn't the Wild West of I need ethics question mark.
6:16
What does that mean?
6:17
It's.
6:18
very practical of these are the solutions.
6:20
These are the operational mechanisms and the protocols that we need in place to be
6:25
able to even enable engaging with any ethics decision-making.
6:29
So for me, starting with that values canvas, which also actually intrinsically
6:35
implies as well, knowing your values.
6:39
So you mentioned the values canvas.
6:41
I assume that's part of your new book, Unresponsible AI.
6:46
So tell us a little bit more, just more broadly about what you focus on in the
6:49
book.
6:51
So the book, it's out in June 2024.
6:54
I'm very excited.
6:56
It's Responsible AI, Implement an Ethical Approach in Your Organization.
7:00
And it's really written for leadership and the next executives in artificial
7:07
intelligence who want to take responsibly AI seriously.
7:11
What this means is the book is designed to take you from A to Z in developing
7:15
responsibly AI strategy.
7:17
In the book, I don't offer, say, specific solutions in terms of technical solutions.
7:22
Instead, I sit on a few layers above it, looking at that full picture view of, what
7:27
do I actually need in place?
7:29
Who is responsible for responsible AI?
7:32
What are our first steps?
7:33
What goes into a full holistic strategy?
7:36
And really as well, how do I conceptualize ethics and how do I find my values that
7:42
need to go into our practice?
7:44
So it covers a lot of these different pieces.
7:46
It's a thick book, but it really is, well, it really will be a great starting point
7:52
for executives to get their foot in the door when it comes to responsible AI.
7:59
How in the book do you make the case for why companies and executives should invest
8:06
in this?
8:07
So that is actually my favorite chapter.
8:10
It's chapter five of the book.
8:11
And I was feeling particularly sassy when I went to write that chapter.
8:17
And I basically had reached the point, I was tired of constantly singing the same
8:23
song, let's say.
8:24
Like there are very, very clear benefits to responsible AI, such as talent
8:29
acquisition, reduce costs in terms of development, reduce failure rate.
8:34
um, increase in brand equity, so on and so on.
8:36
But these are, this is the song that I'd been singing for so long and that a lot of
8:40
others in my field have been singing.
8:42
And so I sat down to write that chapter and I thought, you know, I don't, I don't
8:46
want to write that again.
8:47
I've said it before.
8:48
I give it in my talks.
8:49
I it's been done.
8:51
I'm kind of tired of getting these, these objections.
8:55
And so I said, well, let me actually write, let me actually map those
8:58
objections.
9:00
So I wrote down all of the different reasons that people usually tell me why
9:04
they're not investing in Responsible AI.
9:07
And I broke those down into five categories.
9:09
And so the title of that chapter is actually five reasons why not to invest in
9:16
Responsible AI and one reason why you should.
9:18
I can't in good conscience just tell you why not to.
9:21
But really what I did with those objections is I sat there and I said, OK,
9:25
let's break these down.
9:26
Let's actually put them in context
9:29
the logic behind them, are they well-founded or is this something more
9:33
just a fear that's sitting behind it?
9:35
And you know, it covered everything from, well, responsibly AI takes time.
9:39
Yes, it does.
9:40
Responsibly AI costs money.
9:41
You have to invest into this.
9:42
This is not something that you pull a framework off the internet and say you're
9:46
good to go.
9:47
It does take time.
9:49
It can be confusing when it comes to external communications.
9:53
You're at risk for ethics washing.
9:55
And I really validated all of these pieces.
9:57
But then at the same time, I put it in the context of AI and I compared it to
10:01
companies investment in AI.
10:02
And it comes down to the fact where, as you were reading within the
10:06
contextualization of these points, it was like, well, actually, these are just good
10:11
business practices.
10:13
Why wouldn't I actually invest in this space?
10:17
One of the biggest ones actually that stands out to me is responsible AI costs
10:22
money, yes.
10:24
Well, when you look at investments in artificial intelligence, either an
10:27
adoption or the build of it, the failure rate risk of any AI investment sits
10:33
between 83 to 92%.
10:34
That's insane.
10:36
For the amount of money that gets invested into an AI system that will most likely
10:42
fail, those are huge costs.
10:45
And companies will invest...
10:48
thousands, if not millions of dollars into solutions that will reduce that failure
10:52
risk rate by even 1% to 2% because that margin of failure rate reduction is huge
10:58
in terms of cost saving.
11:00
Responsible AI, thanks to a study done by BCG and MIT, has actually proven to reduce
11:06
that failure rate risk of failure rate, it's a bit of a tongue tire, by 28%.
11:13
That's unheard of, the savings on that.
11:16
So when you contextualize the cost that it takes upfront to put responsible AI
11:21
practices into place, they are tenfold counteracted by the amount of savings that
11:26
you get when it comes to reducing that failure rate in terms of AI investment.
11:32
So the responsible AI practice will actually improve the effectiveness of the
11:37
AI itself.
11:40
How does that happen?
11:41
Well, at the end of the day, Responsible AI is basically just good business
11:44
practice.
11:45
It focuses on standardizing your data practices.
11:49
It focuses on standardizing your documentation of your AI models.
11:53
These are components that should have existed from the start, but are actually
11:57
some of the biggest blockers to companies when they start to try and even scale AI
12:02
in-house.
12:03
They haven't put these good practices in place.
12:05
Other aspects of it include training.
12:07
Your people need to know what they're doing.
12:09
or how they're doing it.
12:10
They need to have these operational frameworks.
12:13
So when you look at it from this perspective, all I'm really regurgitating
12:17
here are good business practices, good development and business practices.
12:21
It's just through a specific lens of being able to translate values into action.
12:26
But that's really at the end of the day, Responsibly AI is a good business
12:31
practice.
12:31
These are going to be standard practices in the future.
12:35
But they're not yet.
12:36
Not yet, exactly, which is always surprising to people.
12:40
I think the biggest one is especially around that documentation case is there is
12:45
a lack of standardized documentation when models are developed.
12:48
So one person can develop the model, leave the company, and it could be a company of
12:53
thousands of people and no one knows what that model does or how it works because
12:57
the documentation isn't there.
12:59
That should be a no brainer that the documentation's there, but it's not.
13:03
This technology was developed kind of in the wild west.
13:05
Now we're putting in best practices and standards.
13:09
Is the biggest challenge just the willingness and understanding of companies
13:14
that they need to document?
13:15
Or is it that there aren't formalized standards or templates for that?
13:21
I would say at the start, there weren't formalized templates.
13:25
Now there are immense amount of online resources to at least get you started and
13:30
examples of companies.
13:31
So even when you're looking at documentation around your data, NVIDIA has
13:36
put out these amazing model cards.
13:37
You can quite literally go online and find these model cards and start using them
13:41
today.
13:41
They're a great example of standard documentation processes.
13:46
The challenge was when we first started deploying AI and companies started just
13:50
first engaging in building AI systems, this was before any of those standards
13:55
existed.
13:55
So it really was the wild west.
13:57
And there is this pressure of we have to develop faster.
14:01
We have to go faster.
14:02
And when you're going faster and faster and faster documentation is one of the
14:05
first things to get taken off the list.
14:07
Cause you're like, I don't, I'll go back and do it later.
14:09
It doesn't really matter.
14:10
So that combined effect of that, that pressure to move fast and break things.
14:15
And the lack of standards to start, or the lack of formal templates to start, was
14:19
really one of the major challenges at first.
14:23
What else has changed in the years that you've been doing this work as we've moved
14:27
into this period of time where there's now just an immense amount of excitement and
14:32
discussion about AI?
14:34
Well, I've been at this for a while.
14:37
I would call myself a veteran of this space, which still sounds funny because
14:40
it's a new space and I still feel new to it, but you know, six, seven years in, I
14:46
guess I've been around.
14:48
Well, when I first started out, a lot of what I was doing was truly, honestly
14:53
building awareness.
14:54
I would say I'm an AI ethicist.
14:56
I would give that introduction and people would look at me like I had a third arm or
15:01
something.
15:01
What in the world is an ethicist?
15:03
What do you do?
15:04
So a lot of my work early on was just trying to explain, this is what ethics is,
15:08
and this is what it is in the context of AI.
15:12
Now, over the years, we've had actually some key points in time that have changed
15:17
that narrative and changed the application.
15:19
So early days, it was a lot of this is what ethics is.
15:23
And even back then, I was really trying to get my mind around what are the solutions?
15:27
It did feel like a Wild West, even for an ethicist.
15:29
What are the practices?
15:30
What does a company, where does a company start?
15:33
It was a lot of conversation, a lot of co-development and building and research
15:38
in those early days.
15:40
Then we hit COVID-19 and we had all of this COVID tech come out and all of a
15:45
sudden the entire world was in a thought experiment around what it means to have
15:51
technology developed from the start.
15:53
And that brought AI ethics and responsible AI out from the shadows and into a main
15:59
line of conversation.
16:01
And then we started seeing some more standards developed and we started seeing
16:05
these policies and let's say high level lists of values come out that we were able
16:14
to start to see some patterns and say, okay, we've got some basic standards set
16:19
down.
16:20
And then we had generative AI hit.
16:22
And obviously generative AI existed before chat GPT, but chat GPT was really the
16:27
start of this boom.
16:28
And now
16:29
At this point in time, companies have actually moved to, we don't need to talk
16:33
about this anymore.
16:34
We actually really need to do it.
16:36
And so this has been an amazing point in time to be able to see the early stages
16:42
and be here where it's now becoming a requirement.
16:47
And now we're really getting these amazing case studies, seeing some really
16:52
fascinating solutions come out.
16:54
So it's been a journey, let's say, over the last years.
16:59
What are some of the best case studies you would point to or examples if someone
17:04
wants to say, here's an organization that's doing it right?
17:07
Well, I actually am about to release a great resource on this because Kevin, I
17:13
get asked this question all the time and you can hunt for them.
17:17
So actually one place that you can immediately look is the world economic
17:21
forum did an entire case study on Salesforce and that is a very in depth
17:25
case study and a really brilliant one in terms of the intricacies there.
17:29
Um, what I'm doing is along with the values canvas that I, I guess, when this
17:35
podcast come out.
17:36
comes out is published, I'm starting to create an archive of case studies.
17:40
So I am about to start the release of a series of nine different case studies of
17:44
companies that do it well, and specifically on different fine-tuned
17:51
solutions.
17:53
This will make sense when you check out the Values Canvas.
17:55
I've got nine different types of solutions on it, hence the series of nine.
17:59
But that is going to be a fantastic resource of companies that do it well.
18:04
And then I'm opening it up as an archive.
18:06
And the idea there is to have companies and individuals be able to submit, hey,
18:11
this is what we've done in terms of this kind of solution.
18:14
Just to create that body of case studies, because this is a question I get asked all
18:18
the time.
18:19
And unless you hunt for them, they're far and few between right now.
18:23
And it's hard to find that central database.
18:28
Can you generalize it all across the companies you see that are doing
18:31
responsible AI well?
18:33
Absolutely.
18:34
So the companies that I see it doing really well, either, okay, this is on the
18:38
larger end, because of course we've got startups all the way to Fortune 100
18:43
companies.
18:44
And I would say in general, companies that do Responsible AI well have two main
18:50
factors.
18:52
One is they understand that Responsible AI is a holistic practice.
18:55
They are not just looking at technical solutions, they're Responsible AI
18:58
initiatives.
18:59
do not just exist in their engineering or programming or developer departments.
19:04
It exists across the entire organization.
19:07
And that's really important.
19:08
The people and the process behind the technology are just as important, if not
19:13
have more impact at this point in time on the outcomes than, say, the actual
19:17
technology.
19:19
So companies that do it well understand that.
19:21
They have a robust governance system going in.
19:24
They have training going in.
19:26
They have more.
19:27
They are incentivizing their employees to engage with these practices.
19:30
It's a holistic initiative.
19:33
That's the first factor.
19:34
The second factor is companies that do this well have a team and this can look
19:40
different depending on the company.
19:42
Some companies will spread out their team and have people placed within different
19:46
departments.
19:47
Other companies will go for more of a centralized structure and have the team
19:53
concentrated in one department.
19:55
But the bottom line is someone is responsible.
19:58
Someone's in there under the title of head of responsible AI, chief ethics officer,
20:04
lead AI architect, AI ethics architect, these kind of positions.
20:10
Because what it does is it creates the budget and the gravitas and the
20:14
credentials and resources that are needed internal.
20:18
It's showing that commitment and it's putting at least a mind or two on a very
20:23
complex problem.
20:25
You need, this isn't something that you do at the end of the day or that you tack on
20:30
to another person's job.
20:32
It is really complex.
20:33
You do need someone or the larger the company gets, you do need teams to handle
20:38
this.
20:40
And conversely, other than not doing those things, what's the biggest mistake that
20:44
you've seen companies make in this area?
20:48
Well, from the startup perspective, I know I'm a bit all over the place on up and
20:52
down on the size scale, but from the startup perspective, and I'll see the
20:57
sentiment carried as well into larger enterprise is the sentiment, the kind of
21:04
mistake where someone says, well, I'm an ethical person, therefore I'm fine.
21:10
It's not a matter of whether or not you are ethical.
21:13
It's not a matter of,
21:15
whether or not you're a good person and make good decisions.
21:17
That is completely separate from actually applying ethics and working through
21:22
responsible practices.
21:24
That's actually a very important separation to make because people can get
21:28
very sensitive and nervous to engage in this topic of, well, I'm a bad person.
21:32
That's not true at all.
21:34
Your ethics and your ability to execute on ethics within a company structure are
21:39
completely different.
21:41
So,
21:41
If you have this kind of mental blocker of I am a good person, therefore I don't have
21:45
to worry, you are missing the fine tuning of this ethical decision making.
21:50
You actually need to be able to surface those ethical decisions going into the
21:54
company and going into the technology in order to be able to work with them.
21:59
Otherwise, you are.
22:00
Excuse me.
22:09
I dusted right before we did this, which I think was a mistake.
22:14
Exactly.
22:15
I had to, I had to pick up my thought.
22:26
If you're not aware of the decisions that are going in, you are creating these
22:30
biases that are, you are embedding biases that you are not even aware of.
22:35
And those biases can be for good.
22:36
They can be for not so good.
22:38
but you need to be aware of them and you need to be able to fine tune them.
22:40
So I would say that's actually a huge blocker in terms of the cultural
22:45
perspective of this is, um, I'm an ethical person, therefore I don't need to do this.
22:50
And the other one again, is that mindset conversely, uh, companies that do it well,
22:55
it's a holistic solution.
22:57
Companies that are stuck are ones that are saying we're going to solve this in the
23:00
code and that's it.
23:02
And we don't go beyond that.
23:05
Yeah, that's an important point.
23:06
And it's one of the things that I look at where, for example, we talk about
23:10
algorithmic bias.
23:11
And there are lots of these examples of systems that have racially biased effects
23:16
or gender bias.
23:17
And it's always a question, well, is that what the company really intended?
23:23
Or did they just screw up?
23:25
Or was there just some simple thing they didn't do?
23:29
And of course, you're not necessarily going to be perfect with some of these
23:32
systems.
23:33
You can't be absolutely certain there will never be any, any bias.
23:36
So how do you advise a company in terms of the right way to think about it?
23:40
The challenge.
23:41
Yeah, I would actually say to look at the bigger picture and take into account the
23:48
worst possible scenario.
23:50
So let's say you do push into production a model that ends up having an extreme
23:55
racial bias.
23:57
What are your contingency plans?
23:59
How do you then go back and fix it?
24:01
What feedback loops do you have?
24:03
This is actually something very important.
24:05
Do you have tight enough feedback loops where you can actually catch that problem
24:10
early on?
24:11
Because if you have the feedback loops early on where you're seeing, where you're
24:15
actually in communication with the people that are being affected by the model or
24:19
the people that are using it, who can give you that live feedback of saying, Hey,
24:22
there's something off here.
24:24
We feel like that there is a bias here.
24:26
You can catch it before it becomes a bigger problem.
24:30
So it's less about trying to completely eliminate bias.
24:34
That's impossible.
24:35
And it's more of what are you doing outside of that?
24:37
How do you get the feedback in?
24:39
And then if something goes wrong.
24:41
How do you then go back and fix it?
24:43
How do you then communicate externally?
24:47
You have the steps afterwards and assuming the worst, but hoping for the best, of
24:52
course.
24:54
Absolutely.
24:54
That goes back to what you said before.
24:55
It sounds like that's also just good business about having mechanisms to
25:00
monitor and have oversight of things in deployment.
25:02
Yeah, exactly.
25:04
When you say it like that, sounds like a very simple thing to do.
25:07
Sounds good when you say it too.
25:08
So absolutely.
25:11
What about the maturity of the community that you're in, the community that's doing
25:15
AI ethics or responsible AI?
25:17
I know one thing that you've done is created an association for people in the
25:21
field.
25:23
How far along are we in terms of the development of sophistication and standard
25:28
practices and so forth for AI ethicists and responsible AI practitioners?
25:33
Yeah, that's actually a great thing to bring up.
25:35
So Kevin, actually at the start of this year, we launched something called the EI
25:40
Network, which you are an amazing member of, but we have the EI Network and that
25:45
really came from this sentiment of I was going, I do a lot of professional speaking
25:51
and I end up at a lot of people's events.
25:53
And I have tons of people coming up to me saying, oh, you're the first ethicist or,
25:59
you know, I also work in Responsible AI.
26:01
I don't know.
26:02
anyone else who ever talks about this.
26:05
And so while I was having these individuals coming up saying, oh my
26:09
goodness, I'm so excited to find you.
26:11
There's someone like me.
26:12
There's someone that does a similar job to me.
26:14
I was sitting on the other end going, you're the 10th person that's told me that
26:18
today.
26:19
What in the world's going on here?
26:20
We need a place for us to all come together.
26:22
And so that's ended up spurring the start of the EI Network, which is this growing
26:27
global community of responsible EI and ethics practitioners.
26:31
The community itself, both in the EI network and beyond the responsible AI
26:36
space, is really a diverse group.
26:39
We're coming from all different angles.
26:41
We're coming from all different specialties.
26:43
So I say, I'm an AI ethicist.
26:45
My focus is on ethics.
26:47
I come from a traditional ethics training background versus I have colleagues that
26:51
come from engineering backgrounds.
26:53
They're more on the technologist side.
26:55
I have colleagues that are lawyers by background.
26:58
I have colleagues that are design.
27:00
designers by background.
27:02
It's this holistic picture and people coming from all these different angles,
27:07
which makes it honestly a really, really fascinating space to be a part of because
27:11
of the interdisciplinary nature of it.
27:13
You have to have these cross communications.
27:15
And so you meet a lot of people, a lot of people that I will lovingly call very
27:20
quirky in terms of interests, because when you're working in Responsible AI, we have
27:26
a lot of different interests.
27:27
We're polling.
27:28
insights from different fields.
27:29
We're not just siloed in one, we actually have to have that communication throughout
27:34
the different fields and perspectives that go into building AI.
27:38
So it makes for a very rich tapestry, let's say, of a community.
27:44
Given that interdisciplinarity though, should a company that's thinking about
27:48
hiring someone, should they ask themselves, do we want someone from an
27:51
ethics background or someone from an engineering background or is that not the
27:54
right way to think?
27:55
That's actually a great way to think about it.
27:57
And it requires the company to really understand what their need is in the first
28:01
place, which is a different challenge.
28:04
If the company doesn't understand first and foremost, hey, we need someone that's
28:08
going to be in with our engineers and working on the actual product, or hey, we
28:12
need someone more from a business analyst background that will help us design the
28:16
business models, or we need an actual ethicist that's going to be able to
28:20
research and give guidance on key decision-making.
28:24
If the company doesn't understand that yet, I would say the best place to start
28:28
is with, with someone that comes from more of a strategy background to be able to
28:32
build out that entire, entire initiative.
28:36
There are different specialties and they are very different specialties.
28:39
For example, I do not code myself and it is something that has never stood in my
28:46
way.
28:47
So the people that are listening that want to work in this field, you don't have to
28:49
code and have it, and you can still have a very great career in this space.
28:54
Especially with generative AI now, it'll write the code for you.
28:57
Exactly.
28:58
But at the end of the day, I'm not going to take up projects that require me to be
29:05
working on the actual development and code of these systems.
29:10
I have colleagues that I work very closely with that I have a good rapport built up
29:15
to be able to execute on that kind of communication.
29:21
How do you think that the work that you and the others in this community do is
29:25
going to be influenced as we now have significant amounts of legislation and
29:30
regulation like the European AI Act that are going to come into force putting
29:35
obligations on companies in this area?
29:38
Well, there's definitely going to be a huge amount of us working on compliance
29:43
and regulation for sure.
29:45
But I think this is also something to really highlight here is we're going to
29:50
start to see a huge difference in terms of companies that are just going for
29:54
compliance versus companies that are going for responsible AI.
29:57
The compliance and the regulation is, think of it as like the foundation for the
30:01
house.
30:02
That is your bare minimum.
30:04
That's your baseline.
30:06
It's very important.
30:07
to be compliant.
30:08
I am not dismissing that at all.
30:11
It is very important.
30:12
Those are regulations for a reason.
30:15
But that is not the full story.
30:17
You can be perfectly compliant and still very unethical.
30:21
Responsibility and Ethics is the house that's built off of those foundations.
30:25
So when it comes to working in this space, we're going to see a lot of people working
30:29
on audits in the future.
30:31
We're going to see a lot of people working on compliance and risk mitigation factors.
30:35
But then we're going to see also this beautiful, what is my favorite part to
30:41
this space, but this beautiful side to it where people are working more on that
30:45
strategy and that design and looking at ethics by design and looking at getting
30:50
these responsibly our practices in place so that compliance actually becomes, I
30:54
think it's like 10 times easier at that point to be able to comply with these
30:59
incoming regulations.
31:00
So it is going to impact our space.
31:02
I think the biggest way that it, that
31:04
it will impact our space is, or at least in the close future, is the need for
31:11
people like me in this space to really be able to stress and to highlight the
31:18
difference between regulation, compliance, and practicing ethics and responsible AI.
31:23
We're really going to have to make that clear.
31:28
And last question, what do you see going forward as the biggest challenge that you
31:33
face in the work that you do?
31:36
Woof, that is actually a great question.
31:39
I think right now on a personal level, my biggest challenge is having enough hours
31:44
in the day.
31:45
I think if I had 48 hour long days, that would be perfect.
31:48
But for me, my...
31:54
not necessarily challenge, but my next area that I'm diving into is further down
32:02
the rabbit hole of ethics by design.
32:05
But I'm specifically looking at instead of looking at the causes of ethical outcomes
32:15
in design features, I want to look at actually how people experience these
32:21
ethical values.
32:22
and how people experience, say for example, users experience privacy or
32:26
fairness.
32:28
And then how do we then reverse engineer that into looking at what design features
32:35
need to be put in place for that.
32:36
So flipping things on their head, instead of looking at the cause for these
32:41
experiences of ethical values, I wanna look at the effects and see how do we
32:46
reach those effects that we're going for.
32:48
and expanding too beyond privacy and fairness and looking more at some human
32:53
values like empathy and trust and loyalty in those.
32:58
So I wouldn't say that's a challenge in terms of what's being, what the big
33:06
mountains are to climb, but it's definitely the space that I think is, the
33:11
time is now and is opening up and I'm really fascinated and digging in deeper
33:14
there.
33:16
Really look forward to seeing what you come up with.
33:17
Olivia, thanks so much for the conversation.
33:20
Thank you so much, Kevin.