Latest / The Road to Accountable AI / Heather Domin: From Principles to Practice

Transcript

Heather Domin: From Principles to Practice · The Road to Accountable AI

• 0:00
  Heather, welcome.
• 0:01
  It's really a pleasure to have you on the podcast.
• 0:04
  Glad to be here.
• 0:06
  Tell me a little bit about your background.
• 0:08
  How did you get involved in the area of AI governance?
• 0:13
  Yeah, so for me, it's been a number of years now of interest.
• 0:18
  And in fact, going back before AI governance was a thing.
• 0:22
  um think, you know, even as I reflect back on ah my early years of education, I always kind of skewed my projects and the things that I studied towards um the ethical trust
• 0:38
  areas and how they intersected with technology.
• 0:42
  And so, you when the opportunity came to uh actually work in the space, it was very appealing to me.
• 0:49
  had the opportunity to, while working at IBM, uh help establish many of the foundational practices that, you know, we were setting up at the time around AI ethics and governance.
• 1:04
  and have worked then with a number of groups like the IAPP to help establish the profession.
• 1:11
  I was one of the original subject matter experts brought in to set up the AI governance professional certification.
• 1:18
  I now co-chair that certification.
• 1:21
  And I now lead the Office of Responsible AI and Governance at HCL Tech and get to continue working in this space.
• 1:29
  So it's a pleasure.
• 1:32
  Let's go back to the beginning.
• 1:33
  You said you always were interested in these ethical questions and you have a PhD, I believe, in data science.
• 1:41
  what appealed to you about these kinds of topics back then?
• 1:46
  Yeah, so I think it was the types of questions that were coming up in, you know, common conversations with my family.
• 1:58
  think it's one of the ways that, you know, I have a lot of academics in my family, a lot of uh people who like to think deeply about tough questions.
• 2:08
  And um at the same time, uh
• 2:11
  You know, my brother is in um computer science and we have lots of like technology focus.
• 2:17
  And so it's this intersection of, you know, this stuff is amazing, but, you know, why are people concerned in the news?
• 2:26
  Like what's happening and kind of dissecting that.
• 2:28
  So it's a very interesting, um you know, philosophical challenge, m social challenge, as well as a technical challenge in many cases.
• 2:40
  And so
• 2:41
  I was always curious then, okay, what made the system do that?
• 2:46
  Why did you get that outcome?
• 2:48
  And so that uh particular answer was something I've always been interested in trying to answer.
• 2:58
  Yeah, that question, yeah.
• 3:00
  Do you feel like data scientists and engineers and computer scientists more generally are more willing to accept now that there are ethical and other dimensions of what they do as
• 3:12
  opposed to saying, well, I just build the system to work effectively to its specifications.
• 3:17
  That's for someone else to
• 3:20
  Yeah, I've seen a major uh shift in that even um in the education space, like early on, you know, in my data science education journey, know, ethics was maybe one chapter in one
• 3:37
  course.
• 3:37
  It really wasn't covered much.
• 3:39
  I'm now seeing many more programs, you know, people that I work with and that I have the pleasure of interacting with.
• 3:49
  have the opportunity to go much deeper and learn much more about the ethical aspects of technology and there's full-fledged programs around it.
• 3:59
  I think it's also, you know, in the workplace, in the workforce, it's now we have a growing number of AI literacy and educational programs and efforts that help
• 4:17
  people actually in the development jobs understand the potential risks and how to mitigate them.
• 4:23
  is still though, I would say a challenge because there is, we're still pretty early, especially in the AI space, right?
• 4:34
  This particular technology of adoption and putting it into full fudge production offerings and deployments.
• 4:45
  And so,
• 4:47
  that because it's like, okay, we have to get this AI off the ground, this AI system, we have to make it effective.
• 4:54
  ah We have to make sure it's accurate, it's reliable, it does what it's supposed to do, which is all by the way, very important from a responsible AI governance standpoint.
• 5:04
  But there's additional factors as well that need to be tested.
• 5:08
  But understanding like, you know, the 70 plus bias metrics out there is still a specialty area.
• 5:14
  The good news is that we are seeing also the, I would say, wide expansion of inbuilt methods for testing and tools for testing within the platforms that developers and data
• 5:32
  scientists are actually using, which is, I think, enabling them in a way that they were not enabled in the past.
• 5:43
  So if you know how to turn on your content safety filters and you know what to look for and you use your governance platform for running the test and the APIs and things like
• 5:55
  that, that all is relatively new and I think um pushing us forward quite a bit.
• 6:04
  One of the challenges in this area is that uh even though they're, as you point out, dealing with the same concerns, the engineers and the technical people tend to speak a
• 6:14
  different language than the lawyers and the compliance people who are both involved in this area.
• 6:20
  So as someone who's really sat on both sides of that, how do we effectively get those conversations to happen where they can realize where they're addressing the same?
• 6:31
  Hmm.
• 6:33
  Yeah.
• 6:33
  So I think we do need cross functional groups and it's one of the early ah
• 6:42
  challenges that I think when people were looking at AI ethics, they understood, right?
• 6:46
  This is a socio-technical system.
• 6:49
  There are multiple aspects.
• 6:51
  We're going to need cross-functional teams to address them.
• 6:54
  um so part of the way we see that in practice is, of course, we want, um you know, within teams, a certain amount of cross-functional expertise and experience, but also
• 7:10
  in the management system that you have around it and how you actually govern at an organizational level.
• 7:16
  ah Like I chair our AI ethics committee at HCL Tech and we have, ah you know, in many other organizations, boards and committees that help bring together a number of different
• 7:30
  groups and weigh in and collaborate.
• 7:34
  Now, uh I think those sort of forums are incredibly important
• 7:39
  And then that's also at a policy level and an organizational level where you can uh influence and get people to work together and form as many of those leadership level
• 7:52
  connects that you need when it comes down to a given model or system, having the engineers and the legal people talk to each other.
• 8:03
  So you build those leadership connects, open the doors, um
• 8:09
  You can also enable, uh for example, even during an assessment process that you have specific points of contact from legal and from the data scientist team that are
• 8:22
  responsible for that given um product line or that given offering that you have.
• 8:29
  So it's combination of, I'd say, lower level connects that you need to be happening.
• 8:36
  within the organization, the front line, and then also the um leadership connects.
• 8:45
  As you said, you spent a number of years at IBM uh doing some of the early foundational work in this area.
• 8:51
  It's probably an unfair question, but what are some of the things that you took away from that experience that you brought into your current role?
• 9:00
  Yeah, it was a fantastic experience for me.
• 9:04
  I really enjoyed the energy and the collaboration um that we had during the time and the excitement around a new area.
• 9:16
  um so part of what I took away was just an excitement and a love of the uh topic and the challenges that come.
• 9:26
  And it's actually, it's something that I see and as I've
• 9:30
  hired my team and helped build my team, the people that get excited about the challenges and want to help find solutions, this is a great thing to bring into this type of role
• 9:46
  because so many of the challenges are new and also we continue to see more new ones all the time.
• 9:54
  And so I think
• 9:58
  how you address this is a um key takeaway for me.
• 10:02
  So a key part of being successful and effective in this type of field.
• 10:08
  um And then I think also, um there were so many things that uh I learned from the researchers that we worked with, so incredible IBM research folks and um the many diverse
• 10:26
  teams that we came across.
• 10:28
  So I loved, you know, working in this global organization.
• 10:31
  So obviously now working in another global organization and understanding how to connect these diverse groups and how they can work together.
• 10:40
  mean, every company is different.
• 10:42
  So that's an interesting thing.
• 10:43
  But there are some commonalities in terms of how you can bring uh team members together and what types of roles ah need to be at the table and connected.
• 10:55
  So it's m
• 10:57
  It's definitely a lot of lessons learned that I've been able to take forward.
• 11:01
  Great.
• 11:02
  I want to get onto HAL Tech, but I want to just follow up on one thing you said, where you talked about people being excited about doing this work.
• 11:10
  I think some people on the outside would think, responsible AI, it's about saying no, or it's about saying there's a risk this could go wrong.
• 11:17
  um So what is it that excites people and excites you?
• 11:21
  Yeah, so I think it's uh meaningful work and these are big issues, important issues that need to be solved.
• 11:32
  And so I think it gives you that sense of, you know, really tackling something that's important.
• 11:40
  And it's exciting to see also, I think what's exciting about it is that often, at least for me,
• 11:47
  The technology side of this also that keeps shifting and enabling better outcomes.
• 11:53
  mean, the rate at which the technology is evolving is amazing.
• 11:57
  uh And also the research and the solutions to help solve that are also rapidly um growing and changing all the time.
• 12:06
  And so I love, you know, um like when I was helping to roll out the first AI impact assessments at IBM and it was
• 12:17
  so much more manual and now we have all these amazing tools that can support that and enable the process in ways that it's like, wow, just a couple, you know, a few years ago,
• 12:27
  this was a much harder challenge to solve.
• 12:31
  uh so when you see that continued progress, that's also exciting.
• 12:37
  Yeah.
• 12:39
  Even though the technology is advancing super fast, you think the response mechanisms are advancing even faster.
• 12:46
  I do, I do.
• 12:47
  Yeah, because I think, you know, it's a...
• 12:55
  something actually probably, you know, obviously I haven't worked in a lot of other technology areas, but I understand that this is a unique aspect in that the widespread
• 13:06
  awareness and social pressure and awareness around the potential risks is driving more focus than in other previous, you know, development of technology.
• 13:19
  So it's pretty exciting.
• 13:22
  Yeah.
• 13:24
  Yeah, look, I'm excited too.
• 13:25
  I'm just playing devil's advocate and asking these questions.
• 13:27
  But I think it's a really critical question how fast things are moving.
• 13:32
  again, I would tend to agree with you.
• 13:34
  And certainly the point you're making is important in terms of how good the tooling has gotten.
• 13:38
  But I hear a lot this assumption that it's all going so fast that we can't possibly keep up.
• 13:45
  Yeah, yeah, that's a concern I hear often as well.
• 13:50
  And I think what fuels that in particular is that all the time people see news, like news stories, and then, you know, the tech company will respond with, well, we didn't know that
• 14:05
  that was there.
• 14:06
  But then you also see if you check like a few days later, sometimes like they fixed it already.
• 14:12
  It's incredible.
• 14:14
  You know, and so.
• 14:15
  um
• 14:18
  It is, you know, a challenge we have to, you know, that's certainly it's not that it's not a concern because, you know, we don't want ideally and it's also hard to know how many of
• 14:31
  those things actually got fixed, like how many things we not see in the news, right?
• 14:35
  Because I think and no, right, so much more is getting fixed before it even gets out.
• 14:42
  Now, I think but what does get out scares people.
• 14:44
  I think it's just the scale.
• 14:47
  at which these systems are being rolled out, um that we start to see more of these um situations hit the news.
• 14:55
  And it's also a headline, right?
• 14:59
  So it's getting a lot of attention um these days.
• 15:04
  Tell us a little about HCL Tech.
• 15:06
  It's a very big global company, but not everyone, especially outside of tech, knows what it does.
• 15:12
  uh And what are some of the particular issues that you have to contend with on doing responsible AI and governance there?
• 15:21
  Yeah, absolutely.
• 15:22
  So, HCL Tech is a global technology company with more than 220,000 employees operating in 60 plus countries.
• 15:32
  And we have clients all over the world and work on many different aspects of technology, um including, you know, managed services and how we help clients actually deploy.
• 15:44
  You know, we're an uh integrated service provider, we can help to
• 15:50
  um manage very difficult technology challenges that many other companies value that we can do.
• 16:00
  This is a very valuable service as well as we have AI capabilities m and platforms that we um put out and help develop.
• 16:11
  So we develop ourselves.
• 16:12
  We also help clients develop AI systems.
• 16:14
  yeah, um what's kind of unique from
• 16:19
  maybe other companies is that a lot of it is it's very client oriented work.
• 16:26
  So from an AI governance standpoint, right, it's ah much more about the client uh project, right.
• 16:35
  So we're dealing with a lot of individual client projects, ah which all have very unique aspects, right.
• 16:42
  It's a little different, I guess, if you were in an organization that ah developed one product or, you know, certain type of
• 16:49
  product set and you have to manage the AI governance risks around that.
• 16:54
  We're dealing with uh many, many clients all over the world in different projects and um that does present uh unique challenges.
• 17:04
  um Luckily, the company has very, very good enterprise risk management support already, right?
• 17:13
  So there's a lot of ingrained.
• 17:15
  Also quality control, because basically the history of the company is engineering, right?
• 17:21
  So very engineering uh heavy uh background.
• 17:26
  so especially on the technology aspects, that tends to be very easy for the uh teams that I'm working with to understand.
• 17:38
  that even more so than some of the other things is actually the technology solutions, which is very cool.
• 17:45
  What is the fact that you have all these different client engagements mean in practice for the work you're doing on AI governance?
• 17:54
  Yeah, so it means that we have to be sensitive to the fact that, for example, if a client um has a preference for a given environment, so not every client is going to want to adopt
• 18:07
  um governance controls that you've used in another um client that may be using a different cloud provider and a different base platform, they want to operate in their environment.
• 18:22
  They also have their own
• 18:24
  tend to have their own AI risk management policies and governance policies they want you to comply with.
• 18:31
  So we have our own for our company, our own standards and our own policy that we follow just across the board, but then also have to manage and consider what the client wants.
• 18:45
  And so it's a combination.
• 18:46
  You're often working with two sets of policies that you're working to follow.
• 18:55
  And how much variation is there?
• 18:57
  Because every company now has a set of AI principles, and everyone understands this is a risk.
• 19:03
  So how much does that really matter?
• 19:05
  So typically it's not that much of a difference in terms of the high level controls and expectations.
• 19:15
  I think that there can be differences, especially based on jurisdictional focus.
• 19:21
  So if it's a client in Europe, they may be very concerned about particular laws like the UAI Act.
• 19:28
  then in other cases, if we're in Canada or wherever, they have their own specific
• 19:36
  uh jurisdictional differences, I'd say, and what they put the most focus on, which might mean that you're going to apply more controls, more ah or even specific controls for that
• 19:48
  given jurisdiction.
• 19:49
  um And so that can be something that is certainly possible and does happen.
• 19:58
  What um did you feel was the most important starting point for putting together an AI governance program?
• 20:06
  So I think for me coming in new to the company, I really wanted to understand ah the need and the gap from where we were.
• 20:17
  What do we need to address?
• 20:18
  ah What things were already being handled very well and what things could we do better ah immediately?
• 20:29
  And also to make sure that, ah for me personally, to build the relationships that
• 20:37
  I knew would be important from the beginning, right?
• 20:41
  So a lot of this when you're operating in a large company, you have to form a lot of partners.
• 20:48
  It's all, you know, nothing happens in a silo.
• 20:51
  ah I think even in small companies, but you know, especially in larger uh organizations in my experience, it's just, um you know, understanding and having those relationships so
• 21:03
  that you can be effective.
• 21:04
  um
• 21:05
  And certainly we don't do this alone, this type of role and work alone.
• 21:13
  There's a lot to do.
• 21:15
  So for me, a lot of it was making sure that the AI governance program was appropriate and effective um in the context of the organization.
• 21:30
  so understanding the organizational needs and the people.
• 21:35
  And then from there, right, because it's a process then also of, you know, prioritization and some of it's also just what's possible, right?
• 21:47
  So you can do things sort of in any given order.
• 21:53
  Obviously, one of the first things we did was set up an AI ethics committee and, you know, really make sure that we started educating people a lot on
• 22:05
  the AI governance um policy that we have and how do we put in place those controls.
• 22:11
  So lot of it is um some of the things that you would expect anywhere, but just making sure also that it fits for the organization.
• 22:23
  So a lot of AI governance work on the ground is how do I fit it into an existing process?
• 22:30
  So there's an existing process, how do I insert um
• 22:34
  the AI governance requirements, you know, in our procurement process and our, you know, what we do with clients and the actual life cycle of that client interaction, right?
• 22:46
  So a lot of it is process-based understanding.
• 22:50
  And then also, I think from a technology standpoint, understanding how the development teams work and at what point do they need in their process, a control or a test run.
• 23:03
  And so it's a lot of the details around that stuff.
• 23:11
  How do you answer those questions?
• 23:14
  ah Well, it's not an easy question.
• 23:18
  Those questions are not easy.
• 23:19
  They're basically um context specific, which is often what we hear in the case of AI, right?
• 23:26
  So it's uh what is the use case?
• 23:28
  What is the context?
• 23:29
  And so I think for it's the same way like deciding on the governance controls.
• 23:35
  um We have um quite a bit of, um you know,
• 23:44
  if the um control itself needs to be um adapted or...
• 23:53
  So given the type of work we're doing with the client, for example, could be something that we need to think about.
• 24:01
  um And also where we have client teams that are working on particularly high risk industries, um making sure that we have...
• 24:11
  um
• 24:12
  clarity because there's often an intersection between existing laws and highly high risk, highly regulated areas and the AI governance laws.
• 24:20
  So what is that intersection point?
• 24:22
  Like we've done a lot of work like with our life sciences and healthcare groups.
• 24:27
  you know, that becomes a question of two.
• 24:30
  mean, this, this is a control you're inserting that is also in the context of many, many others.
• 24:39
  You mentioned earlier that the tools and automation capabilities for responsible AI have advanced a tremendous amount over the past decade or so.
• 24:51
  What are the things that you uh most want that you don't have in terms of capabilities or tools to do the work you do?
• 25:00
  So I think that we have a lot to do on agentic AI still.
• 25:08
  I think this is just because the technology is so new.
• 25:10
  think that I've seen a lot of really good research and a lot of good early um product capabilities around this.
• 25:19
  I think we still have quite a bit more that we need to do and we'll see in the coming years.
• 25:27
  And that is the technology too.
• 25:30
  again, is something that we are seeing picked up very rapidly.
• 25:36
  But I am seeing a lot of movement in the research and development of uh governance controls for agents.
• 25:45
  think also um
• 25:50
  some of it is still in the process.
• 25:52
  What's really interesting about AI governance itself as a function and as a profession and the area where people work, it can fit organizationally in many different parts of uh a
• 26:06
  company or organization.
• 26:11
  depending on where they sit, whether that's a uh privacy office or a legal group or a technology office or a security office.
• 26:20
  All of that um presents differences in the operationalization of the process and also, I think, the potential for automation.
• 26:33
  um And also, I think that it's still a challenge with the number of different...
• 26:42
  We talked about the importance of cross-functional teams, but um enabling...
• 26:49
  good quality data sharing and um integration, right, across those groups, still I think is a challenge.
• 26:59
  And part of that is honestly just the human aspect and um the risks of data sharing and merging and that sort of thing, right?
• 27:10
  So there's a lot of it is maybe not even the technology itself.
• 27:15
  It's just, if you can't get access to the data,
• 27:18
  from that group, you may not uh have it as readily available um for AI governance type of work.
• 27:30
  Where does your group sit within the organization?
• 27:32
  So we're part of the chief technology office.
• 27:35
  So this is a technology group.
• 27:38
  Yeah.
• 27:39
  So it's pretty common, I think, for organizations to um have a technology executive that is responsible for this.
• 27:53
  Or it can be a risk and compliance or privacy type of executive.
• 27:59
  And that's where it was for me at IBM was more in the legal and privacy area.
• 28:06
  And yeah, it's a very different um skill set that's brought to the team.
• 28:16
  And um there's pros and cons uh both ways for sure.
• 28:21
  absolutely.
• 28:21
  But I'm curious, in your experience, what does it mean to be doing this work within that technology organization as opposed to more on the privacy or legal side?
• 28:32
  Yeah, so I mean, you know, it's, it's, uh I think we all bring our lens, right, our lens of the world.
• 28:38
  And so technologists tend to say, how can I solve this with technology?
• 28:43
  How can I measure it?
• 28:46
  What measures?
• 28:47
  What metrics do you have?
• 28:49
  What methods?
• 28:49
  What tools can I use?
• 28:51
  Whereas, um you know, somebody in a area like law or compliance or privacy or risk,
• 29:01
  is asking questions about what are the risks?
• 29:05
  How do I measure?
• 29:06
  How do I comply with the law?
• 29:07
  What is the law?
• 29:10
  What is the specific clause?
• 29:12
  What is the definition?
• 29:14
  And so they're asking questions that often framed a little different way.
• 29:20
  think um what I've seen is that they're both critical.
• 29:27
  So when I was operating in like more of the
• 29:31
  privacy, legal risk type of area, it was always, where is the technology group so that we can get our technology questions answered?
• 29:39
  And same thing when I'm operating in the technology area, it's always like, okay, but we have to comply with this law lawyer, please help us with those definitions and the
• 29:51
  understanding of it, right?
• 29:51
  So um you need both.
• 29:55
  And um I don't think that it necessarily matters where you place
• 30:01
  the group as long as there can be effective collaboration and support between them.
• 30:08
  You mentioned a little bit ago agentic AI, and it's obviously something that a lot of people are talking about.
• 30:16
  What's really different from a governance perspective about agentic systems?
• 30:22
  Is there something that really makes it harder, or is it just asking the same questions for agents that you would be asking for other kinds of systems?
• 30:30
  Yeah, so it's, you know, it is an AI system, clearly.
• 30:34
  So we ask um many of the same foundational questions that we do about any AI system.
• 30:39
  So that's very clear, and that's true.
• 30:43
  uh However, there are some unique challenges, um but like AI systems, it is really risk-based.
• 30:51
  So depending on the situation which you're deploying,
• 30:56
  those AI agents, it's going to matter what kind of controls you have in place.
• 31:00
  But I think just the level of autonomy that the agent has is really driving that concern.
• 31:08
  So the ability to create compounding impact over time, operating independently, um without proper oversight can be um an issue.
• 31:20
  Now, the speed as well is also a concern.
• 31:23
  um
• 31:25
  When we used to think and talk about human oversight um of AI systems, that was much more feasible.
• 31:35
  In this case, with AI agents, we're starting to look to more automated controls and oversight mechanisms as well so that we can catch potential issues um at the speed at
• 31:49
  which the technology moves itself.
• 31:52
  And so having, you
• 31:54
  other agents that essentially govern agents underneath them is part of the picture.
• 32:01
  so uh that also, and I think also there's challenges around the um complexity of the environment and the number of connections um that they have.
• 32:17
  that root cause identification can sometimes be a challenge.
• 32:21
  So you really need good logging and tracking and understanding the source of potential issues.
• 32:29
  And so that is a key component.
• 32:31
  If you're operating in multiple environments and calling multiple APIs and taking many different actions, you'll need to have in your workflows for the agent, good controls and
• 32:47
  checkpoints as well as
• 32:49
  Like I said, maybe oversight agents and things to help control.
• 32:54
  And what else is on your roadmap or agenda in terms of issues that you're focused on for the next year or two?
• 33:03
  Yeah, so a lot of what I'm hearing and seeing from clients is a...
• 33:15
  And this also in my own world, in my own experience.
• 33:18
  I think just generally in the industry, there's a continued um interest in adopting AI.
• 33:27
  So it's how do we um get the most value and have the most benefit come from our adoption of AI?
• 33:37
  Everybody understands um the importance of that.
• 33:42
  And then um
• 33:45
  what does that mean in practice?
• 33:47
  But I think we're moving from a period that was, ah so we talked a lot about AI ethics and principles, the 2018 sort of timeframe.
• 33:57
  ah And then it was all about moving from principles to practice.
• 34:03
  So how do we put that into practice?
• 34:04
  And I still think large majority, and there's lots of studies to um highlight the fact that we're still in this
• 34:13
  phase of getting it into practice.
• 34:16
  There's no doubt.
• 34:17
  did actually a released a study earlier this year with MIT on the uh gap that uh we see.
• 34:26
  So like 87 % of executives feel that responsible AI and AI governance practices are important, but only 15 % of them really felt ready to implement the necessary controls and
• 34:40
  capable of managing the risks.
• 34:43
  huge gap in terms of awareness and recognition of the issue and then being ready to actually do it.
• 34:49
  I still, especially for practitioners like myself, we are still very much in this, you know, um moving from principles to practice and continuing to stay abreast of the changes
• 35:02
  that continue to happen in the technology and the new challenges all the time.
• 35:06
  So that is, I would say we're still there, but then
• 35:10
  There is a bit of a shift, I'd say, that's happening more around, you know, how do we actually move from um practice and sort of self-regulation, right?
• 35:23
  We didn't have many laws and frameworks.
• 35:27
  Now there's many more frameworks out there, you know, the NIST AI Risk Management Framework and the ISO 40-2001 standard, and all sorts of good frameworks and standards out
• 35:37
  there that we can look to.
• 35:39
  now.
• 35:39
  um And also we're starting to see more, obviously laws being passed all the time.
• 35:49
  And so it's starting, I'd say, looking at like the proof points.
• 35:54
  So okay, you have it in practice.
• 35:56
  Now, how do we prove it?
• 35:58
  What, you know, maybe there's audits happening, uh you know, governments are asking and lawsuits and all sorts of things, right?
• 36:08
  That it's
• 36:09
  um Less about like, so we sort of moved from ideas and principles to how do we put this into practice?
• 36:15
  Let's figure out some ways to now show me that you have it, right?
• 36:21
  Show me, show me.
• 36:24
  Great.
• 36:24
  Well, so that's a great place to end.
• 36:27
  And it's still a long way to go, but we've made a lot of progress.
• 36:32
  So Heather, thank you so much.
• 36:33
  been a pleasure speaking with you.
• 36:34
  Absolutely likewise.