Latest / CXOTalk / AI Ethics, Governance, Responsible AI: Views from the CIA's Deputy Privacy Officer | #863
Transcript
- 0:00Welcome to CXO Talk Episode 863. I'm Michael Krigsman, and today
- 0:08we're discussing AI governance and ethics.
- 0:12Our guest is Jacqueline Acker, who is a Deputy Privacy and
- 0:16Civil Liberties Officer with the CIA.
- 0:20She will share her personal perspectives on these topics.
- 0:24I think we all know the CIA, the Central Intelligence Agency, but
- 0:30we're on the outside. So tell us about the CIA and
- 0:33tell us about your role and and what you do.
- 0:35We work on issues ranging from running a traditional privacy
- 0:39program, like every federal agency has, to implementing
- 0:43statutory requirements. Some of those include making
- 0:46sure that the public has the ability to submit complaints
- 0:50that their rights have been violated.
- 0:52We're protecting Americans information under executive
- 0:55orders as well as developing AI governance, that is with your
- 1:01the intelligence community, but also with the wider federal
- 1:04government requirements. And we also support CI as chief
- 1:08transparency officer. So the goal there is to make
- 1:11sure that the public understands our mission, our authorities and
- 1:15our oversight. And then we also work with
- 1:17executive branch oversight and congressional oversight and
- 1:21publish reports for the public. So that's just a piece of what
- 1:23we do. There's so much more to it and
- 1:26I'm so proud of my team. Some of the other work that I do
- 1:29includes being a faculty member at the International Association
- 1:33of Privacy Professionals, which is actually been re rebranded
- 1:37recently as IPP because the scope has expanded.
- 1:40As you've probably heard from many of your guests, your
- 1:43privacy work is getting so much more integral into other aspects
- 1:49such as cybersecurity and of course even AI governance.
- 1:51So they're trying to tackle a lot of those intersecting
- 1:54issues. And then I just wrapped up my
- 1:566th semester teaching at American University for their
- 2:00Washington College of Law, their Master of Legal Studies program.
- 2:03So I was really happy to to be able to do that and also just
- 2:07like I was very proud of all of our students.
- 2:09We all like the spy shows about the CIA and conspiracies.
- 2:15Do you folks in the CIA watch these shows and what do you
- 2:17think about these shows? Well, I can't speak for
- 2:20everybody, the whole CIA, but I am so and I, you know, I don't,
- 2:24I haven't seen every show there is out there, but there's a few
- 2:27that I follow pretty closely. I've really enjoyed the new Jack
- 2:31Ryan series with John Krasinski. I love the Americans.
- 2:35I watched every episode of that. And then I'm kind of followed
- 2:38Kerry Russell on her journey through the Diplomat.
- 2:40But I think lately my favorite has just been slow horses
- 2:43because I it just cannot get any better than Gary Oldman.
- 2:47Long time favorite of mine so. So when you watch these shows,
- 2:51do you say to yourself, oh, that's a close reflection of
- 2:56what we really do in the CIA, or do you say that is total science
- 3:01fiction fantasy? Like how do you relate to these
- 3:04things? I would say more grains of truth
- 3:06in them than others. Like I'm not running around, you
- 3:10know, in foreign countries. Like, I definitely sit at a desk
- 3:13most of the day. So for me, it's always just a
- 3:16different, a different perspective maybe.
- 3:20But I think it is amazing seeing how Hollywood has always
- 3:24portrayed a lot of what we do and and same thing with like
- 3:28our, you know, the counterparts in Britain and some of the
- 3:30amazing, you know, movies, James Bonds and things like that.
- 3:34I personally don't see myself as one of those, but I do enjoy
- 3:38watching them, especially when you're every once in a while
- 3:41there is something where you're like, oh, that has a great of
- 3:43truth in it. And for me, those those moments
- 3:47are usually like how amazing the Co workers are and how much they
- 3:50take care of each other and things like that.
- 3:52That there's kind of a say that as CIA we're family.
- 3:56And I think that's the part that I like seeing in these Hollywood
- 3:59portrayals. So that sense of taking care of
- 4:03each other that resonates with you as being real.
- 4:07That part, yeah, absolutely. I've been at CIA for over 11
- 4:11years now. And I think that's, you know, I,
- 4:13when I started there, I didn't know how long I'd be there.
- 4:15I was like, oh, sounds like a cool job.
- 4:18I had a friend who was like, hey, we're hiring lawyers, you
- 4:21should come over. And, and I was like, yeah, of
- 4:25course I'm going to go do that. That sounds so cool.
- 4:27And then I got there and I was like, well, we'll see how it
- 4:29goes. You know, I'm, you know, fresh
- 4:30out of law school trying to figure out, you know, how I want
- 4:33to spend my career. And I just came from a really
- 4:36wonderful firm out in Oklahoma. So it was kind of a big leap for
- 4:40me to try something so drastically different.
- 4:42But the people are are why I've stayed and the mission that I've
- 4:45been able to to work on as well. Why is AI ethics important to
- 4:51the CIA? For us, what's special is it's
- 4:55about the trust of the American people.
- 4:57So if we don't continue to use AI responsibly and ethically,
- 5:01then we could lose that trust and lose the trust of our
- 5:05partners around the globe, around the country and industry
- 5:09and things like that. So at the, you know, at the very
- 5:11bottom line, it's really all about trust.
- 5:14Trust in the agency and the the role of AI ethics is ultimately
- 5:21to maintain that trust. Trust is a lot of things, but
- 5:25you're part of it is getting it right.
- 5:27Part of it is making sure we're protecting those American
- 5:30values, your privacy, civil liberties, things like that.
- 5:33And when we don't, you know, if we don't maintain that trust,
- 5:37then the consequences could include things like losing our
- 5:40authorities, our authorities to collect data and things like
- 5:43that. And that would make our mission
- 5:45of protecting the public that much harder.
- 5:47So your trust is it's almost a fundamental part of what we have
- 5:52to do. When you think about AI ethics,
- 5:57can you describe what that encompasses for you?
- 6:01I think you that can change from organization to organization,
- 6:06from person to person. So in addition to trust, I think
- 6:10it's also making sure that you others understand what you're
- 6:14doing so that you is part of transparency and making sure
- 6:19people understand that your product is trustworthy, making
- 6:22sure they understand you, how it was created, what were the rules
- 6:26that you were following when you made it.
- 6:28And you know, for the people I talked to in industry, you know,
- 6:32that can also mean you explaining that to a consumer,
- 6:35explaining that to your insurance carrier even.
- 6:39So you being able to make those guarantees, whether it's for
- 6:43sales or to, you know, government oversight.
- 6:46I think the transparency piece is a big part of the ethics.
- 6:50It's also, I think there's so much more to, to ethics and
- 6:55creating responsible AI. But I would say I'm having a
- 6:59good program in place, having processes, having, you know,
- 7:02some sort of governance and oversight mechanism, and those
- 7:06are all really, really key pieces to it.
- 7:08So in terms of having a a program in place, can you
- 7:13describe what are the components of an effective AI ethics
- 7:19initiative? It really does depend on the
- 7:22organization. So you know, factors could
- 7:24include, you know, what context you're working in.
- 7:27Are you in healthcare? Are you in education?
- 7:29Where are you located? Are you in the EU, for example,
- 7:33your organization's opinion, what do they value?
- 7:38And in all of those things play into building your program
- 7:42because you have to really define first like what do does
- 7:46your organization mean by AI ethics?
- 7:49So for example, in the intelligence community, starting
- 7:52in about 2019, we wrote the principles for intelligence
- 7:57transparency. And those principles were based
- 7:59on your general principles for ethical intelligence that had
- 8:04been around for a while, but we use that and lifted it and
- 8:07shifted it for artificial intelligence.
- 8:09So we built on something that already existed, this kind of
- 8:12ethos that we already had. And then we said, OK, well,
- 8:15that's great, but how do you actually do that?
- 8:18And so we worked really hard with data scientists, lawyers
- 8:22and people from at the time. I think there's 17 different
- 8:26intelligence community agencies. I don't think Space Force was
- 8:29was there yet, but we worked with people across the entire
- 8:32community to create a framework that is basically a set of
- 8:37questions meant to work with any agencies, authorities and be
- 8:42very flexible even with what kind of technology it's about so
- 8:45that everybody has a common script to go through for.
- 8:49OK, well, how do we make sure that we're upholding these
- 8:51values? Like what do we mean by, you
- 8:54know, what is ethics, what to what are our data requirements
- 8:57and things like that? So that was a really key piece
- 9:01for us in the intelligence community.
- 9:02It got signed in 20, 20 in June, which was kind of a crazy time.
- 9:08But it was really important that even in the midst of all of the
- 9:12big changes, we were setting that foundation.
- 9:15Because having a governance framework out early and also
- 9:19having one that it's, it was created by so many people and,
- 9:23and therefore can be adopted writ large was really key to us
- 9:27because we, we knew we were going to have to use this
- 9:29technology. And if you set, you know, some
- 9:32minimum guard rails, you can go faster, right?
- 9:36So like cars originally you didn't have brakes and so they
- 9:39couldn't go very fast, but once they had brakes, they could go
- 9:42faster. And for us, a lot of that's also
- 9:45still about that trust and, you know, making sure things were
- 9:47safe and everything. An organization like the CIA,
- 9:52when you say that it's signed or or adopted, where does that
- 9:58adoption or authority to adopt sit inside the organization?
- 10:04For the ethics framework that was signed by the then Director
- 10:08of National Intelligence. So that was at the Director of
- 10:13National Intelligence is part of the DNI and they kind of their I
- 10:18would say like the integrator across all of the IC.
- 10:21So for IC wide policy that's really the appropriate place
- 10:24there. That's not going to be the same
- 10:26at, you know, a lot of different, your organization's
- 10:30maybe large companies have subsidiaries and things like
- 10:32that. But but for governance and for
- 10:35ethics, even like within the federal community, there's a lot
- 10:38of different ways to do it. You can have that be more
- 10:41centralized or you can have it be decentralized.
- 10:44You could have kind of hybrid approaches where there's one
- 10:48group that creates the, you know, the overall policy and
- 10:51then the people who actually carry it out are maybe
- 10:54integrated into different, you know, parts of your organization
- 10:57or, or agency. So you've this framework, we had
- 11:01to write it to be able to work in all of this different
- 11:04environment, which was a, a really fun challenge.
- 11:06But I think, you know, we're, we're here, you know, 4 1/2
- 11:09years later and it still works really well.
- 11:11So very proud of that. Please subscribe to our
- 11:15newsletter. Check out cxotalk.com.
- 11:19We have excellent shows coming up.
- 11:22Subscribe to our YouTube channel.
- 11:24We have an interesting question that has come in from Twitter,
- 11:28and this is from Arsalan Khan, who's a regular listener.
- 11:32He asks great questions and he says who defines the ethical
- 11:37boundaries? Who should or shouldn't define
- 11:42these ethical boundaries? At every organization that's
- 11:46going to be different and I, you know, I talked earlier about how
- 11:49the organization's culture matters, but also your things
- 11:52like what about you? Where are you located?
- 11:55So geography, for example, I think some ethical boundaries
- 11:59are just built into the laws where you are.
- 12:02So we've seen a lot of that in the United States with different
- 12:05state level rules on AI. Of course, you're with the EU,
- 12:10there's the EUAI Act. So you have it at your AI
- 12:14specifically at those kind of levels.
- 12:16But I would say even, you know, even more granularly you there
- 12:20are rules that you maintain data requirements, right?
- 12:25So you're a lot of AI. I would say ethics is actually
- 12:29just data ethics. You know, what data are you
- 12:32allowed to have? You know, how long are you
- 12:34allowed to have it? What protections does it have to
- 12:36have? So some of that is just your
- 12:39baseline legal requirements. Some organizations, you go a
- 12:43little further, they take things like making sure that you have
- 12:46the minimum amount of data necessary to complete your job
- 12:50very seriously. And others are kind of like we
- 12:53need all the data. So if that organizational
- 12:56culture is really a critical a piece of that on top of, you
- 12:59know, what is your baseline, like what can you do under the
- 13:02law? What do you have to do?
- 13:05I think the more fun area to be is in that, you know, in
- 13:08between. So you know, what should we be
- 13:10doing? What you know, what are ways
- 13:12where we can be ethical but also meet our mission.
- 13:15So that is a harder thing to figure out who decides that.
- 13:21And again, you know, for your organization, it might be your
- 13:24shareholders, it might be your CEO, it might be somebody else.
- 13:29We're seeing a lot of organizations start to have
- 13:31chief ethics officers, even chief AI ethics officer.
- 13:35So I think like there's still kind of some norms forming
- 13:39around that in the AI space, but I don't think of it as being
- 13:42really all that different from where we were, you know,
- 13:45previously where we already had these data issues, ethics issues
- 13:50about technology, you know, who should have access to it and all
- 13:54those things. So, you know, AII think the big
- 13:57changes obviously been with generative AI and kind of the
- 13:59focus on that. And so there's been a bigger
- 14:03shift towards formalizing some of those choices on who is a
- 14:08decision maker. But you know, even one person in
- 14:11an organization should be working with stakeholders across
- 14:14that organization to, to try to make those chances and, and
- 14:17paying attention to, to just like what's happening in the
- 14:20world, you know, what are, where are ethics norms forming, you
- 14:26know, around wherever your organization is or you know,
- 14:30maybe where your clients are, things like that.
- 14:33Are there unique aspects of this that are specific to
- 14:39intelligence organizations that might be different from what
- 14:43we're used to in the private sector?
- 14:46There's some pieces that are unique.
- 14:48I would actually say there's some pieces that you know, could
- 14:50be borrowed by the private sector.
- 14:53So the intelligence community, you were part of the federal
- 14:57government and for over 50 years we have been doing privacy
- 15:02thanks to the Privacy Act of 1974.
- 15:05So there's, there's a lot of requirements in that, but a lot
- 15:09of a lot of the basic requirements include making sure
- 15:12that, you know, information about U.S. citizens and lawful
- 15:15permanent residents is protected appropriately.
- 15:18And what does that mean? So it means in certain systems,
- 15:21if you're able to retrieve information by a unique
- 15:24identifier, you have to have transparency to the public about
- 15:28you, what that system is, why does it exist, what protections
- 15:32are on that? And you have to publish that on
- 15:34your federal agency's website so that the public can go request
- 15:37records about themselves. And they basically have this
- 15:40right to access their records and then amend them if they're
- 15:43wrong. So there's been a big check on
- 15:45the federal government for 50 years.
- 15:48And that requires a lot of of work to make sure that your data
- 15:51is protected appropriately. And there's a lot of
- 15:54accountability there. So I think there's a good chunk
- 15:58of that piece that, you know, companies can learn from how you
- 16:02can go to any federal agency website and see examples of
- 16:05that. Department of Justice, of
- 16:07course, has, has some amazing and resources as well.
- 16:10But even things like a privacy impact assessment, you know, a
- 16:14lot of organizations are looking at the federal government's
- 16:17privacy impact assessments and say, oh, we can shift this into
- 16:20an AI impact assessment. It meets a lot of the standards
- 16:23or, or they're using those already for their internal
- 16:26privacy programs. And then just like building on
- 16:28that. So yeah, the there's that piece
- 16:32that's kind of special to the federal government that I think
- 16:34can be borrowed from. But the other pieces that are, I
- 16:37think challenging are in the intelligence community.
- 16:41We can't always be as transparent as we would like
- 16:44because we have to keep sources and methods safe.
- 16:48So we can't always share the ways that we got information or
- 16:52who we got it from and and things like that or like the
- 16:55technology that we're using with the general public because then,
- 17:00you know, we wouldn't be able to continue using those.
- 17:02So instead we have this really robust system of governance and
- 17:07oversight with people who are cleared.
- 17:09So there's cleared members of Congress, there is the Privacy
- 17:12and Civil Liberties Oversight board, there's even my office.
- 17:15So we have our own internal oversight requirements.
- 17:18So that is a little bit different, but although I say
- 17:22it's different, I actually see a lot of that in private companies
- 17:25as well. You may want to have you your
- 17:28own kind of internal auditing or internal oversight so that
- 17:32you're not sharing trade secrets and things like that.
- 17:35So there's actually a lot of parallels, even though, you
- 17:38know, I think it, I'd like to say that, you know, it's all
- 17:42very unique and very special and we're the only people who could
- 17:44do this. But, but at the end of the day,
- 17:47I I think that there's a lot of great ideas that the private
- 17:50sector can and has borrowed from those processes.
- 17:53And of course, what you're doing at the CIA has a lot of
- 17:59maturity. And so there are lessons I'm
- 18:02sure that the private sector can learn from that.
- 18:05Yeah, absolutely. I mean after 50 years of doing
- 18:08the Privacy Act, federal agencies really have some some
- 18:12great experience there and it's not always been perfect.
- 18:14There is, you know, definitely with any new rules and any new
- 18:17technology, there's challenges and there's growth.
- 18:20But I think that if I always think about this actually how
- 18:25like the Privacy Act is older than most tech companies by like
- 18:28quite a bit. So the maturity of the federal
- 18:31government is something that should definitely be lifted and
- 18:33shifted and, and used where it can be because the, the
- 18:39experiences are you in many ways universal.
- 18:43But you're right, the maturity level is, is certainly there and
- 18:46you can see that in the just in the documentation that exists
- 18:50and also just in how the transparency has worked.
- 18:53And there's also you've been this accountability, not just
- 18:56with the public, but you know, with Congress and and other
- 19:00oversight through that. So if for any organization that
- 19:04is, you know, going to potentially you have that kind
- 19:08of scrutiny, like they should absolutely be be looking at the
- 19:10mature programs from the federal government as a starting place.
- 19:14We have a couple of more questions that have come in.
- 19:17And interestingly, both of these questions now relate to
- 19:21international relationships and boundaries.
- 19:24So let's try these. And this first one is from Greg
- 19:28Walters on LinkedIn. And he says it's his contention
- 19:34that the AI revolution, including privacy, is being led
- 19:38by the US and that the US is the best hope for an ethical and
- 19:43free deployment and use of AI. I'm just reading his question
- 19:47and he says this. What?
- 19:50Or he asks this? What real threats to freedom and
- 19:54liberty can our adversaries implement with AI aside from
- 19:59drones? There's a values challenge
- 20:02internationally right now. And yeah, the US has definitely
- 20:06been trying to lead on this. We saw this even with the
- 20:09National Security Commission on AI report.
- 20:12It's like 700 something pages long.
- 20:14It came out a few years ago before, again, the generative AI
- 20:17boom. But that report talks a lot
- 20:19about how this is a values competition and how you there
- 20:24are so many different threats to values.
- 20:27And one of the things that I personally find interesting
- 20:30about all this is you the the need to protect American values.
- 20:35That's obviously part of what my office does, but the
- 20:39intersection of that need with also the need for the United
- 20:45States companies to be successful.
- 20:48I think there is a lot of there you can read 8 different news
- 20:52stations articles and get 8 different answers about like
- 20:56what part of that is important and who should win out And, and
- 20:59you know what that means for the international community.
- 21:02But I really like that. I think that report is something
- 21:05that still stands pretty well on explaining, you know, why some
- 21:10of those values systems could be, you know, I don't want to
- 21:14say like in peril, but what those competitions really are
- 21:18that international stage and a lot of the kind of predictions
- 21:22that it had, I think could have been playing out.
- 21:24So so if if you don't have 700 you know or time to read 700
- 21:28something pages, I'm sure you can find a an overview somewhere
- 21:33or just read the executive summary.
- 21:35It's very interesting to hear you talk about this values
- 21:42competition and then the relationship between that and
- 21:47the need to maintain an ethical stance.
- 21:51Is there is there a a tension or or conflict but between these
- 21:57two different goals? I think there's so much
- 22:00alignment here. I, I strongly believe that
- 22:04creating technology, using technology and doing that in
- 22:07alignment with your values actually helps, you know,
- 22:11because it helps maintain that trust.
- 22:14It helps create an ecosystem where you can work with, you
- 22:17know, partners, whether that's in the private sector or sector
- 22:20or overseas. And and that really helps bring
- 22:24these kind of like alliances together of, hey, we all have
- 22:27the similar value structure. So like, let's work together.
- 22:31I think that's a huge strength. I think that's a huge American
- 22:34strength, but you that is, that's part of why it's so
- 22:38important to get it right. And that report really does a
- 22:41fabulous job of, of kind of explaining, you know, all of
- 22:44those things. And, and you know, but
- 22:47personally, I, I really do believe that there's alignment
- 22:49there. I think that values are a big
- 22:51part of our strength. And we have another question
- 22:55again from Arsalan Khan relating to this international set of
- 22:59issues. And he says, when data is shared
- 23:03across organizations or countries, whose ethical
- 23:08boundaries should be followed? It's so important to have these
- 23:12frameworks, these international frameworks that basically say,
- 23:16hey, here's our shared set of values.
- 23:19And we've seen a lot of these, you know, OECD definitely has
- 23:23several. We're seeing things out of the
- 23:26UN and otherwise. But but yeah, I mean, I think
- 23:28that the, the best way is to have those standards and have
- 23:33people adopt them and say, like, yes, we're signing on to that
- 23:37because it shouldn't necessarily be like who, you know, whose
- 23:40standards are winning. It should be an agreement.
- 23:43And those agreements are important because it says, hey,
- 23:45like my, my values align with your values.
- 23:48Now we can work together. And we've seen that anywhere
- 23:52from people adopting this internationally to ISO
- 23:55standards. So see, I mean, that's that's
- 23:59kind of how I would approach that instead of thinking about
- 24:01it as like a who's who's winning scenario.
- 24:05What steps do intelligence organizations take to mitigate
- 24:11bias in AI systems for intelligence gatherings or
- 24:17threat assessments? Any part of just the data life
- 24:22cycle, not even getting to AI, you know, there's pieces that
- 24:26that are important. So we have strict authorities
- 24:29about what data we're allowed to get in the 1st place.
- 24:32So that's kind of one area where that, you know, bias could be
- 24:35introduced or it can be reduced. But typically, you know, what
- 24:39we're looking at is something that is a biased, you know, it's
- 24:43very biased towards what we're allowed to do, which is actually
- 24:45pretty narrow. So, so there is some bias there.
- 24:49But the one where you're looking at AI and developing AI, there's
- 24:53a million different questions that have to be asked because
- 24:56you you bias can actually be something that's helpful.
- 25:02So if you need to be able to find, you know, X, you know, X
- 25:08topic, you know, say like, I don't know widgets or something
- 25:13and you need to be able to find those widgets in pictures from
- 25:16anywhere, you kind of want an AI that's biased towards finding
- 25:20those widgets really well, so the first question I think you
- 25:23have to ask is what kind of bias are we concerned about?
- 25:27So there's bias that can be good and then there is a lot there's
- 25:30obviously bias that we really don't want.
- 25:32So bias that could harm, you know, that could harm
- 25:35individuals or harm civil liberties or privacy or, or
- 25:39maybe harm a specific group or not have equal application.
- 25:44But, but yeah, I mean, we have, our ICAI ethics framework has an
- 25:49entire section on mitigating undesired bias and has a set of
- 25:53questions there. And that's a, that's a really
- 25:55good starting place for AI. But also I mentioned earlier, it
- 25:59applies to a lot of technologies.
- 26:02And so anytime you're using a lot of data, I think those are
- 26:05great baseline questions to start from.
- 26:07And then of course, you know the NIST standards are are always
- 26:10fantastic. What about the internal
- 26:13processes when it comes to ensuring that that data
- 26:19algorithms don't have undesirable bias?
- 26:25Can you talk at all about how the intelligence community
- 26:28manages that? A couple of different things
- 26:31coming together. So organizations are different
- 26:33sizes. They need to be able to work
- 26:36within, you know, the existing programs that they already have.
- 26:40So some for some organizations, they're going to be working
- 26:44those issues through their existing like accreditation
- 26:47process or you're sometimes it's under a CIO's office.
- 26:51Some of that you could be woven into you with the data
- 26:55scientists who are working on the cybersecurity side.
- 26:58Sometimes they're the ones thinking through.
- 26:59OK, well what, what are, is this data set even complete?
- 27:03Like what do we really need this data set to be?
- 27:06And then how do we need to protect that data?
- 27:09How do we need to protect that algorithm?
- 27:10What are different potential injection points into that?
- 27:13So is sometimes it's kind of done by OK, well, we have the
- 27:16experts and they're already over here.
- 27:18So that's who's going to do it. But it really does depend based
- 27:22on, you know, I would say if you ask again, like if you ask 8
- 27:25different organizations, you're going to get eight different
- 27:27answers. But we are seeing more
- 27:31coalescing of some of those and some more professionalization
- 27:35and formalization and organizations, you know, around
- 27:39that. So you might have more of a
- 27:40process for, OK, and AI is going to be created, you know, we're
- 27:45going to document what the purpose is and you know, what
- 27:48the potential risks are, what the potential biases are, what
- 27:50the mitigations are, and then have kind of different people
- 27:54who are responsible for each piece of that, including the
- 27:57documentation piece. So it's getting, I think it's
- 28:00getting a little bit more formalized, but I would say it's
- 28:02still very much organization dependent.
- 28:05Arslan Khan comes back yet again.
- 28:08He's really on a roll, and on this point he says if a handful
- 28:13of countries develop AI ethical standards then that itself
- 28:18creates an A, creates a bias since other countries
- 28:23perspectives might not be included and is the UNA better
- 28:29place for AI standards to be developed for that reason.
- 28:34I honestly have not put much thought into that.
- 28:36So I would like to defer that one, but I will.
- 28:41I don't know. I, I think it's a really good
- 28:42question. It's one that I should have
- 28:44asked myself. So I'm going to be thinking
- 28:46about that probably all weekend. Last week on this show, we had
- 28:50as a guest the Assistant secretary General of the UN for
- 28:54Technology. His, his actual title is Chief
- 28:57Information Technology Officer and I would imagine that he
- 29:03would argue that the UN is the right place to do that.
- 29:06But it's a, it is an interesting question because so many
- 29:11different countries have different levels of maturity
- 29:16when it comes to the development of privacy regulations and
- 29:20issues around AI ethics. But as you mentioned earlier,
- 29:27there is also divergent in terms of their goals and their there
- 29:34is competition for for perspectives here.
- 29:37It's been on my To Do List to watch last Friday's episode so
- 29:40I'm a little behind. For an organization that wants
- 29:44to start implementing a program of AI ethics, where should they
- 29:50begin? And I mean implementing it in a
- 29:52serious way, not just not just giving Idlib service.
- 29:55That means having your C-Suite being involved, first of all,
- 29:59making sure that there is executive backing.
- 30:02It doesn't always mean having, you know, somebody whose entire
- 30:06job is to be the chief AI officer.
- 30:08We've seen a lot of that at some organizations, particularly
- 30:11bigger organizations. But sometimes that that title is
- 30:15added on to like the chief data officer or the chief privacy
- 30:18officer. So depending on the size of your
- 30:20organization, it's really going to depend.
- 30:23But the important thing is having somebody in the C-Suite
- 30:26who is backing that. And I've seen a lot of success
- 30:29for the models that are doing you having somebody who's the
- 30:32chief AI officer to coalesce everybody.
- 30:35But a, the success of an AI governance program is really
- 30:40built by so many more people than that.
- 30:42So you have to get your SISO on board, your CIO, your chief data
- 30:47officer, you know, if those are all the same person, then
- 30:51congratulations, you're probably also the AI ethics officer as
- 30:54well now. But really figuring out just at
- 30:57a baseline, what do you already have in place?
- 31:01I mentioned earlier like assessment and authorization
- 31:03processes, any sort of like data, you know, data impact
- 31:07assessments that you already have.
- 31:09What are your privacy requirements?
- 31:11So you really have to figure out at a baseline, what do you have?
- 31:14And then what do you need to add to that to be able to do AI
- 31:18governance. So kind of what's the, what's
- 31:20the gap analysis that you need to do?
- 31:22And for some organizations that that really is enough.
- 31:26Hey, we're going to add a few questions to our privacy impact
- 31:30assessment. And now it's an AI impact
- 31:31assessment. We need somebody to sign off on
- 31:35the cybersecurity aspects of AI and the data and make sure that
- 31:39that is, you know, that part's being considered a little bit
- 31:42differently. But one of the big changes I
- 31:45think that we see with AI is that you can't, and not that you
- 31:49should be doing this for regular systems anyway, but you can't
- 31:52just like set it and forget it and come back to a system like
- 31:55once every year or something. How often you come back to that
- 31:59is going to be periodic and it's going to depend on a lot of
- 32:02factors, including like, what is the purpose of the AI?
- 32:05How often is that algorithm changing?
- 32:07Is it machine learning? And it's changing some every
- 32:09day. And what are the consequences?
- 32:12So if either the consequences are are really significant, you
- 32:16can't just like look at it once a year and see if it's still
- 32:18working. So that I think it's going to be
- 32:20one of the pieces that has to change the most for a lot of
- 32:23organizations is having enough of a process in place to go back
- 32:27and look at your AI on a periodic basis that's
- 32:30appropriate for that exact use case.
- 32:33I have to imagine that will be a significant change for some
- 32:36organizations because if you're used to dealing with older non
- 32:41AI technologies, you put them in place and of course it's
- 32:45basically static. But with AI, and especially
- 32:48generative AI, as the models change and as the prompts
- 32:53change, the results can be very, very drastically different.
- 32:58They can be completely different.
- 32:59And then, you know, the other thing that's changing almost to
- 33:02as quickly it feels like is the, the legal compliance regimes
- 33:05around us. So the, the, there are new laws
- 33:08coming out. It feels like every day, new
- 33:10standards, you know, being created.
- 33:13It feels like every day and then boom, you know, suddenly there
- 33:16is a text to video generator and you have to rethink everything.
- 33:20So making sure that there are people who have the time and
- 33:23space to be focusing on, OK, well, what are the new
- 33:26technologies that are coming down the line and can build
- 33:30something that's flexible enough to anticipate, you know,
- 33:33significant change that might happen short term.
- 33:36But things that probably aren't going to change are things like,
- 33:39OK, well, we need to have some sort of data impact assessment
- 33:42or if we're procuring things like we need to have, you know,
- 33:45some baseline procurement processes.
- 33:46And there's been some great work being done on procurement by a
- 33:50number of different organizations.
- 33:52So I think those are those are really critical pieces for a lot
- 33:55of organizations. Making sure that you have your
- 33:58lawyers involved, obviously, especially in that procurement
- 34:01or if you're selling something. But yeah, also having people who
- 34:04have the space to focus on, OK, So what laws might be coming
- 34:07down the Pike? Because if you are working on
- 34:11creating a governance program, but there's going to be a big
- 34:14compliance regime, you know, being implemented, just build
- 34:18towards it now and think through to, you know, it's, it's easy
- 34:22for me to say all of this, right?
- 34:24But nobody has nobody has time to do this for every single AI.
- 34:29So really being able to rack and stack like where that risk is
- 34:33the highest and focusing your energies on that.
- 34:36You've raised a really interesting point with AI in
- 34:40particular. You can't, as you said, take a,
- 34:43a static view because it is changing so rapidly, which
- 34:48places an additional burden on folks who are dealing with these
- 34:51issues to not only manage and lead what's inside their
- 34:57organization and the ethical issues around that, but to a
- 35:01very large extent also be looking outside into the
- 35:05environment, into the changing legal environment, the social
- 35:07culture environment and the expectations.
- 35:10So that's what that's like a very large, it becomes a large
- 35:13mandate. I think that's part of why we're
- 35:15seeing more organizations have somebody who is a chief ethics,
- 35:20you know, AI ethics officer or you know, why we're seeing some
- 35:23expansion. And I, you know, I, I don't want
- 35:26to put words in their mouth, but I think that this is one of the
- 35:28things I've really liked about how IPP has been expanding their
- 35:33offerings is understanding that a lot of this has been falling
- 35:36on your privacy and compliance officers.
- 35:39So they need to know more about AI, They need to know more about
- 35:42cybersecurity, They need to know more about these different
- 35:45areas. So you're, I actually help them
- 35:47develop the AI governance professional certification over
- 35:50the last couple of years. And I helped teach that class to
- 35:54make sure that, you know, we're kind of bringing up the next
- 35:56generation of AI governance professionals.
- 35:58And the people I've had in my class hail from far beyond
- 36:01privacy and I so it, it really is expanding, but kind of
- 36:06flexibility and an anticipation are the new nature of the game.
- 36:10Like you said, it's not static anymore.
- 36:12You can't do like a waterfall development process.
- 36:15You really have to be more elastic.
- 36:17And it's so interesting that if you're, if you're not, if you
- 36:21don't have your ear to the ground when it comes to likely
- 36:25potential regulations and social or cultural expectations, you
- 36:32can really be screwed right out of almost out of thin air
- 36:36because you weren't watching the broader view beyond your narrow
- 36:42domain. That's really true.
- 36:44And there's there's so many pieces that either still norms
- 36:48forming around to you. So you know, anticipating some
- 36:51of them, but also like if you have the ability and in your
- 36:55organization to be able to actually do some of this and
- 36:59figure out how to do things like, OK, well, what should an
- 37:02AI incident response program look like?
- 37:04Is it the same as a privacy incident response or a cyber
- 37:07incidents response program? Probably not, because where an
- 37:10AI goes wrong could be the data was wrong.
- 37:13So if you're going to have a different set of stakeholders in
- 37:16place, but if you can help create processes that work,
- 37:19those might start becoming the norms, right?
- 37:22And go teach people, hey, here's what's working for us.
- 37:25You know, we had an incident and now, you know, because we had
- 37:28this process, we knew where to go back and fix the algorithm.
- 37:32We got rid of, you know, this, this errant data stream and now
- 37:36it's working again. Or, you know, whatever your
- 37:37system is. But yeah, I think if you have
- 37:40time to focus on some of those more nascent issues, you may
- 37:44also be able to help influence, you know, what those norms are
- 37:47and make sure that it actually works for an organization like
- 37:50the. And I say this as a policy
- 37:52person, but like I try to bring a data scientist along with me
- 37:55to every conversation I have because like nobody wants the
- 37:58lawyers or just the policy people to be writing this like
- 38:01it has to actually work at the end of the day.
- 38:03So I would very much foot stomp having an interdisciplinary team
- 38:08on that. And on that exact topic, we have
- 38:12a question from Elizabeth Shaw on Twitter who says, what is the
- 38:19cross impact of the ethical use of AI on corporate or CIA ethics
- 38:26programs? The impact on each other, You,
- 38:29you touched on this a little bit earlier.
- 38:32There is kind of a cultural aspect to this where you to
- 38:37enable adoption of AI, which organizations all need to do,
- 38:42right. You have to have an
- 38:44understanding of what your organization's values are.
- 38:47What does ethical AI mean to you?
- 38:48What does governance mean? And that can also be like
- 38:52everybody. So even if you aren't somebody
- 38:55who is developing AI for your organization, if you were
- 38:58thinking about using Copilot or you are using ChatGPT or Google
- 39:02Notebook or whatever you are using to help you do your work,
- 39:06you need to understand what your organization's kind of ethics
- 39:09and requirements are around that.
- 39:12And what data might you be sharing with, you know, with
- 39:15open AI or Microsoft or anybody have have an idea of what your
- 39:19company or organization is comfortable with for that.
- 39:23But also, you know what, if any requirements are there for
- 39:27double checking things, making sure that you're explaining that
- 39:29something was created with the help of AI, just like
- 39:32cybersecurity, right? Every organization has
- 39:34cybersecurity training, I'm assuming if you don't, let's
- 39:38let's talk afterwards. But you people to take that so
- 39:42they know to not like, you know, click on a certain e-mail and
- 39:46you and infect something. And that same sense of
- 39:49responsibility and same sense of, you know, shared values
- 39:53should be applied for AI as well.
- 39:55But I also think it's just important because that is part
- 39:58of your organization's ethos and, you know, part of your
- 40:01culture and being able to say, like, here's kind of where we're
- 40:05coalescing around and here's your part of that is really
- 40:08strong. We have another question from
- 40:11Twitter, again from Marcel and Khan.
- 40:13Arsalan is really on a roll, and he says, should AI be regulated,
- 40:19and if So, what about the impact on innovation?
- 40:24I really do believe that when there's the right balance of
- 40:29regulation and governance, it can help actually speed up some
- 40:34of that innovation. So, but this is it's so much
- 40:37harder with AII think then with some other technologies, because
- 40:41you've some regulations are very much looking at just the AI
- 40:46technology that exists today and some of the AI technologies that
- 40:50people are building towards. Maybe you need a different type
- 40:54of governance all together for. So I think you we have to be
- 40:58very conscientious to get that balance right as a society.
- 41:02But going back to the car analogy, I mean, cars didn't
- 41:06have brakes and so you can only go a few miles an hour and then
- 41:09once you added brakes, you could actually speed up.
- 41:11OK, we'll extend that out. So, you know, if somebody said,
- 41:14OK, well, we should have roads and we should make it where, you
- 41:17know, in some countries you can take a right on red and others
- 41:21it's, you know, not not allowed or whatever.
- 41:23But like those things allow different cars with, you know,
- 41:27with different drivers to be on the road at the same time and
- 41:30still move around. And I think that's where we need
- 41:33to be able to get. Is this your kind of nice
- 41:37equilibrium that allows innovators to know, OK, well, I
- 41:41know it's not what I know what's legally allowed soon.
- 41:45I've, I know what space I have to move within that so I can
- 41:48move fast and not worry about somebody coming in afterwards
- 41:51and say, OK, well, you can't do any of that.
- 41:53Your whole company is gone. And we, we have to have some
- 41:56amount of of, of kind of rules of the road so that they can
- 42:01work together. But also I think that's going to
- 42:03be important because AI is going to have to work with AI, right?
- 42:08It's not all about humans anymore.
- 42:10Like we're going to have to have some standards so that, you
- 42:14know, we can say, OK, well, this AI is going to be interacting
- 42:16with this AI. And, you know, how do they know
- 42:18that they can do that? So we have to be able to build
- 42:21towards like the world is going to exist instead of just the one
- 42:25that we're in today. And that's a, it's a huge
- 42:27challenge. I don't pretend to know the
- 42:29answer. I think we're all working to get
- 42:32there a little bit at a time every day.
- 42:34It's a very interesting point that having clarity around
- 42:39what's allowed, what's not allowed, enables
- 42:42interoperability to take place and predictability as well,
- 42:47which is so important in for any organization.
- 42:51Yeah, and, and for sales too, right.
- 42:52Like if somebody is going to, I mean, I, I don't know why this
- 42:56is what just came to mind, but I had a Furby as a kid and I don't
- 42:59know if anybody even remembers those.
- 43:01But there was a lot of controversy when, you know,
- 43:04everybody bought them and then they kind of did different
- 43:06things and people were like, OK, we didn't, we didn't know what
- 43:10we were getting into. Like, I would like to return my
- 43:13Furby if we're like, we're like my kids.
- 43:15Furby is now in a closet buried underneath like a bunch of coats
- 43:18or something. So like nobody can hear it.
- 43:21That's not what you want, right? Like you want your technology to
- 43:25be adopted and trusted. And, you know, for people to
- 43:29say, hey, this is this was really well done.
- 43:31Like I understand how it works. I have as much transparency as I
- 43:34can. That trust is still, I think at
- 43:37the, you know, at the baseline of a lot of this.
- 43:41And you have to have, you know, the trust has to go both ways.
- 43:44The technologists have to trust that the policy people aren't
- 43:47going to just change their minds all of a sudden, right?
- 43:50So that conversation really has to have all the parties at the
- 43:53table. And I mentioned this earlier,
- 43:55like I like to bring a data scientist with me basically
- 43:58everywhere. If I could actually just bring
- 43:59one everywhere, that would be fantastic.
- 44:01But it it's, you know, we have to make sure that just because
- 44:05something sounds like a great solution, it will actually work.
- 44:09And a lot of times you that's a conversation.
- 44:12And I'll be honest, like sometimes the data scientists
- 44:15are the ones who come up with better policy solutions then the
- 44:17policy people because they understand really the nitty
- 44:20gritty of like, OK, well, here's what's possible or like, hey,
- 44:24and you know, that's not what we're we're building towards.
- 44:26So making sure that that is a conversation, not just happening
- 44:30kind of as a blue sky exercise with a bunch of policy folks is
- 44:33really critical. That's definitely important.
- 44:38If you if you start making policy without having a a pretty
- 44:45complete understanding of the implications of the technology,
- 44:49the data, how it's actually being used, and what the
- 44:52implications are, then your policy is going to miss the
- 44:55mark. Yeah, absolutely.
- 44:57Can you talk about governance? You've mentioned governance a
- 45:00few times. Can you drill into what do we
- 45:03mean or what do you mean when you talk about AI governance?
- 45:07If you are still struggling to get your data governance program
- 45:09in place, I would say continue that work, but with an eye
- 45:13towards what you might need to do for AI as well.
- 45:16But your governance means a lot of things.
- 45:20It, I think you're at the baseline, it's processes that
- 45:24enable you to have responsible AI, ethical AI.
- 45:29But some just very high level parts of governance in my
- 45:34opinion, are things like, OK, define what is it that you're
- 45:37actually trying to do with this system?
- 45:39What is the scope of it? What is its operational
- 45:42environment and what specific tasks are are it going to
- 45:45perform? Because if you define those,
- 45:48then you can then do this analysis of does that actually
- 45:51work within our organization's ethical principles and our
- 45:55goals? Because sometimes things sound
- 45:57like a great idea, but then when you start taking, you know,
- 45:59taking a step back or peeling that onion back, you realize
- 46:02like, well, maybe that the impact of that is actually not
- 46:05aligned with our goals. So having, you know, governance
- 46:08to me is even just sitting down and having that conversation up
- 46:12front of like, what are we trying to do here?
- 46:14And is that part of our organization's principles?
- 46:17And then from there, OK, well, you know, what data do we need
- 46:20to do that? Are we allowed to have that
- 46:21data? What our stakeholders think
- 46:24about this, you know, whether that's internal stakeholders or
- 46:27external stakeholders? Are you in a highly regulated
- 46:30environment? What rules apply?
- 46:33What rules might apply going forward that we should plan for
- 46:36now? There's some really interesting
- 46:39work being done by various different, I would say like a
- 46:43lot of former government people too, where they're saying like,
- 46:46hey, if if you're trying to build something, if you're
- 46:50trying to build something that's going to do X, you should
- 46:53probably figure out how to do, you know, privacy protections
- 46:55upfront. So you're being able to have
- 46:58anonymization and, and things like that.
- 47:01So you have an idea of what rules might apply tomorrow as
- 47:06well as what rules apply today for just your plain data
- 47:10requirements. So that's a big chunk of
- 47:12governance as well. And of course, bringing in your
- 47:14lawyers is always good advice. So making sure that that
- 47:18compliance is happening not just with any of your policy folks,
- 47:22but also with your lawyers. And then, you know, I mentioned
- 47:26earlier like common ethical principles, whether that ends up
- 47:30being OECD or the UN, as you talked about, making sure that
- 47:34you're also thinking bigger than just your organization.
- 47:37OK, So what what governance do we need to have in place to have
- 47:40trust with people who are in this country or, you know, in
- 47:45this part of the world? So be thinking like bigger than
- 47:48just that as well. You know, that could even
- 47:51include things like transparency requirements or security
- 47:55requirements and accountability. And then also, you know, not
- 47:59just doing all of this because it's like the right thing to do,
- 48:03but making sure it really is part of your organization's
- 48:07ethos and value system. So I mentioned earlier you
- 48:11having nobody has enough resources to do this for like
- 48:14every single little piece of AI because it's just, it's going to
- 48:17be part of a lot of what we do. But making sure you understand
- 48:20like where you really, really have to do this and you have
- 48:23processes in place for, for actually going through some of
- 48:26that governance. And then you, you talked about
- 48:30bias several times. Like that's absolutely part of
- 48:33governance. I would say it's not just about
- 48:35bias, but it's also about fairness.
- 48:37So making sure your data sets are representative and that you
- 48:41understand like who could benefit from them and who might
- 48:44not benefit from them. And write those down and figure
- 48:47out what mitigations you need to have.
- 48:49And then have that ongoing monitoring to make sure that
- 48:52you're still being fair and being not having undesired bias.
- 48:57And then having transparency, you know, whether that's with
- 49:00your internal stakeholders or you might be required to have
- 49:03that with external stakeholders at the key part of governance as
- 49:06well. And then accountability.
- 49:08So who like basically who is responsible for making sure that
- 49:12this use case goes the way that it should?
- 49:15That's a really critical piece as well.
- 49:18You know, who's going to continuously monitor this?
- 49:21Who's going to do the auditing if there is an incident?
- 49:24What is your process for responding to that and
- 49:27mitigating that and maybe even briefing it out and documenting
- 49:31it and remediating it and everything?
- 49:34So those are all really critical pieces.
- 49:37I would also say just iteratively improving on
- 49:39something. So not just checking to make
- 49:41sure it doesn't go wrong. But as we have new technologies
- 49:43or as we have new standards, you know, how are we going to
- 49:47integrate that into our system? So governance can mean more than
- 49:51that. It can again change kind of
- 49:53depending on the organization, but I think those are some some
- 49:57kind of basic areas where most organizations align about what
- 50:01is governance. Wow, you've just given us the
- 50:03PhD course on governance. Oh, hardly.
- 50:07There's so much more to it. I can see you're very passionate
- 50:10about this topic. I could do it all day, every day
- 50:13actually. That's kind of my job.
- 50:14So plus, you know, plus privacy and civil liberties and all that
- 50:19other stuff. But no, it's, I've been really
- 50:21lucky to be able to work on this for over six years at this
- 50:24point. I like to say I did it like
- 50:27before, it was cool. But it, it is amazing how with
- 50:31the advent of generative AI, you know, so much more focus has
- 50:34come to, you know, come to bear on these issues.
- 50:38And I think it honestly, a lot of it's just gotten more
- 50:41complex. And for me, that makes it more
- 50:43interesting. But I'm really glad that people
- 50:45are having this conversation. I think, you know, to kind of
- 50:49together we're all creating those norms and helping move the
- 50:52world forward on it. What advice do you have for
- 50:57business people? What can business people learn
- 51:01from government efforts? There's so much out there that
- 51:05you can pick and choose from and kind of make sure that you're
- 51:08tailoring it to your organization and you looking at
- 51:13anything from, OK, well, what are transparency reports looking
- 51:17like? I mentioned data impact
- 51:19assessments, privacy impact assessments as well as you, I
- 51:23briefly mentioned like system of record notices that explain to
- 51:26the public, here's how we protect your data.
- 51:28Here's the requirements there. Those are all fantastic
- 51:31resources to get started. If you're like, I don't know
- 51:34what to use, but also we, I think, you know, not just
- 51:38because the government has experience, but they also have
- 51:41lessons learned. So I would, I would learn a
- 51:44little bit about the history there, but also maybe learning
- 51:48about, you know, what's not the same.
- 51:50So if we're, you know, we're working for the American public,
- 51:54maybe not for shareholders. And what is that Nexus?
- 51:57And like what? What might you want to do a
- 51:59little differently? Very interesting.
- 52:01Well, a huge thank you to Jacqueline Acker, who is the
- 52:06deputy Privacy and Civil Liberties officer with the CIA.
- 52:10Jackie, thank you so much for taking time to be with us.
- 52:13Your very interesting comments. Oh, thank you.
- 52:16You know, I'm, I'm glad to be here and share kind of my
- 52:18perspective and thanks for all the great questions to you.
- 52:21I, I know I will be thinking myself about some of those all
- 52:24weekend. So.
- 52:25Yes, thank you to everybody who watched you, folks who asked
- 52:28such great questions. You guys are really smart.
- 52:31Before you go, please subscribe to our newsletter, check out
- 52:37cxotalk.com. We have excellent shows coming
- 52:41up. Subscribe to our YouTube channel
- 52:44and we'll see you again soon. Everybody, thank you so much and
- 52:48have a great day.