Natalie Tecimer: Navigating Global AI Treaties

0:00What can 160 years of experience teach you about the future? When it comes to protecting what matters, Pacific Life provides life insurance, retirement income, and employee benefits for people and businesses building a more confident tomorrow. Strategies rooted in strength and backed by experience. Ask a financial professional how Pacific Life can help you today. Pacific Life Insurance Company, Omaha, Nebraska, and in New York. Pacific Life and Annuity, Phoenix, Arizona. Welcome to the AI Chat Podcast. I'm your host, Jaden Schaefer. Today on the podcast, we have the pleasure of being joined by Natalie Tessimer, who is a fellow at the NYU Law Center for Cybersecurity.

0:40She was formerly at the U.S. Department of Justice, where she was a lead for AI, representing the U.S. in global treaty negotiations at the Council of European Convention on AI and a federal prosecutor. She also worked at the U.S. Department of Defense on big data and AI issues. She has policy experience from CSIS, and she received her bachelor's from the University of Southern California and her JD from Georgetown University Law Center. We're super excited to have you on. Welcome to the show today, Natalie. Thanks so much, Jaden. I'm really happy to be here and have a little chat about AI. Super excited to have you on, as I mentioned.

1:16What I wanted to kind of kick this off with was asking you a little bit about, you know, if you could tell us a little bit about how your experience as a federal prosecutor, how that kind of shaped your perspective on the ethical implications of AI? Yeah, absolutely. You know, one of the fascinating things about being a federal prosecutor is you are working for law enforcement. You are a law enforcement officer in some ways. But the people I was working with were so concerned about the ethical implications of every single thing that we did and about every side. And so that means that even though I was on the law enforcement side, I feel like I had a much more well-rounded perspective than I might have had otherwise.

1:54And a big part of that is a credit to the Justice Department. We were always bringing in other people to kind of hear their thoughts and and kind of civil liberty sides of things. But so I was in the computer crime and intellectual property section. And one of the big things that I was really focused on is, you know, major cyber threats. And I was very exposed to child sex trafficking cases. And that is where I really saw, I think, to me, one of the biggest benefits of AI, because our prosecutors and our enforcement is able to make incredible strides in finding child sex crime perpetrators using facial recognition and using AI in a way that is night and day without the technology.

2:37On the flip side of that, though, I also saw how people would try to use facial recognition for the other side and try to use it for oppressing civil liberties. And so one of our jobs as a prosecutor is to try to make sure that we were always looking at the ethical implications of everything we were doing and doing kind of a balancing of equities. Yeah, that is so, so fascinating. That's incredible. You know, a lot of times I hear from people talking about the downsides of AI and kind of like, especially with the facial recognition stuff, I think we hear and we talk a lot about like countries like China and the ways they're kind of implementing facial recognition in smart cities to, you know, ding your social credit score if you jaywalk when you're not supposed to, or of course, like worse things.

3:22And I think a lot of people kind of get concerned about some of the technology from that perspective. But that's so it's so cool to hear from your perspective, like some of the incredible use cases that are really helping people, of course, like an area like child sex trafficking is such a difficult field to work in. And any tools that you have that can help with that, like make an incredible, you know, have some incredible benefits to really helping that. So that's so interesting to hear your perspective on that. I'm wondering if you can share some insights from your time at the US government in regards to kind of the role of AI and national security specifically?

4:01Absolutely. You know, there's so many different kinds of AI tools. And I think that, you know, in my various jobs, whether I was at the Defense Department, kind of doing a little bit more of intelligence work or at the Department of Justice, I got to see a lot of different kinds of it. So I mentioned facial recognition, object recognition is a really other interesting case. I think that, you know, there's a lot of battle space awareness that can be done in a different way with AI algorithms and AI tools that make it safer for troops who are on the ground. There's obviously a lot that happened on the drone and autonomous weapons side of things.

4:37And I was really proud that the government has always had ethical AI policies, I think, far back as 10 years ago, on those kinds of issues. But once again, a lot of what I saw is, how do we use AI and technology and national security to minimize any kind of risk to civilians and also to U.S. people? And then the other interesting side of this, too, is protecting personal data. And that gets into cybersecurity. And I think that when folks think about intelligence and national security, they don't always think about cyber defense. But AI is playing a very big spot in that space to make sure that we're protecting our critical infrastructure.

5:17So that's amazing. And I'm super happy to hear from you, you know, really the commitment the U.S. government's made. I think probably not a lot of people know exactly where the U.S. government stands on that. So that's definitely very comforting because it's definitely a concern I'm sure a lot of people have, especially as we're building out a lot of these systems. and yeah again just so many incredible use cases that you know really have the ability to help so many people that have the ability to help our troops and and so many others so really really fascinating something i'd love to ask you about um is a little bit about you know what are some of the key challenges you faced when you were representing the u.s in global treaty negotiations I think specifically at the Council of Europe Convention on AI.

6:04The biggest challenge was the definition of AI. And it's very nitty gritty. It is word by word specific, but it changes everything. Are we thinking about AI as being any kind of prediction making tool? That would include a word predictor in Microsoft Word or on your Outlook email. Or are we only looking at tools that have a sense of autonomy and that are really making kind of human-like perspective choices? That closes that box a lot. And so there's a very big debate in the international community and even within the United States about what those definitions look like because it changes how we regulate and how broad regulation can be.

6:43So that's the first one. The second one is, I mean, I call it the China problem. That's my own personal label for it. But what did we do about countries that are using AI, like you mentioned, in ways that are extremely averse to protecting civil liberties? And how do we create a regulation that allows other countries who are kind of going through the appropriate levels of checks and balances and ethical review to have similar AI capability, but not get into this standpoint of where China is? So a great example of that is China has an AI prosecutor. and so you mentioned they do social scoring it's more than that they've got an ai algorithm that's automatically recommending people to have charges faced against them we would never do something like that right we would never do something like that in the united states because we require so much more in our judicial system to even bring a case so i feel comfortable with that but then if we think about facial recognition we and you and i both want facial recognition when we go to a large sporting event.

7:46We want every single person who's coming in to be scanned against the top 10 terrorist watch list. That isn't our safety. But then what happens when you have a company like Madison Square Garden, who's using the same technology to then pull out people who are working for a locker room representing against them? And so how do we regulate that? And can we regulate the technology? Or are we trying to regulate the use case? But then you can't really regulate the use case because it's really about how it's used and for what purpose. And that gets really hard when you're sitting at a negotiating table.

8:17And then the last thing I'll say that I was a little surprised. These were my first international negotiations. And it's very slow. Everyone has to get clearance on every word they say, which is good. It shows these are a lot of westernized countries that have democratic processes. But it means that when you're standing up on the floor, you've got your three sentences you're allowed to say. And then if that's it. And then if we want to change something, we have to go back in the United States and get a review from 10 individual agencies and the White House and kind of public input to change what we're going to say, which is hard.

8:52Yeah. Yeah. That that definitely sounds incredibly challenging, especially when you have multiple countries doing that. Yeah, it's so fascinating, you know, talking about, of course, the Madison Gardens situation was, I think a lot of people think it's like kind of like funny, but also ridiculous. And it's like, okay, we definitely should should fix that kind of problem. But then yeah, you don't want terrorists coming into a sports event. So like you understand why the technology is there. I think China is definitely a big issue. A lot of people are concerned about. They've just rolled out a bunch of regulations recently, which I'm sure you've seen where like, essentially, they have regulations and rules for training AI models.

9:29you're not allowed to have like a certain percentage of the data in your AI model that goes against like their, their, uh, you know, the communist parties, like ideologies and lines. Like if they can find more than a certain percentage of like the total data included. So it makes it really difficult. You're right. You have like, of course, open AI, which I scooped up the whole internet. And so we have these massive data sets to create, like Chinese companies are really going to struggle in, in those regards. And in addition, of course, it's like they can't include anything in the training data that the Chinese Communist Party is blocking with their, you know, the Great Firewall or whatever.

10:03So anything they're trying to censor from the government. So it's just like very fundamentally different ways we're looking at AI. But China is kind of the other big player producing some of these multi-million, multi-billion dollar AI models and plays. And I mean, my personal concern is that China is going to try to compete with us at a very high level because they see what the stakes are. They're going to create these powerful AI models and then they're going to get rolled out and American consumers are going to take like, well, this like really cool GPS tool with like a Chinese AI model embedded in is really useful.

10:34So like, I'm just going to use it because it's cheap and cool, but then there's going to be like the almost like ideological warfare or some of the concepts that us as a country may not agree with, but they're embedded into the AI that we have taken. It's almost like a Trojan horse. So that's what, that's, that's one of my concerns in this whole regard. In any case, in regards to this whole kind of geopolitical arena and area, what I'm what I'd love to kind of get your opinion on is like, how do you think AI will affect global geopolitics, especially in terms of, you know, international law and treaties?

11:06Well, Jaden, you brought up the transfer of data and how Chinese data sets are kind of being shaped differently than U.S. data sets. And that's a great place to start here because the other side of that is a little bit more of the national security side. How is how does a company like OpenAI, who has these massive data sets that can in some ways represent a lot of U.S. purchasing power and a lot of consumer demand? And then that data set can then be used by a foreign adversary, which could be a potential threat. And I think that that kind of ties into how the U.S. and other countries should think about AI from a global side, because all of the sudden data, and we've seen this in other ways, but data becomes the ability to cross borders and be used to make predictions against someone's population potentially.

11:58So that's one thing I see. I also think that there's this piece of how other companies from different countries and how the country regulation themselves are going to want to use AI tools differently. And we see that a lot. The U.S. I think has been very at the forefront of wanting to use AI tools to improve Defense Department production. I absolutely love the op-ed that Deputy Secretary Kat Hicks wrote just about all the possible benefits of the Defense Department using AI. But there's other countries that might not think that that's appropriate to use AI in any relation to military, Whether we're just trying to improve a human operations and human resources side of things.

12:45And that's a sticky point. And then on the other side of this, there's countries that are using AI much more for tax benefits determinations and social welfare determinations. And we in the United States might not think that side is appropriate because so many of these cases are so fact specific based on the individual person that how could you even create a generalized model to be able to answer a question. So that factors in a lot on the regulatory side. And I think that we will also see some potential shifting of alliances on this specific issue. And we see that across other sides. There are some countries that we are very strongly aligned with on specific issues.

13:25And I think we might see that a little bit in how this geopolitical game popped up. But a lot of this is going to end up being, I think, Western countries countering China. Yeah, I tend to agree with you there. Like I cover a handful. I pretty much produce like three or four podcast episodes a day covering a ton of different news. And I've been doing this for the last year. And the overwhelming trend I'm seeing here is it's like the West versus China on a lot of these AI issues. If we're just looking at who's putting the most resources and most manpower, the most regulatory thought into this, there's a lot of different aspects.

14:02I think China is the major adversary, if you call it that. I know. Yeah, anyways. So I think it's very interesting. And a lot of the issues that you brought up, I think are really important to think about. I feel like we are seeing a bit of a shift maybe here in the West. Not totally, but there was a time a number of years ago where a lot of the top tech firms in the US refused or a lot of employees at them refused to work with the Department of Defense. Google, of course, is a very notable one that was doing some deals and a lot of the employees kind of revolted and they stopped doing that. So there is this kind of like concept, but I do feel like the tides are shifting a little bit in a sense that we see, you know, there is a need.

14:44There's obviously like a lot of geopolitical strife. There's Russia invading Ukraine. There's Hamas and Israel. There's China threatening Taiwan. There's a lot of areas. And I think people are realizing like we really do need to make sure that we are taking every competitive advantage we can in probably a military sense, regardless of your opinions on war and peace and how to bring that about, making sure we have the strength. And so I think that transfers from the tech side. Now AI is the big focus of the tech side. So I'm seeing a lot more people open to that. There's a lot of ethical questions.

15:14And I know you brought up a lot of issues there. So that kind of brings me to my next question I would love to ask you about, which is like, if you can elaborate a little bit on the concept of responsible AI and how you think it kind of differs across industries. Absolutely. Responsible AI is a buzzword that gets thrown everywhere, as I'm sure you're known and as you know. And people are always saying, well, we think AI should be more responsible. My question to companies and to different kinds of public service folks is always, responsible to who? Because if you've got a company who wants to be responsible in AI, responsible to their investors might look very different than responsible to the public.

15:51And so that brings up a very interesting balancing of equities problem. Responsible to the public might be really taking every possible step to protect data and to not do any kind of sourcing from places that might have any wishy-washy data regulation and PR issues and potential privacy problems. But that might not make the most amount of money. or responsible to the public might be, we are not going to release this latest model that might be a video AI algorithm that could create a deep fake. But the investors might want that or the board might want that or even consumers might want that. Some companies, there's a few different video AI companies that are really very successful in creating video avatars for consumers, for companies to use for training videos.

16:44but those video avatars could potentially be used for something else depending on which company we're looking at so that's always hard i think that it does there needs to be a balance and test between the the rights of the rights and the interest of the public the rights and interest of consumers and the rights in the interest of investors and i'm always probably going to come out a little bit more on the public side but i understand that that's not always sustainable for companies. And so there's a bit of that. I do think that we're going to see that the companies that are tracking a lot of these upcoming regulations, which really are geared toward making companies more responsible in their AI, having training for all of their employees coming into contact with these algorithms, having different kinds of transparency requirements, those are going to be the companies that make it.

17:35And those are going to be the ones that are around in 10 years. Yeah, I 100 % agree with you. And I love your concept, your question of responsible to who when talking about building responsible AI. So that is a really pointed question a lot of companies and people should be asking themselves, I think. One thing that I struggle with when looking at this issue is when you're talking about responsible AI, and I want to ask you a little bit about some of the ethics in corporations, but the thing that I struggle with is a lot of times when I hear people be like, oh, you need to make sure your AI is responsible, They're like, OK, well, X, Y and Z technology we could use for bad.

18:14We need to kind of shut that down. I don't I believe that inevitably if the technology is there and exists, people will create it and it's going to go open source and just not be as controlled if that's the way it goes. So like the thing that I struggle with is like I've kind of removed myself from my like, oh, you can't do certain technologies because of the bad implications of like bad implications, which might have been like more of my opinion before to like, OK, we have to just let everything everything right out in a way we have to build everything. But because whether we do it or China does it, someone's going to do it.

18:47And then we just have to ask ourselves, like, how can we make this tool as safe as possible? Because it's inevitably going to get built. People are going to make bootleg open source models of whatever we don't want them to do. So like, you know, what are our offensive approaches to defend against that? Anyways, that's, I'm sure a little bit more of a philosophical question, but something I would love to ask you kind of in that vein is, What are the most pressing issues in AI ethics that corporations need to address today? Absolutely. Data privacy is one of the biggest ones because all of these companies are, we've seen that the data privacy side has not always been at the forefront.

19:29And when it's not at the forefront, then we have problems. and an example of this might be you know the earliest models of chat gpt that where the api could be taken by any company and then that api data was being sent back to the main company well what happens if you're a small health care company and you've gotten you so you you have this api of chat gpt embedded in your website you've got someone coming on there and saying you know i'm 60 years old i have this problem in my leg i have xyz and then the chat gpt algorithm is kind of helping them figure out what kind of doctor they need to see, as an example.

20:00Well, when all of that data flow was still continuous with not just OpenAI, but it wasn't so strongly integrated and protected only in that small API user, that means that now we've got potentially this person's medical information. And we're living in an age where no privacy information can actually be de-anonymized. That doesn't exist anymore. Now there's always a way to find out who someone is, what their background is. And there are a lot of people who say, well, so what? Well, that now gives information for potential hackers. And we see that happening a lot. We see a lot of organized crime groups.

20:38And this is something I saw a lot as a prosecutor. We see organized crime groups who are taking people's names and identities. And then they're able to find their credit card information. And then we've got really small mom and pop shops around the world getting scams out of$100 ,000. Well, that's everything to these people. And these kinds of consequences really are detrimental to the functioning of society in this kind of digitized world. So that's my big concern is data privacy. The other big ethical concern is, as you're saying, technology that goes a little too far. I think that there are technologies that are being developed that might be pushing that line.

21:21I understand what you're saying about how you can't always stop the development of that technology. But I also think that there is some responsibility to not push it out and not commercialize it in a way that really makes it available to everyone. We don't do that for chemical weapons. I think that, you know, we all know that probably there are many places in the world that have the ability to develop a biological weapon, but we're not pushing those out. And I think that in some ways we need to look at some AI capabilities in a similar way. An example of this might be a real time instantaneous, no human review required drone.

22:03That does not exist. but that would be detrimental but we could very easily create that technology but we shouldn't and if it is being created we should not commercialize it we should not publicize it so i do think companies have a duty to do that yep i i i do agree with you especially on the company side um especially when you look at like what are the barriers to entry well to create something like chai chibiti it's hundreds of millions of dollars so like yes we shouldn't create like bad versions of that technology and spend millions of dollars developing tools you know of course if our adversaries or other people bad actors want to develop that like there's not as much we can do to stop them we'll try our best um but definitely not giving them uh you know a leg up on that as far as the autonomous drones i did see reports that ukraine i believe uh was launching some of the first autonomous ai drone technology um so like not a lot of comments on that, but it does look like it exists and it's out there.

23:04If it's possible, people will make it. And so my thing is just like, what do we do? If all the things that were possible were made, like what are the safeguards? What are the things we put in place? Something I would love to ask you about is, how do you see the role of regulatory bodies evolving in the context of AI development and deployment? A lot of what regulatory bodies are going to end up doing is preventing companies from being able to produce certain kinds of technologies if they want to sell them to different audiences. So that's going to be, I think, the EU AIX impact on this whole space, for example.

23:41Now, companies that don't have the ability or the desire to make some of their algorithms completely transparent or to share kind of a lot of their API algorithm, and there's a very long list of requirements, those companies are not going to be able to operate in Europe without the threat of being fined. Now, some companies are probably going to take the fine. We've seen that with GDPR. But GDPR has been kind of effective in some ways because we're seeing some companies really do have to change their practices. Whether that's just a banner that says, you're now entering a GDPR zone, here are the different privacy laws that apply, that's still something.

24:20So I think we're going to see that a lot. We are going to see, I think some difference in the types of companies that are able to do contracting with the government, whether it be the federal government or European government, based on different kinds of requirements. We're already seeing that a little bit. I mean, there's no U.S. law on AI right now, but there's a lot of drafts out there. The White House published their blueprint for an ethical bill of rights, which I think that companies are not necessarily following, but they're paying attention and they're noticing them. And also, it looks a heck of a lot better for a company to say, oh, I am actually providing a right to notice when someone is being targeted by my AI system.

25:01I think, you know, that kind of goes in some ways. So I think we're seeing a little bit of that. It's going to be really interesting to see where the U.S. ends up finally coming out on AI regulation. My guess is, unfortunately, that so many of these things are going to be watered down because there's so many different viewpoints on every side. And I think it might take a little while for us to really have something that is changing the landscape in a substantial way. Do you think the U.S. will just kind of default to copying the regulation out of the EU, for example, in a similar way that the EU came out with GDPR and now essentially by default, most American companies follow that because they do business in the EU?

25:44Do you think that it'll just kind of like be adopted or do you think there's going to be some fundamental differences the U.S. takes? We're still figuring out what the final version of the EU AI act is going to look like. I think it's going in a positive direction. from the first draft I saw a couple of years ago, it's really come a long way in bridging some of our concerns with the US government and concerns from US companies. One of the initial concerns about the UAI Act was that it was very stifling for innovation, and they backed off on some of those provisions a lot. So I think that, yes, we might end up in that situation.

26:18The other one that I'm personally biased about and excited about is the Council of Europe AI Convention, because I was kind of doing negotiating on that. And once again, I think that if that treaty comes out in the next couple of years, the U.S. would hopefully sign on to it. That's never a guarantee, but hopefully sign on to it. And that might become a kind of a de facto law in the way that we follow international treaties. I do think that these regulatory advancements are going to be made abroad before they're made at homes. Okay. Yeah, that's very interesting. Okay, one thing I would love to ask you along this vein, because I think this is probably one of the areas that has gotten the most criticism regarding AI and regulation.