The Black, White, and Gray of AI: Ethics, Risks, and Regulation with Guest Cyrus R. Vance Jr.

Voice over (00:01):
Amicus Curiae with host Daniel K. Wig, a podcast series from the New York County Lawyers Association. Join us monthly for candid, useful, inspirational and entertaining conversations with leaders of the bench and bar and those who serve the legal profession.

Daniel Wiig (00:22):
So Cyrus, let me ask you this. I guess it was close to a year ago, but does it surprise you or whatever reaction you may have that when Pope Leo the XV was elected, he raised AI as one of the preeminent issues that concern him and his role as Pope. Did that surprise you? What are your thoughts

Cyrus Vance, Jr. (00:42):
On that? Well, it’s interesting that you asked that question because I attended and spoke at a conference at the Vatican, at the invitation of the Pope and the Vatican, and I can’t remember the date, but let’s just say six months ago. And I think the conference was convened to talk specifically about AI and its challenges. And the Vatican and the Catholic Church, I think in many respects, because it deals with so many children and the responsibilities of interacting with those children, I though it was very well-timed. And so if the question is, what do I think about the Pope becoming interested in AI? It doesn’t surprise me. And I think frankly, he is an important voice. And in that regard, Daniel, I propose in our conversation what I thought was a real opportunity for the Catholic church to engage with the issues around child safety and AI.

(01:36):
And it was to, I think a lot of this when you look forward is an education question. How are we going to make young men, women aware of AI and how to use it intelligently and how to avoid using it unintelligently that the Catholic church and its parochial schools, which I think are collectively the largest non-governmental school systems in the world. So we’re talking about millions of kids all over the place, that they could develop a curriculum for very young lower school, middle school, and high school to essentially reach directly to millions of young men and women across the globe, across cultures, across governments to talk about responsible use of AI. So I think the Vatican has an important role to play and I don’t think the Vatican’s being political about it. I think this is an opportunity to assess there are risks around child safety.

(02:35):
I think we all agree with that. I think the companies that create AI and all that can do understand that there are those issues as well.

Daniel Wiig (02:43):
So we will get into this and more. I’d like to welcome our listeners to Amicus Curie, NYCLA’s series of casual, comfortable, and friendly conversations with leaders of the bench and bar and those who serve the legal profession. I am absolutely thrilled today. We’ve been planning this for a while to welcome Cyrus Vance Jr. To the show. Cyrus is the global chair of Baker McKenzie’s cybersecurity practice and co-chair of the firm’s North American government enforcement practice. He co-founded New York City’s Cyber Critical Infrastructure Task Force, establishing one of the first public-private partnerships dedicated to mitigating cyber risks on critical infrastructure. Cyrus also founded the Global Cyber Alliance, a global nonprofit consortium providing expert guidance on cyber risks and prevention to members worldwide. He served three consecutive four-year terms as the Manhattan District Attorney. He’s a graduate of Yale University and the Georgetown University School of Law.

(03:36):
He joins us today to talk about the moral issues concerning AI, accountable leadership in tech, concern for public safety and responsible partnerships with government. Cyrus, welcome and thank you for joining us.

Cyrus Vance, Jr. (03:48):
Thank you, Daniel. I’m very happy to be here.

Daniel Wiig (03:50):
And as a reminder to our listeners, Cyrus is sharing his views and not necesarily those of NYCLA, its officers, its directors, its members, or me for that matter. So I think there are people out there, and I’m not just talking about older folks, let’s say, who aren’t involved with this, but even others who don’t really understand what is AI? Sort of like we were talking about the Vatican and maybe quoting from Genesis in the beginning, let’s talk about what exactly is AI? And before you answer that question, let me throw something in there. Would you consider Google a primitive form of AI? Meaning if I type in Cyrus Vance Jr. In Google, I get all these websites or wherever you’re mentioned, your bio, et cetera, from maybe perhaps different areas of your life, that might be a primitive form of AI. So how would you describe AI to someone who has no idea what this concept is?

Cyrus Vance, Jr. (04:41):
AI is obviously a tool that is being applied usefully to users all over the globe. And essentially it runs on algorithms which are code and it is not a brain, it is not a human. It is a system, a computer system that has, as its foundational base, billions of facts, documents that have been loaded into, let’s assume it’s memory. And what it does in essence is when you type in the question, it responds to that question essentially one word at a time. And I think that some of the magic of it is it makes the decision about what word is next based upon its analysis and evaluation at the machine speed of all the data that has in it and then puts down what it believes is the next most probable word to answer your question. So what I’d say, Daniel, is it’s a computer program.

(05:49):
It’s enormous. It’s not human, which I think is important always to remember. And none of us, certainly myself, really understand exactly how the algorithms work in the AI program, which is I think what is one of the interesting and concerning aspects of what AI is and how we react to it.

Daniel Wiig (06:13):
Believe it or not, I used ChatGPT for the very first time recently, really to prepare to chat with you. And I asked it, I’m talking like it’s a real person, like you said, it’s not a real person, to write my bio for me. And I didn’t feed it any information. It found information about me, I presume, on the worldwide web and within seconds wrote what could be viewed as a professional bio that I would submit if someone wanted to. But it screwed up because it actually said that I was an assistant district attorney in your old office for a few years, which is not true. So obviously it pulled incorrect information or was confused. So I guess in that case, that’s what it did. It’s super fast lightning speed called webpages and put a bio together for me using whatever this algorithm is that searched for me.

(07:01):
Is that fair to say?

Cyrus Vance, Jr. (07:02):
Yes, I think that’s fair to say.

Daniel Wiig (07:04):
So talk to us a little bit about… We’re going to talk about the good, the bad, and the ugly of AI, and then we’ll get into your views on how we should, I’ll use the word regulate. Maybe that’s not the right word. What are some of the good ways that AI is being used or could be used? I know, for example, you mentioned politics before, not to get political, but I believe the Trump administration is taking a very positive view of AI. They’re sort of emphasizing, I guess, how this could be a force to assist in commerce, assist in business. Talk to us a little about what you see as the good use of it in line perhaps with that thinking, and then we’ll get into the not so good uses of it or the concerning uses of it.

Cyrus Vance, Jr. (07:45):
The good uses of AI I think are many, and I certainly don’t know them all, but with a knowledgeable coder and someone who understands how to prompt a system, it has the ability to answer questions with incredible quickness, with relative accuracy. And as applied to science and medicine and other very complex and broad subject matters, it has the capacity to produce answers to questions that have hitherto not been achieved by the human brain. So applied, it can solve problems around health, around development of materials, around all sorts of issues that are useful. And my use of it often is personal use is to prepare when I am speaking somewhere on the history that relates to the topic. And so my use of it is relatively simple, but it has very complex uses that are also very useful.

Daniel Wiig (08:50):
You started mentioning before. We don’t maybe know who, maybe I think you answered who we don’t know, but maybe who’s creating this? In other words, who’s the mind behind these algorithms that makes it easy for you to pull a history of a topic that you’re talking about or is able to teach it to search the internet within lightning speed and draft a bio of someone based upon public information? I presume these are computer programmers. D we know anything about these folks that are working for Claude, ChatGPT or Elon Musk is, I think it’s pronounced Groke if I’m not mistaken. What do we know about these obviously very intelligent men and women?

Cyrus Vance, Jr. (09:30):
I will say I’m not sure I know exactly who is on the keyboard in each of these companies and creating the program that we access and use for all the reasons that we talked about earlier. But there are many companies obviously that are now using AI, creating AI programs for specific uses, industry, medicine and the like. Who those programmers are, I don’t know. But we do know that the big companies that have taken the lead in this have legions of coders. So right now it seems to be a competitive situation with giants and yet there are probably hundreds of other AI companies in quotes that are using large language models that they have created to be essentially the pot of information that they search. And the program then creates the answer to your question by using a very program-related function where it really puts on paper one word after the next, depending upon what the probability is of the correct word following another.

Daniel Wiig (10:34):
There’s also what I’ve called the dark side of it where it’s being used to harm or to create false information out there. We’ll get into that in a second, but there’s also this sort of gray area where it may be good, it may not be good. I’d love to get your thoughts on this. So I’m sure you’re familiar with the story, and I may get some of the facts incorrect so you can correct me, of a lawyer who drafted a brief for a court in the Southern District of New York and it was filled with fake cases. He was citing to cases that never existed or didn’t exist, I should say. And I believe the court sort of figured this out doing their own independent research, making a decision on presumably this motion. That’s a problem. I mean, it may not be evil, harmful to children, which we’ll get into, but there obviously are issues where either it goes too far or it’s being used improperly or it’s being used dishonestly.

(11:27):
Talk to us a little bit about your thoughts, what you’re seeing, and what do you think a better use of this new tool is?

Cyrus Vance, Jr. (11:34):
Well, the technology that exists now is value neutral. Its goodness or its badness depends upon who’s behind the keyboard and who’s trying to use AI and for what purpose. AI itself generally speaking, I think is value neutral. If you are a criminal and you are trying to find a way to defraud senior citizens or citizens of any age with a program that reaches out to them personally and ultimately enables AI to obtain and to influence those individuals to provide money to an organization that AI has fabricated or any number of things, it can be used in a way that is harmful. And in terms of its good use, the program obviously because of its ability to work so quickly and to have such power in its application has enormous positive uses as well. There are challenges I would say about there are areas where we’ve seen that AI has to be very carefully guarded.

(12:37):
Predictive policing, for example, uses historical arrest data to forecast where a crime would occur. And yet the problem is with predictive policing is that the arrest data reflects where the police are deployed, not necessarily where crime happens and algorithms amplified existing enforcement patterns rather than correcting them. Risk assessment, and these are all related to my work previously as the Manhattan DA. The Arbor Foundation’s risk assessment tool called COMPASS are used in bail and sentencing across dozens of jurisdictions. Yet there has been concern by organizations who have investigated COMPASS and how it’s used that COMPASS was significantly more likely to flag defendants who were Black as high risk and the cool then therefore encoded the biases presented in the historical data they’re trained on. And face recognition will be the third. It’s used extensively by US law enforcement, but there is no federal regulation and multiple documented cases of wrongful arrests have been noted based on misidentification, which often disproportionately has affected Black men.

(13:47):
So those are three areas where AI documented and expressed concerns about its accuracy and its application have been publicly raised.

Daniel Wiig (13:58):
I was going to ask you about the bias and you beat me to it.

Cyrus Vance, Jr. (14:01):
Yeah, but from my own experience running a major prosecutorial office, the appeals of these tools are real. Prosecutors and judges, for example, face enormous caseloads and the imperfect information when they’re trying to make a decision about a case. AI promises consistency and efficiency, but the lessons of the dangers of its application are sobering. You talk about bias. That’s one of the things that is very important to realize AI and its coders and then the consumers need to understand. The AI systems do not often correct for systemic inequality based upon the information they put in. They sometimes formalize it by having biased information be part of the code that AI searches. So you have to, and I don’t fully understand all the details, have to have rigorous ongoing audit requirements to avoid entrenching disparities while giving the judges the sense and a veneer of scientific objectivity.

(15:03):
Transparency really matters in terms of the concerns. Defendants and their lawyers have really no ability to examine the algorithms that produce the risk score used in their case. And this is a due process issue in criminal justice that is very important. And the third problem issue is absence of regulation in the US itself. The US approach to application of AI has been largely permissive. Companies deploy first, and if there are issues around transparency or bias, that gets litigated later. The EU’s AI Act by contrast classifies law enforcement AI as a very high risk operation and requires conformity assessments before deployment. These are opposite philosophies, but the examples I’ve given I think are examples that are current and real and judges, prosecutors, defendants, anyone has to understand that AI has potential for inherent negative biases and concerns about the outcome.

Daniel Wiig (16:14):
We’ll get into regulation in a bit and your thoughts on that, but I want to touch a little bit about what I consider to be the really dark side of AI. And I’ll use a real life example that we discussed with the prior guests on this show, and you’re probably familiar with it and I may get some of the facts inaccurate, so apologies in advance. Case in New Jersey where a high school boy took a photograph or used a photograph of one of his female fellow students. She was underage, under the age of 18. He used some system, imposed her face on a woman in a pornographic video, created a video where it looked like this underage girl was engaging in a pornographic video. And then he disseminated that to his school, to his fellow classmates. That’s clearly probably don’t get much darker than that.

(17:05):
And then we’ve also seen these, and they’re rampant I think on Instagram and X and all these other places where they use these fake videos. There was one we spoke initially about the Pope. It was the prior P, Pope Francis. There was a picture of him in a white puffy wrapper jacket, which I saw, I though it was real. Maybe he went, I don’t know, he went skiing, though I don’t think he had the ability to go skiing. He was certainly not in his later years. But this is a dark problem. When I say dark, I mean there’s almost, again, we began talking about Vatican and the Catholic Church, almost like an evilness to this where you’re going to put a young girl’s image in a pornographic video and send it out to your classmates. Talk to us about what you’re seeing and what you’re aware of these real issues that send the term fake news has been coined in our lexicon over the past decade or so.

(17:53):
Talk about fake news. I mean, this is really bad stuff. Your thoughts on this?

Cyrus Vance, Jr. (17:58):
One point is that this is all still new to us in the sense that when I began my time as the Manhattan DA or even decades before when the computer technologies were accessible, this is new. And from the potential victim’s perspective, they really don’t necessarily understand the risks that exist when they are using either the internet or AI. So it can be a very effective tool for fraud and that is clearly one of the risks. There are problems with AI in broader context besides criminal justice. Consider benefits and welfare determinations, for example. These very consequential use where automated systems decide who qualifies for public housing or for disability or food assistance in healthcare. And even in immigration and border control, AI used in visa screening or asylum flame processing and border surveillance across the US are really important uses of AI where you can have negative impact.

(19:07):
So AI can be used obviously for good and it can be used intentionally for bad and its operation unintentionally still has risk for consumers and the public because the programs as applied may in fact embed risk or bias that exists, but people who are using the programs do not yet understand

Daniel Wiig (19:33):
It. So let’s now turn to what you’re really, I think, interested in is how you curb the abuses, bridging the private sector with government, something that you have experienced within the cyber area. In terms of the tech companies, we use the term accountable leadership. What would your recommendations be if you had their ear? What could they do to help curb what you and I are talking about as being some of the abuses or the biases or the quasi, if not criminal activities that are done using this technology?

Cyrus Vance, Jr. (20:11):
Let me perhaps talk about where I think lines should be drawn. And again, I will stick with my experience as a prosecutor. The use of AI, and I think the companies, those who use these technologies in government, I think there are some red lines that are really not the province of the company, but the province of those people who use AI and apply it to the public. First, Daniel, I think AI systems in law enforcement should not be the sole basis for a decision that deprives someone of liberty. A risk score from a judge’s decision informs a judge’s decision, but it should not replace the judge. Second, I think I’ll just put that out there. We should not be making individual risks assessments solely on AI. It is a tool. It is not a replacement for human evaluation and decision-making.

Daniel Wiig (21:17):
And you mentioned before, protecting children. There are efforts that the private companies are doing to help curtail or protect either children’s access or certainly using of images. What are your thoughts there?

Cyrus Vance, Jr. (21:32):
I think it is the responsibility of companies who create products that are being consumed by the public that the companies should be aware of and have a responsibility for applications and uses of their products that may cause public harm. Aside from technology, let’s just put AI and the companies that produce it, which most companies do incredibly good things certainly in their use of AI and other technology. Almost every other consumer product that is available in the public market has some regulation associated with it. You can’t buy a baby carriage unless that baby carriage has been tested for the anticipated problems with it. You can’t buy a toaster without the product being tested for safety and security. And so I think historically since 1990 to today, using very general numbers, I think technology companies have not been regulated and have not had to answer questions as other companies producing consumer products have.

(22:46):
So I think really this is two things. One, it’s the responsibility of the government to decide what regulation is required and what standard of safety is required with these products and their application for consumer use everywhere. So long way of saying I think it’s the government’s responsibility to draw the lines on what dangers may exist in the use of these products and then how those products should be adapted to eliminate or reduce that danger. And I talked earlier about Meta. I think it was Meta, but there have been concerns that the application of the programs have produced harmful results without certain regulation. The ability to access children at all hours of the day providing AI-created content automatically is obviously a big challenge. And that’s an example of where I think companies should, and I think many are trying to meet a standard of security for users who are young men and women.

(24:01):
And I think that’s their responsibility as well, not just the governments.

Daniel Wiig (24:06):
Wha efforts are you aware of that any local state or even the federal government is taking in line with what you think should be done? And maybe none are, but

Cyrus Vance, Jr. (24:17):
I think recently New York State and California have both passed acts focusing on technology companies and programs and applications that are being accessed by children to limit, for example, based upon the age of the young child, the ability to produce content for that child at certain hours of the night, not 24 hours a day, but to have a period where it’s not accessible, to not use data that they have on children in order to market products algorithmically to that child. So I think my answer to your question is there are ways that the government now is trying to put guidelines and guardrails around what is provided to young men and women and when and how.

Daniel Wiig (25:07):
So maybe to wrap up our conversation, a little bit about you. You’ve had a storied legal career, and I ask this of many people that I interview in the show. You’ve sort of made a jump, if you would, from being the prosecutor for three terms here in New York to being in private practice and working on mainly cyber that seems to be your bread and butter right now. How do you make that jump? I think that’s maybe a lesson for our younger listeners, people who make jumps from one area to another. What advice would you share to someone?

Cyrus Vance, Jr. (25:38):
I think every decision, professional decision about what I’m doing, I have made and the motive behind it has been to continue to do what I like to do and what I think is important to do, impactful to do. And I don’t mean that in any arrogant way, but I often get asked by young lawyers who are in law school or thinking of going to law school, “What should I do? What direction should I take?” And my answer is recognizing that there may be limits economically after someone graduates from law school with huge debts that have paid for their education. But putting that aside, my guidance is always quite simple. Go to the first job that you think you might love, that you like, love in the law. And if you pursue those kinds of employment, even if they are not the employment where you make the most money as a young person, I think it’s a first step that will be the foundation for a law career.

(26:41):
You will do a good job if you are doing work that you think is important and that you like. And that first job or jobs in law I think really helps define what you’ll end up doing later down the road 20 years later. At least in my case, I started as a young assistant DA and took that job with really very little information, not as thoughfully as many people do today, but I was very lucky because I really loved it. I found just the working with the people and the judges and trying cases was really interesting, emotionally satisfying, and I felt important. I don’t mean that in any arrogant way, but the work that I was doing as a young lawyer was important to victims of crime, to those who were accused of crime. And so I did a job that to my surprise I loved.

(27:30):
But having that positive experience really led me through in each stage of my career to follow a pursuit that I thought I would like, but that is important. And leaving the prosecutor’s office in Manhattan, there were many reasons that I went to Baker McKenzie, but one of them was the continuation of the public safety role and heading the global cyber practice with many talented lawyers in the firm. I felt that was in part a public safety mission, that safety was helping to protect companies from being attacked and being the victims of cyber-related crimes and individuals. I felt and feel today that guiding some of the largest companies on how to protect themselves is a public safety mission and it indirectly but it sometimes directly affects the communities that those companies work out of and helping keep the marketplace safe from cyber attacks and being able to guide companies on how to prevent them is really the motivation for me coming to Baker McKenzie, which is a global firm.

(28:49):
And I think the mission is, although not expressed directly in my mind, was a public safety mission.

Daniel Wiig (28:55):
And on that note, Cyrus Vance Jr., This has been a pleasure for me to chat with you about these issues which are to be continued. This is not the final word on any of this. So thank you, my pleasure. To our listeners, thank you for tuning in. I’m Dan Wig, your host, and that is a wrap.

Voice over (29:12):
Thank you for listening to NYCLA Amicus Curiae: Candidly Speaking with host Daniel K. Wiig, a podcast from the New York County Lawyers Association. New and previous episodes are available on Apple Podcasts, Android, Stitcher, and Spotify. If you enjoyed this episode, please subscribe, share, rate, and leave a review on your listening platform. To find out more or to join our incredible inclusive legal community, please visit nikla.org. We hope you’ll join us on the next monthly installment of NYCLA Amicus Curiae: Candidly Speaking. Be sure to follow our Twitter feed @amicusNYCLA for information on on upcoming episodes.