AI Human Impact: Toward a Model for Ethical Investing in AI-Intensive Companies

Other speaker: Welcome to the AI Asia Pacific Institute podcast. The rise of AI presents important legal and ethical challenges for society. In this podcast, we invite leaders from different industries and creators of new AI to debate the big questions. This is the AI Asia Pacific Institute podcast. Welcome to today's podcast. I'm joined today by James Brusseau. He is the author of books, articles, and digital media in the history of philosophy and ethics. He has taught in Europe, Mexico, and currently at Pace University near his home in New York City. As Director of Data Ethics Sites, a research institute currently incubating at Pace University, he explores the human experience of big data technologies. Welcome, James. It's great to have you here today.

James Brusseau: Thanks. I'm happy to be here, Kelly.

Other speaker: So I feel a large focus of our conversation today is going to be around the recent paper you've produced, which involves investments in AI. So you suggest a new model of looking at these challenges and perhaps a good place to start by inviting you to share what is AI human impact?

James Brusseau: Right. So this is a strategy for rating AI-intensive companies for investment purposes and rating them in human and ethical terms, as opposed to traditional financial terms. Also, it's a structured and rigorous way of examining companies. So in that sense, AI human impact resembles traditional investment scoring. What it adds is humanitarian and ethical information to supplement the traditional financial tools. So it's an attempt to add information for investors so they can get a fuller picture or view of the kind of companies that they would choose to invest in and then just very quickly sometimes people ask me well what is since this is aimed at this is a system which is aimed at AI intensive companies people sometimes ask well what counts as AI intensive by that I don't mean companies that exclusively do artificial intelligence, but instead I mean any company that has artificial intelligence at the core of its function or operation. So this would include well-known companies like Amazon and Tesla and so on. But I think perhaps more importantly, there are a whole slew of startups and new growing companies in fields like insurance, finance, healthcare and elsewhere that function to a significant extent through artificial intelligence. So the idea is to select or weed out those companies, just take that group of companies and then evaluate them with a set of customized criteria which is tailored to artificial intelligence and the kinds of human dilemmas that arise within that particular part of our existences.

So maybe just, I could just give a very quick example. Of course, as we're talking about artificial intelligence, privacy is going to be an important criteria for evaluation. So as a thumbnail sort of beginning, we could say, let's just take two companies. We could take, say, Apple Computer and a company that's been in the news recently, TikTok, which owned by ByteDance. And we can say, well, in terms of privacy, Apple does fairly well. We could investigate or gain information from the assessment list for trustworthy artificial intelligence that's been produced by the European Commission. We could look at their cybersecurity expenditure, how we could look at public records about Apple and privacy and data protection and so on. And I think probably we would come to a conclusion that Apple does fairly well on that front. If we did the same thing with TikTok or with ByteDance, we would come to a very different conclusion. So within an AI human impact system or rating, Apple would score positively in privacy and TikTok would score poorly. And significantly, that measure is entirely distinct from a financial measure between the two companies. It may be that in terms of price to earning ratio or revenues or so on, that one or another company is preferable. But that kind of financial rating runs parallel with, though it is entirely distinct from the kind of rating that we want to do with AI human impact. So that's a starting definition of what we are doing.

Other speaker: And how do you think you connect with the history of investment?

James Brusseau: Right. I think that probably one thumbnail way of talking about this is as a sort of ESG theory for artificial intelligence. I suspect that many of your listeners will be familiar with environmental society and governance investing, which is a movement that began maybe 20 years, well, further back, but began in earnest, say about 20 years ago. And this is an investment rating system promoted by companies like MSCI, Sustainalytics, State Street, to a certain extent True Value Labs. There are all kinds of companies that work out what is called environment ESG investing, or sometimes it's called responsible investing, sometimes it's called sustainable investing. And what they have been doing over the last two decades is they have been trying to set up a parallel non-financial way of thinking about companies in terms of their impact on, as the name states, the environment, society and governance. So, for example, if you're going to rate a company in terms of its environmental impact, you might talk about its carbon footprint or the kind of toxic waste it produces or doesn't produce. If And if you're going to talk about its social impact, you might think about a company, let's say Philip Morris, a cigarette manufacturer. That company might score poorly in terms of its social impact. And then governance includes things like executive pay, international bribery, and so on. So there is this established area of investment that exists to cover non-financial aspects of active companies.

What we want to say is that that idea fits well to artificial intelligence. However, the criteria that are being used for traditional ESG investing simply do not fit with artificial intelligence reality. Just for example, think of the environment. No artificial intelligence company is going to produce the kinds of toxic waste that a strip mining company would produce or produce the kind of environmental damage that a clear-cutting lumber company would produce. Instead, the kinds of questions that surround artificial intelligence have to do with issues like privacy. And that's something, of course, 20 years ago, no one cared about the privacy of the consumers.

No one was even interested in the personal information of those people who were buying products and services. If you think further back to some of the original industrial corporations, what was it that they used to say about Ford Motor Company? That you could have whatever color you want as long as it's black for the Model T. So those kinds of companies, those kinds of mass-producing companies, they were not interested in specific consumers. They were interested in segments of, say, middle class consumers who lived in the suburbs or something and needed a car. But AI is completely different. AI is all about personalization, the force and the success of artificial intelligence, whether it's in health care or insurance or in retail. All of those things stem from the ability to gather very very specific details about individual consumers, and then connect with them immediately using those details.

So in general, the point that I'm trying to make is that in terms of the history of investing, there is over the course of the last 20 years, this development of non-financial, say more social or individual or ethically oriented investing, which has had some success in the marketplace. And what we are trying to add now is this idea that there are this new set of companies which could also participate in that way of thinking about investments, but in order to do so, we need completely new criteria to evaluate those companies. So I think that's how AI Human Impact connects with the history of investing. It takes that old ESG foundation and changes the criteria for investing.

Other speaker: Yeah, and I feel what you're saying about the history of privacy there, for example, with consumers, I feel that recent events have really brought that to light, right? People didn't used to think about that before Speaker uncertain: And now we're being forced to think about that. Speaker uncertain: Scenarios like Other speaker: Scenarios like Cambridge Analytica really forced us to consider what the impact of these companies are having on the population.

James Brusseau: Right, or another example is that the question, I think it seems to me that this is one that will become very forceful in the next five or 10 years in healthcare as we become increasingly capable of discerning what kinds of infirmities you and I will suffer from in the near future from data which can be gleaned from us like electrocardiograms and so on, it's going to become very difficult for traditional kinds of insurance companies to continue functioning because they are going to know. It's going to be very difficult for insurance companies to continue functioning in the traditional way because they are going to know quite a bit more about those who are applying to be covered. So the kinds of concerns that you and I have as individuals who perhaps want to buy health insurance, we never had to worry about protecting our electrocardiogram information or our blood pressure information or so on. But now if we want to buy life insurance in five or 10 years, we better be pretty careful about what data we allow who to see. So that's a very, right?

I find that example works pretty well because that's something that affects us on a very human level, right? This is biology. This is our health. This is our lives. Cambridge Analytica, which is also an excellent example, but it's somewhat abstract in the sense that it happened to other people and it occurs in some other place and so on. But health is very hard-hitting. And I think that many of the issues we want to talk about in AI human impact and in AI ethics, one of the first places where they're going to have resonance is with respect to our physical, biological bodies.

Speaker uncertain: Yeah.

Other speaker: And also, we already had examples of that going wrong, right? So we had recent examples of companies using DNA, testing DNA, and then that data has been breached, right?

James Brusseau: Right. Right. And then there's a whole, I had curious, I just had this debate recently with someone who was selling artificial intelligence insurance and what they're selling is highly dynamic insurance. That is it's insurance that tracks you on your iPhone and is able to determine things like gathering from various sources, is able to determine things like you are taking a ski trip and you are about to go down the double black diamond run, which is very, very dangerous for some people. Broken legs happen all the time on these runs. And what they were trying to do is they're saying, look, what we can do as an insurance company is at that moment that you're at the top of the hill, we can send you a notification on your phone saying, if you take this run, your life insurance premium goes up a little bit. But if you say, forget it, I'm going to go on the green run, the easy run, then your life insurance premium goes down. And what he was saying is, and he had a very strong point, he said, look, this is good. He said, this gives people more control over the insurance premiums they pay and over their finances. And he was right, and he was right about that.

But on the other hand, it's also true what I said to him in response is it's also true that the reason we get insurance in the first place is because we want to live our lives vigorously. The reason I get health insurance is just so that I can do the double black diamond run, because I know that if I break my leg, I'll have insurance to pay for it. So if we have this notification system, which allows me to raise and lower my premium, it also means that I'm going to think twice before the double black diamond, and maybe I won't go. So maybe this artificial intelligence advance, which on the one hand gives us more autonomy and freedom over our finances it also limits the vigorousness the vitality of our lives so that's the kind of I think that's the kind of situation which will be the most interesting ones where it's not so much that there's a clear right or wrong or good or bad but there's a change in the dynamic that we will need to make sense of yeah it's

Other speaker: It's not a straightforward debate, right?

James Brusseau: Right, right.

Other speaker: So what are the criteria principles of AI human impact investing? And what are the ratings that are based on?

James Brusseau: Right. So this is the key idea. As opposed to the ratings of the ESG investors that use the environment and society and governance and then they use this further criteria many of them use criteria from the united nations 17 sustainability goals so that's on one side what we use well let me say first we were talking before the interview about how quickly this field of AI ethics has been shifting and growing And there have been over the last five years or so, loosely, about 80 sets of principles, ethical principles that have been published by governments, by institutes, by private institutions and so on. So we have a lot of information or a lot of options when it comes to thinking about the values or the principles that we should apply to artificial intelligence. And the group that we decided to use is very similar to the recent production by the European Council, the Guidelines for Trustworthy Artificial Intelligence. And it's also very similar to the recently produced German Commission on Data Ethics. They produced a white paper which they called an opinion.

So the kind of criteria that we are using are right in the mainstream. And we divide those criteria up into three parts. There are personal values or ethical rules, there are social values and ethical rules, and then there are technical values and ethical rules. And for us, the personal rules or ethical values are autonomy, dignity and privacy. And then the social rules or values or ethical guidelines are fairness, solidarity and sustainability. And then the technical rules are performance, that is how well the machine works, safety and accountability, which also includes explainability as accountability is the idea when something goes wrong or right for a artificial intelligence.

I saw just today, I believe it was a Tesla, another Tesla accident, though I don't want to say that there are a lot of them. Actually, there are not. The Tesla autopilot function does seem to work quite well. But I believe it was just today that there was another one of these where the person was watching a movie on his iPhone coming home from the late shift at the hospital and his Tesla side swiped a police car. So, something went wrong there. And in order to determine what went wrong or in order to attribute accountability, we need to have something called, well, as you're familiar with, explainability. That is, we need to have some ability to trace how the autopilot and the Tesla car made the decision it did to go left instead of right and so on.

So those are the nine principles that we use to grade companies. We say, look, let's take a company like Apple or like Tesla or a health company or an insurance company like Prudential. I mean, your Prudential, by the way, there's two Prudential insurance companies and the Prudential in Asia is quite invested in artificial intelligence. So you take companies like that and then you evaluate them in terms of what they do for the autonomy of their users, what they do for the dignity of their users, what do they do for privacy? What do they do for the fairness of society? What do they do for solidarity in society? How do they rate in terms of sustainability? And then the technicals, I say, how do they perform? How safe are they and do they have good accountability? So what we do is we try to rate companies in terms of each of those nine criteria. And there you have some sort of model

Other speaker: Model where people can look at and quickly be able to perceive what companies are doing well

Speaker uncertain: And what companies you should probably avoid. Exactly this is the goal and then

James Brusseau: I just want to add one quick point here about that last comment you made about the companies to invest in those to avoid. I do want to say this about what we're trying to do. We are not trying to paint companies in terms of good or bad or right or wrong so much. What we are trying to do is provide information for individuals, sort of as you pointed out in that first part of the question, provide comprehensible and clear information to individuals and investors as to how these companies perform in each one of these sectors. And what we want to say is that then investors, the value we're trying to add, what we're trying to say, the reason investors should pay attention to what we're saying is because we can give you as an investor options. We can give you a better understanding of the companies that you want to invest in. And that better understanding lets you work or invest out in the world just like the companies do. You as an investor can pursue profits by taking the information you have and choose to invest in one or another company. It's not so much that our effort is to sway investors toward one or another company, or to create new regulations, or to create new laws. It's much more a libertarian project in this sense. What we are trying to do is give investors clear information so that they can make their own decisions in the marketplace.

And then, so that one preface, now I can get back to the question you asked. So what we try to do is we produce comprehensible scores that allow people to look at a company and understand how they perform in each one of those nine categories and overall. Frequently, it's going to happen that there are trade-offs.

For example, one of our technical categories is performance, and one of our personal categories is privacy. And in healthcare, for example, it's really difficult to get both. You're going to... Let's use a different example. Let's use retail. Let's just take a simple thing like that it's 7.30 here in New York City where I'm at, so perhaps tonight I will watch a Netflix movie and I will ask Netflix to suggest something to me. The suggestions that Netflix gives me depend quite a bit upon my revealing my personal interests to them. So if privacy is defined as control over access to your personal information, the problem you have with Netflix is once I tell them, well, I like this kind of movie and not that kind of movie, and I have this age, and I have this family situation, and so on. The problem I have is once I turn that information over, well, now I've lost control over it. So I've lost privacy. So for Netflix to perform well, they are going to have to be fairly demanding in terms of privacy. So frequently, as I say, we're going to get trade-offs. And as a result, We try to do two things with our reports.

First, we give clear scores for each company in terms of each of these nine categories. A consumer or an investor might say, look, privacy is very important to me. It's more important than performance. I'm willing to sacrifice a little bit of performance in order for individuals out in the world to maintain control over their personal information. They'd be able to look at a company, let's say Apple, and see how Apple performs in terms of how Apple is rated in terms of privacy and in terms of performance and see if that fits their own values and their own way of thinking about the world.

So examples of this, there are simple examples like autonomy is one of our categories. And it's easy to see how, let's say, let's take an example from Facebook. Facebook is well known to use dopamine spurts, right, these rewards for likes and so on to try to increase engagement, right? So they're looking at the kinds of postings you do. And if you haven't posted for a while, then when you post right away, they will find some way to give you some likes. So that encourages you to post more and so on. So something like that is going to result in a low autonomy score, right? Because it's not that the user is determining for him or herself when and why he or she wants to post. Instead, Facebook is manipulating that for their own benefit. So that would be something that, so that's an easy one. Companies that do that, they use what a computer scientist named Arvin Narayanan has called a dark pattern. Companies that use these kinds of distribute likes and so on in order to manipulate users, they will score very poorly in terms of autonomy. As I said, that's easy. But then there are others which are more curious.

For example, another of our categories is dignity. And a case that we were discussing just the other day was a dating website named OKCupid. And one thing OKCupid did is they ran an experiment where they told people who their own data implied were not good matches. And they told them, they told these two people or these groups of people, well, you are a good match. And then they monitor or followed their online activity. So usually, of course, on the dating sites, right? You exchange all the messages and then the company, whether it's Tinder or OkCupid or so on, they have some sense about how the relationship goes depending upon how many messages you interchange and kind of what those messages sound like. And if you have dates and come back and have more messages and so on. So OkCupid ran this experiment where they put together people who they thought were good matches and people who they said they thought were not good matches, but they told them they were good matches to see if the results would be different, to see if people could, what would the phrase be? The power of positive suggestion, right? To see if people could be convinced that they should like this other partner. Then they got the results and it turned out that no, people really don't respond. People really do either like or not like other people and the power of positive suggestion doesn't work so well, a little bit, but not so well for dating. So we have this question. Okay, so they did that. And then there's this category that we have for evaluating a company like OkCupid, which is dignity. And dignity is defined as, dignity means treating other people as ends in themselves and not as means to an end or not as tools for someone else's projects or ends. And when you look at this, this looks like a clear violation of dignity. I mean, it looks like the people who are running OKCupid were just manipulating their own users for their own vaguely perverted interests, right? To kind of see what would happen and so on. So it seems very abusive in terms of dignity. But then they produced a blog post, and what they said subsequently was, look, this information that we gleaned actually helped us find true romances for our users. So it's sort of an interesting case, right?

On the one hand, it's very manipulative and abusive in terms of dignity. But if it's true that it does help their users find romance, commands, they use that information for that reason. Well then what kind of dignity score do we give them? And so we were talking about that and one point that came up is we did notice that OKCupid after they ran the experiment, they told people that yes, we did this and you were one of the people who experimented on it. So something like that, maybe you think in that case their dignity score is say neutral as opposed to what we initially thought it was going to be which is terrible because they were doing these things. So something like that. So I've given a couple of examples, right? The dopamine example with autonomy, the dignity example with OkCupid.

I could give other examples, but something like that is the way the analysis works. And then as I say for each company, then we can attach a score for each one of these categories as well as a total score, because ideally, I mean, it would be possible. I said before that we're not so much looking to separate good and bad because that's impossible in the sense that frequently there are trade-offs. But ideally, at least we could imagine the possibility that there would be an AI company that would score well on everything. So in light of that, we also provide a total score. But the main value we're adding is we are allowing investors to make their own decisions about what values are important to them by presenting comprehensible ratings of these companies in terms of customized criteria for artificial intelligence.

Speaker uncertain: Great.

Other speaker: Yeah, I feel that's such a model. Speaker uncertain: Would, Other speaker: It would essentially empower not only consumers but investors as well to understand a lot of these principles and values, which, you know, we still have a lot of awareness to build in the industry, right? So there's still a big gap in terms of understanding. And I was mentioning to you before this conversation that I had this week with the founder of Affectiva, where she was sharing the challenges of raising funds and staying true to the technology. And one of the key points that came out of that conversation was the need for consumers to really understand this, ask the right questions so that they can start making a decision, right? Because that's how trust would be built. Essentially people expect something that companies will be, you know, encouraged to pursue, right? Just like you used the example of Apple before, and I feel Apple is, because they took a different path, slightly different from the other big tech companies, what they have now is a lot of people tend to trust them more. And I feel we're still sort of going towards that way where these companies are now being rewarded for building that trust. Other speaker: And in a few years, perhaps that difference will be stronger, right? Speaker uncertain: Right. Where you on the other hand, you've got companies that breach that trust and it's very difficult for them to build that again.

James Brusseau: Right. Such as Facebook. Right. And then, of course, a key element of trust is deciding what we are going to trust them to do. Right. Are we going to trust them to be a high performing company or to be a very safe company or to do well with privacy, user privacy and so on? So I think also there's that this aspect of trust only occurs when we can specify the specific ways in which we look to companies to do well or poorly.

And then I think also, it was easy five years ago when there were only a few major artificial intelligence companies in our lives to develop ideas about which ones we wanted to be associated with as consumers and as investors. But it's going to become extremely difficult as this flood of artificial intelligence applications takes over every aspect of our lives, from the transportation we use to the health care we have to the movies we watch and to the people that the jobs we have linked in, romantic partners and so on. And all of these across this entire range of our experience, I think that for that reason, there is an opening for the kind of project that we are undertaking here, which is, again, a way of making comprehensible the ethical aspects of these companies for consumers to make their own decisions about where they want to, investors primarily, but also consumers of these products to make their own decisions about where they want to put their investment money or their consumer money.

Other speaker: Yeah. Like you were saying before, we do have a lot of principles available and values, right? We're not sure that they're available all around the world. They might be slightly different depending on the framework. We still have a large gap reaching those values to practice.

James Brusseau: Right, and it's a lot of work, right? So I'm just going back to that OKCupid example. In order to grade a company in terms of dignity, it's easy to say what dignity is, but to actually assign a numerical objective value, Because that's very crucial for us. We want a rigorous, objective system. This is not sort of vague discussion. These are analytically solid ratings that will fit well with financial ratings in order to produce those on the human level because that's what this is about, right? The human experience of these things, that takes a lot of work. That takes a lot of thinking through things like, as I say, this question for OKCupid, well, are they really abusing the dignity of their users if they are helping their users find romance? It's not clear.

Other speaker: Yeah, yeah. And I think you touched a little bit on this question before, but who might be interested in investing and accessing Speaker uncertain: This ratings?

Well, I think primarily our goal is people who are professional investors initially,

James Brusseau: Because I think that one important aspect of finance generally is that that's one place where There is enough money simply flowing to pay for this kind of project because we believe that this project will, it adds value to investors because it gives them options, helps them understand their own investments. But we also believe that it's likely that over the medium and long term, companies that to score well in terms of AI ethics will produce better results. So I think that we can make promises to financial companies that are rewarding both on the ethical and human level, but also on the financial level.

And this is always a challenge for ethics. That is typically when people look at ethics in AI or in almost any field, they feel like ethics will detract from value or it will stop us from doing things or it will restrain or block innovation. We don't want to be any of those things. We want to promote innovation. We want to promote financial gain. And we believe that these principles well understood and presented to investors and to consumers will do those things That the companies that adopt, that work in line with these principles will, in fact, advance more rapidly, gain more investment and ultimately more profit in the medium and long term.

So it is, at least initially, it is the idea that it would be the professional investment area that would be our aim. But we hope of course that consumers and people who are listening to this program who are perhaps not investors would be interested in what we're doing yeah so it's really making

Other speaker: The connection of you know helping people to see ethics is actually you know becoming financially rewarding right right and encouraging I think investors have a large role to play in encouraging ethics and AI. And I feel the model is opening that path.

James Brusseau: That's what we aim to do, yes.

Other speaker: And when we have, you have here in your paper some predictions that, you know, AI will be reaching, you know, there's a calculation in the industry they'll be reaching about 20% of electricity, towards electricity by 2030, right? Right.

James Brusseau: So that gives an idea. Obviously, there's nothing more hopeless than to try to predict the future. But if anything remotely like that is true, then so many of our human questions and ethical questions will be channeling through artificial intelligence that we will need to have just to be happy as individuals out in the world to understand our existence and what is happening around us. We will all of us need to be conversant in at least rudimentary artificial intelligence ethics so I think that would be another way to think about a contribution that we at least hope to make right this is not something which is going to be limited this is something that which is going to overflow our existences over the course of the next generation

Other speaker: Yeah. In a way, it's some sort of foundation for how this progress should be made.

James Brusseau: How people can live and think in that kind of world.

Other speaker: Yeah. And especially like, you know, you made a note there about how these encourages innovation. And I think it's a great way of looking at this because on the other hand we have you know perhaps possibly upcoming regulations which if poorly designed will likely delay innovation and right and so and we have that challenge of you know how would we even design a regulation that could globally you know be aligned with these principles and values is in the right place I think it's much more challenging getting that right than getting a Speaker uncertain: Model like the one you're suggesting here right which is another way into the project I think

James Brusseau: Which is what we are another way of talking about what we're trying to do is we are trying to help developers engineers technical experts we are trying to help them understand their work on the human level in terms of privacy in terms of dignity in terms of the way you and I live. And I think to the extent that they understand that and see the kinds of ethical issues that arise from their work, to the extent that they themselves are aware of that, they will be able to avoid pitfalls. And it's in that way that the advances will be encouraged, not restrained by this kind of AI human impact project. To the extent that we can help designers and engineers understand the human impact of what they are doing, they will steer their innovation more effectively. And therefore, they will be able to innovate more faster. So we do believe in that more innovation, faster innovation, but only because the designers understand the humanity of what is happening around them. That's another way to talk about what we're doing. We're trying to provide insights into the humanity of any specific artificial intelligence and how it works.

Speaker uncertain: Very good, well thank you so much James for taking the time to talk to us today.