In a report that asks critical questions for our future, The Rockefeller Foundation’s AI+1: Shaping Our Integrated Future explores the nature of artificial intelligence, stressing the need for a regulatory framework to shape and monitor AI. The august power of AI must not be left to market forces, the report recommends, but must be a force that helps all of humankind.
To discuss the report’s themes, this webinar we discussed the following themes:
1. The report states: “As we reimagine a way forward, The Rockefeller Foundation is betting that AI will help rebalance and reset the future in a way that addresses current inequities. To realize that outcome, we must develop a regulatory framework to ensure its responsible use.”
“We need to reimagine an entire new rule-making system that guides AI towards society’s goals instead of our current de facto rule-making system that guides AI towards the market’s goals.”
But given how well-financed the market players are, is this really possible? Can the forces of regulation truly overwhelm market forces?
2. There is, to be sure, a threat from AI if it is not governed. Can you touch upon some of the potential downsides of an AI that it allowed to advance without governance?
3. What are some efforts to start to build this regulatory framework? What players might take the lead?
4. Are there actions that certain key professionals can take? Say, data scientists, managers or AI developers?
5. What is your forecast for the years ahead, as we grapple with the increasing power of AI and the need to regulate it? How do you foresee this struggle evolving?
To provide insight into the future AI, I spoke with two leading experts:
Gillian Hadfield, Director, Schwartz Reisman Institute for Technology and Society
Zia Kahn, Senior VP, Innovation, Rockefeller Foundation
Moderator: James Maguire, Managing Editor, Datamation
Download the podcast:
The Rockefeller Foundation Report / The Need for AI Regulation
Kahn: Artificial intelligence has been an interest of the Foundation for actually a while, and we were actually the funders of the 1956 conference at Dartmouth that coined the term “artificial intelligence.” And the whole premise was around, at that time, they wanted to do research into how can we actually replicate the human brain. It was a little bit more of an academic, mathematical approach. And artificial intelligence has its ebbs and flows, but now we’ve seen an explosion in its use, and it’s really gone beyond an interesting technology to something that is just permuting all aspects, and we’re seeing this in the COVID response right now, how artificial intelligence is both being used to accelerate drug discovery and vaccine development, but also highlighting some of the privacy issues as we think about contact tracing and how we can use it in that context.
Kahn: So for us at the Rockefeller Foundation, our mission’s been for 100 years, how do we promote the well-being of humanity throughout the world. And right now as we’re thinking about this COVID and the pandemic situation, thinking about the near-term responses, but also, how do we set the course for a recovery so it’s a more equitable recovery. And we just feel guiding the development of AI now is really important to setting the stage for where we’re gonna go in the future.
Maguire: Let me briefly read this [from the Rockefeller AI report], because I think this sums up the question as I see it, really, it puts it in a true nutshell. It’s, “As we reimagine a way forward, The Rockefeller Foundation is betting that AI will help rebalance and reset the future in a way that addresses current inequities. To realize that outcome, we must develop a regulatory framework to ensure its responsible use. We need to reimagine an entire new rule-making system that guides AI towards society’s goals instead of our current de facto rule-making system that guides AI towards the market’s goals.” And I think that is really the issue, but I think it’s a very difficult issue because there’s very large companies that have enormous budgets, and they are pouring vast budgets into the development of artificial intelligence, applications, platforms, widgets, etcetera. The idea that some regulatory body, perhaps a governmental body, could actually really play referee against such powerful forces seems a little questionable, and I’m doubtful of that.
Hadfield: First of all, really important to recognize, there’s no such thing as an unregulated market. Markets are constituted by laws. We think about regulation just more generally. Markets are constituted by that. So the power that our large tech companies have today is in part constituted by the way the state protects contract rights, intellectual property rights, employment relationships, and so on. And the tools in our toolkit are some of those basic rules and those basic things that are constituting the power of markets.
But the other reason I’m optimistic about the capacity for now regulation that comes in, to say, okay, you could do this with AI, but you can’t do that with AI. You can use facial recognition on a phone, but you can’t use it to check up on your competitors, or you can’t have police departments using it in discriminatory ways. That kind of regulation, it’s definitely challenging to develop that regulation today, but we faced that challenge at the last major revolution in the economy, the early 20th century. That’s when we invented the regulatory state to harness and rein in the power of huge corporations at the time. Anti-trust law comes out at that point. I’m pretty optimistic that we can develop those new regulatory tools. I think they’re gonna look different than what we have now, but I certainly think we can do it.
I find the use of the words optimistic and pessimistic kind of interesting, because I feel like there’s been over time a negative connotation associated with regulation, particularly when it comes to innovation. That regulation is kind a bad, it slows things down. And to build on Gillian’s point, I sometimes use this expression of, the reason we have brakes on cars is not to go slow but so that we can go fast. And when we think of lots of markets, look at the health market. It’s fairly heavily regulated. But you can’t even imagine a system by which you could develop drugs or provide healthcare to people when you’re thinking about their safety without a lot of that regulation. That’s not necessarily a bad thing, it’s kind of striking the right balance.
Potential Downsides of AI
Hadfield: Well, automation definitely changes the way who’s doing what jobs. And again, we’ve been through significant rounds of automation. I’m getting into my historian mode here, but in the 19th century, 70% to 90% of the population is working in agriculture and, of course, that changes over time. I’m not sure that we wanna hold up necessarily also the types of lives that people live in sort of mass manufacturing environments and factories as the ideal of people’s lives. So as an economist I’d say, look, first of all, yes, we should expect to see continuing automation and as we know, automation kinda ups that value chain. My colleagues in law certainly are gonna see some of their work displaced by artificial intelligence as well as a factory worker.
But I think because that creates more value, what we should be seeing is then a change in the mix of what kinds of work people do, how people spend their time. Wouldn’t it be nice to actually have a world where we were producing more or the same amount or more output, but people have more time to spend with their families, more time for leisure activities, more time for the types of creative work that we see unleashed by the kinds of access we now have to social platforms. We can write. We can post videos. We can do artwork. We’ll definitely see a different world. I think the question is, how do we share the surplus and the benefits of these technologies in a way that is equitable and supportive of the flourishing of human lives?
Kahn: I think that the downside if AI isn’t properly regulated, particularly in a context where we’re undergoing a big transformation, will be like what we saw when manufacturing and technology entered manufacturing. Or if we think of dislocations from using coal to using clean energy. These are big transitions that happen in society. And unless we think about what outcomes do we want, and we don’t sort of combine market and government to help guide those transitions so we maintain good social outcomes, then that’ll be a big risk. To indulge your negative, but in… And there’s good reason to be negative, there’s good reason to be concerned. One big concern of mine if we don’t regulate AI is that the current inequities that we have in society will get frozen in, because AI will just replicate all the biases that we have and make them kind of permanent versus just cultural and social. So that’s a big problem. We’re already seeing the growing inequality that’s happening right now and I think AI could just exacerbate that.
“And then to your point, we could see massive job losses and replacements that happen with AI. And if we don’t do that thoughtfully, then all of a sudden you’re gonna have entire groups and large groups of people who find themselves very limited with opportunities. So we’re seeing all these anecdotal issues when it comes to education, you’ve heard about the story about people who are assigning grades and that didn’t work out. And justice when they were trying to use AI to sort of determine whether people are guilty or not. That’s not really working out. Health is a big concern area. So there’s plenty of downsides, for sure, but I wouldn’t wanna throw the baby out with the bath water in ignoring the upsides. And we have a fundamental belief that AI can be a force for removing inequities and guiding a more equitable recovery as we come out of 2020.
Potential for Regulation in the AI Sector
Hadfield: Thinking about it in the AI setting, is something I call regulatory market. So this is a… Can we create a layer of competitive regulators, private regulators, companies that are investing in regulatory methods and technology, but regulate those regulators by having government set the outcomes that they have to achieve. So if we did this in the context of self-driving cars, it’s a very simplistic version of regulation, but you have a politically determined what’s the acceptable accident rate on the highway. Okay, so now I may be a private regulator that says, “Well, you know what, I’ve got a set of rules I think I could implement, and the companies that would have to buy my… They have to buy regulatory services, I would create a regulatory machine.” Zia might say, “You know what, I’ve got a way to do some technology for that. I’ve got a machine learning safe model that will regulate the vehicles.
“And Zia and I both have to demonstrate to government we achieve the target outcome, but now we’re competing to maybe provide you if you’re the manufacturer of the self-driving vehicles, you’re choosing between the methods we are proposing for achieving those outcomes. We both have to achieve the same outcome, but we are investing in figuring out better, more effective, more adaptive, rapid ways of achieving that. So that’s the… I think there’s a way for us to use those tools to get to this more adaptive, agile form of governance.
Kahn: Well, I actually am a big fan of Gillian’s model here, and I think there’s a slight analogy, it’s a little different, but when you think about insurance, which was the government sets the standard, and everyone needs to have insurance, and then you can shop around in an insurance market, there’s some kind of loose analogy to that. So ultimately, government will have to set the rules for what are the social outcomes that we want, that is a political process, and right now we have private sector companies that are in essence setting those rules, and whereas they were comfortable with that rule-setting before, they’re growing increasingly uncomfortable, and we see something like Microsoft, which is now not selling facial recognition technology to police departments because they wanna force the government to come in and set the rules and the norms because it’s out of the scope of what they want to or are able to or should be focusing on.
Kahn: We believe that there’ll be a raised consciousness in the same way that you have doctors have a raised consciousness about what actions they have have impact on people and society. Same with the legal profession. I think there’ll be a professionalization of data science that raises the social consciousness, and that will be a force that we can harness. We’re already seeing it in a lot of the large companies, large tech companies who are responding to their employee concerns as much as others. So in terms of what specifically someone can do, I think there’s a lot of resources out there to just understand what are the frameworks of how to think about ethical development and to engage with their management, to engage with their companies around, how do they think about the unintended consequences of their work, how do they think about choices that they can make and just create some of that internal pressure within companies. Companies themselves had business reasons to be interested in this, civil society has reasons, government has reasons. And that’s something that we’re excited about is we’re seeing lots of people who are aligning to this notion of we need some form of governance. We just don’t know exactly what it is.
Future of AI and Regulation
Hadfield: Well, so I think two possible paths. One is that we don’t address this problem, and what we see are just the exacerbation of the inequalities and power and so on that we’re seeing now. And we see the blunt force instrument of “We’re just gonna ban this stuff because we don’t trust it and we don’t like it.” So I think that’s one not very happy future, and that’s what I think happens if we don’t solve this problem.
Again, I’m back to the optimist, and I’m also, if you’re selling ideas, so I’m gonna be optimistic. I’m gonna believe I can sell these ideas because I am absolutely confident there are paths forward where we get to smart regulation that harnesses the power of AI and allows it to become part of our regulatory environment so we can continue to shape our future as humans. I think there’s some reason to think that we are on that path. In the last year, we’ve started to see the shift from the call for fairly abstract guidelines and ethical principles, all of which is very good, to a recognition we need to start building the concrete regulatory mechanism.
That’s what Zia and Rockefeller and Schwartz Reisman and also with the Center for Advanced Study in the Behavioral Sciences at Stanford, we’re starting on an initiative to say look, okay, let’s start thinking about how we actually build those regulatory. Don’t just call for governments to write new laws. Don’t just call for engineers to be more ethical. It’s a solid regulatory challenge. There are ways forward on that. So I’m gonna be optimistic and say that’s the path we’re on.
Kahn: So I believe what you’ll start to see are some countries or states who actually figure out how are we gonna make sure that AI is a good infrastructure that helps us serve our society on health and education, and also creates interesting market opportunities, and just in the same way that the states figured out how electricity could help in that way, and how the internet and broadband helped in that way. And so I think there’ll be these positive examples out there that will start to become more and more common and that people will seek to replicate more and more.
Kahn: We’re seeing some examples of that. This isn’t exactly the same, but when it comes to digital ID, the countries and the states that are able to create a real digital ID system are seeing so many benefits from that, that more and more people then look to them. So I think five years from now we’ll be hopefully past the state of just those initial little examples, and there’ll be more and more a common and systematic approach of how do we, whether it’s some of Gillian’s ideas like unlock these markets for regulations, but it’ll be more and more common. I think it’ll be more and more of an expectation.