Hello, and welcome to this week’s edition of the Future in Five Questions. This week I spoke with Rebecca Finlay, a longtime artificial intelligence policy expert who now serves as CEO of the nonprofit Partnership on AI. The group, whose mission is to ensure “developments in AI advance positive outcomes for people and society,” represents industry heavy hitters such as Apple, Meta, OpenAI and Microsoft, bringing them together to develop AI standards that meet those values. We discussed the challenges of ensuring various nations’ policy frameworks for AI sync up, the urgency of preparing the workforce for AI and why she doesn’t necessarily think watermarking AI-generated content is always the answer. An edited and condensed version of the conversation follows: What’s one underrated big idea? How we successfully make sure that our AI is interacting effectively with humans. We spend a lot of time talking about the technology and where it's going, but not the question of human-AI interaction. At the end of the day, if these systems are going to do what we all hope they are able to do in terms of driving benefits for people, we really need to be thinking about how they work for humans. From a public policy perspective, that means we need a well-funded public research infrastructure that draws on interdisciplinary expertise from social scientists, sociologists and organizational behavioral managers, and areas of research that are thinking about how these systems work within human and social structures and settings. What’s a technology that you think is overhyped? There is a lot of focus on technical watermarking of AI and synthetic media. You'll often hear that watermarking is a way to solve issues related to misinformation and disinformation when it comes to generative media, and what we know is that it’s really only one piece of the puzzle. Yes, we should have technical solutions in place to both better identify AI-generated media and also to better identify authentically generated media. But at the end of the day, we are going to need to think about how the public, when they are engaging with this media, can be reassured that there is some authenticity, some trust, some way in which they can assess its validity. That is not going to be a technical solution, that is very much going to be a sociotechnical question that comes back to trust in institutions and trust in our information ecosystem. What book most shaped your conception of the future? One is “Power and Progress” by Daron Acemoglu and Simon Johnson. In thinking about the future of AI technologies, they look to the past and make the case that it has never been true that technological advancement alone benefits society, there have always got to be other social mechanisms in place. In the case of the Industrial Revolution, that meant that you had workers who were coming into cities who were starting to get organized, and you also had regulations coming into place to make sure that workers were protected. They make the case that we too often focus on the technological innovation that is driving its perceived benefit, and it is true, but there are always other measures around social, political and economic responses that make sure that the technology benefits us. Then, I find myself often thinking about “Klara and the Sun” by Kazuo Ishiguro. I adore his writing. This is a beautiful story that, yes, is about a creature and their relationship with humans, but it's really about our humanity as we interact with and engage with technology and science. Klara, the automated character in the book, is powered by the sun, and so much of our attention right now thinking about responsible AI development is going to be about how we make sure it will be environmentally friendly, and how we need to be attending to renewable resources and energy consumption. What could the government be doing regarding technology that it isn’t? Making sure that the workforce is prepared to both further integrate predictive AI models and then to integrate new generative AI models. Thinking about what that means from an educational perspective, from a reskilling perspective, but also what it means in terms of ensuring that worker protections are in place and worker voices are at the center of making these systems work. Often what we've seen, and we've seen this in terms of AI systems being deployed in health care settings, is that they fail because they haven't attended to the current workplace practices. Who are the individual operators that need to be part of figuring out how to integrate these systems into their decision making processes? Again, it's the whole question of human-AI interaction. If we're not thinking about how you translate that research into practice by understanding the current structures in place, it is not going to be successful. What has surprised you the most this year? You cannot have worked in AI and its social impact space as long as I have and not be surprised by the policy response since the launch of ChatGPT. It is so heartening to see that policymakers are engaged in asking fundamental questions and attempting to become more educated about the technology and its impact. We focus on a lot of work on foundation model safety, so I think the U.K. has to be commended for getting out in front of this with the U.K. AI Safety Summit, and really catalyzing this. Then the U.S. executive order was a really important piece of this. We're seeing things like the work that's happening at the regional level in the EU with the AI Act, the attention of the G7 and of course, just this last weekend the release of the UN's AI Advisory Body report. For me, it has been very, very heartening to see that the policy community is starting to understand that they need to attend to this.
| 
