Who Decides Which AI Risks Matter? Nick Diakopoulos on Media, Politics, and AI Governance

AI in Chicago, hosted by Khullani M. Abdullahi, JD

Guest: Nicholas Diakopoulos, professor of communication studies and, by courtesy, computer science at Northwestern University; director of the Computational Journalism Lab

Recorded: February 19, 2026 · Published: August 24, 2026 · Running time: 48:19

This transcript has been lightly edited for readability: filler words, false starts, and conversational acknowledgements have been removed, and speaker turns interrupted by those acknowledgements have been rejoined. No substantive wording has been changed. Section headings mark the flow of the conversation. Per-turn timestamps have been omitted: the published audio is an edited cut running 48:19, while the source transcript was timed against a 52:20 recording session, so the two do not align. Where something said on air is contradicted by the underlying record, the spoken words are left intact and a bracketed editor's note appears alongside them.

Introduction

Khullani Abdullahi

Hello, and welcome to the AI in Chicago podcast. I'm your host, Khullani Abdullahi, the founder of Techné AI, a Chicago-based AI governance, risk, compliance and strategy firm. AI in Chicago spotlights our local academics, operators, entrepreneurs, and thinkers, as well as policy leaders who are scaling applied AI from our home in Illinois, but with a global impact. Each episode delivers practical stories and insights, empowering leaders to understand AI, the various risks and use cases, but minus all of the hype.

Today, we're continuing to think about AI at the intersection of different fields. In our last conversation, we discussed AI and quantum computing. Today, we'll dive into AI at the intersection of policy and the communication of AI risks in the media landscape, as well as potential governance options.

Nicholas Diakopoulos is a professor in communication studies and computer science at Northwestern University, where he directs the Computational Journalism Lab and is the director of graduate studies for the Technology and Social Behavior PhD program. [Editor's note: The computer science appointment is by courtesy. Diakopoulos served as director of graduate studies for the Technology and Social Behavior PhD program from 2020 to 2024; the role is currently held by Matthew Kay.]

His research focuses on computational journalism, including aspects of automation and algorithms in news production, algorithmic accountability and transparency, and social media and news contexts. He's also the author of the award-winning book Automating the News: How Algorithms Are Rewriting the Media, published by Harvard University Press in 2018. [Editor's note: The book was published in June 2019.] Long before ChatGPT, Professor Diakopoulos was talking about algorithms. He received his PhD and MS degrees in computer science from Georgia Institute of Technology and his Bachelor of Science in computer engineering from Brown University.

Welcome, Nick. I'm so pleased to have you here.

Nicholas Diakopoulos

Thanks so much for having me. I'm excited for the conversation today.

From computer vision to algorithmic accountability

Khullani Abdullahi

You've been thinking and talking about algorithms since before it was popular. I did highlight that your book was initially published in 2018. Why don't you kick us off with telling us a little bit about your background and the big question that you work on that led you to be so early in this trend.

Nicholas Diakopoulos

I started my PhD back in 2003 and came into it thinking that I would do computer vision and study visual interpretation of information. And when I finished in 2009, I looked completely different as a scholar. I was much more interested in people and technology, how people interact with technology, how they're impacted by technology — so the broader area of human-computer interaction, and how that intersected with media. Journalism and news. My family has a couple of journalists and former journalists in it, so I was always exposed to news and information and media and journalism growing up. It was a natural intersection for me to start to explore how technology was going to interface with information.

A lot of my early work in journalism was around information quality. In 2006, when YouTube was a big thing, people were just starting to realize that there was going to be a lot of social media that had information that wasn't necessarily credible in it. That was the premise for my dissertation — thinking about tools that could support information quality evaluation. So that was an early motivation.

As I got more and more into the journalism stuff, and I was working at CUNY in New York and then Columbia in New York, I started talking more and more with journalists and starting to appreciate and understand accountability as a core facet of what it means to do high-end investigative journalism. And I started connecting that over to my interest in technology, and coined this phrase algorithmic accountability in 2012 — this idea that we should be investigating algorithms in society and understanding the power structures around algorithms and AI in society. [Editor's note: The earliest use of the phrase by Diakopoulos that we could verify is "Rage Against the Algorithms," published in The Atlantic in October 2013, where he writes "I call it algorithmic accountability reporting." The Tow Center report "Algorithmic Accountability Reporting: On the Investigation of Black Boxes" followed in 2014.]

That's you sort of alluding to how I was early to this. I don't even know that I was early. There's people who had been talking about accountability and AI back in the 90s, but it never really caught on. So maybe I just hit it in a moment around the early 20-teens where society was ready to really care about accountability of AI and algorithms.

What held up from the book, and what didn't

Khullani Abdullahi

If I remember correctly, the "Attention Is All You Need" paper also came out in 2017, 2018. [Editor's note: "Attention Is All You Need" was published in June 2017.] There's some degree of technical progress in the machine learning field that also gave you great timing. You were doing the work, and then society was ready to reflect on that work seriously.

I want to have you reflect a little bit on the arc of the book. Correct me if I'm wrong, but as I understood it, the core argument is that automation is not replacing journalists — it's creating a human-algorithm hybrid. You discuss the tension between some of the commercial imperatives. You give the Associated Press and financial reporting from earnings calls as key use cases and vignettes. You talked about journalistic values remaining very central. And in the book, you express some degree of skepticism about the hype cycle, and you thought that AI and journalism would "get boring" — I think in a good way is the quote.

So we are eight years since you published that book, probably a decade since you started researching and drafting it. If we think about the genesis of this book in 2016 and where we're sitting now in 2026 — what held up? What still feels true and accurate in terms of how you were thinking? What surprised you, or changed, or shifted? And where do you think that arc might be going, prospectively, in this relationship between journalism and media and how they're using artificial intelligence technologies, whether it's new products and services or just the underlying algorithms?

Nicholas Diakopoulos

There's a lot there, but it's a great question. To jump back to the idea of the hype cycle — it's so interesting, because when the book came out, we were still in a hype cycle around algorithms and automation and news media. And then I think it was around 2022, before ChatGPT came out, things felt like they were actually settling down. The industry was hitting a steady state. They were doing all the things. They were automating stuff. They were using algorithms. They were using machine learning. They were optimizing their audiences. They were optimizing recommender systems. They were using data. And it had gotten a little boring. It wasn't as hyped. It wasn't as sexy.

And then of course ChatGPT hit, and we went immediately back into a big hype cycle where by 2023, 2024, people were like, what are we going to do? This new technology has new capabilities. It's changing everything.

But honestly, in terms of the core themes of the book and the idea that humans and AI are going to work together and be hybridized, I don't think a lot of that's changed. I still believe that core idea — that you need human judgment, human oversight, humans in the loop in general. And that there are a lot of gaps in the capabilities of these technologies in terms of what we need them to be able to do in order to do good journalism. That might include having access to the right data to be able to know or observe what's going on in the world. Or it could be about having the right judgment about what matters about the data, or how to interpret it. All those kinds of human elements, I think, are still critical. I think that has all held up, and I'm proud that the book has held up.

There are some parts of the book where, when I wrote it in 2017, 2018, I was maybe a little dismissive of the generative technology. It was still really early and I just couldn't see that it was going to work. I didn't totally dismiss it — there's a version of it in there, and I'm sort of like, this is a thing, it could happen, but it doesn't look like it's going to happen right now. So certainly the technology has advanced. But a lot of the human elements of the analysis in the book still hold up.

The layer he missed: management, not capability

Nicholas Diakopoulos

One thing I could have — or maybe I was a little naive about back then — is just how much the management of the technology matters. What I mean is whether or not AI leads to job losses. Yes, it matters how capable the technology is, but it really also matters what's the management of the organization around the technology. Does it want to reallocate time saved to improving the quality of the product, of the reporting? Or does it want to cut hours from people, or cut positions?

Just thinking about the management structure around the technology wasn't something that I really talked about in the book. If we were going to go back and do it, maybe that would be another layer that I would add in.

Why most newsroom job losses were not AI

Khullani Abdullahi

I want to pull on two threads. One is that implicit in the idea of management technology choices are also ethical and normative choices that these companies have to make. In 2018, I think it wasn't clear that there would be a direct correlation between the scaling up of the automation and the use of these machine learning technologies to produce media at a greater scale. But now we are seeing — in the last decade, we have seen — a virtual collapse in the workforce that works in print, that works in digital media, that works in traditional news. And we've seen this collapse even at premier, tier-one news organizations, and we've certainly seen it collapse in local and state-level news media. So there's that workforce component.

And then there's another thread, which was this idea in the book of the hybrid workforce, where the augmentation is — I got the sense — mostly positive. Do you feel that continues to be the case in your own students that you're seeing, in the use of AI as a professor now? If you put on your professor hat, not your researcher hat: what is the relationship that your students have to these technologies? Is that relationship healthy? What is going well, what is not going well?

So, as a professor, I'd love your thoughts on this hybrid AI-augmentation human student. And then secondarily, on the workforce component and the choices that the news media organizations have been making that have hollowed out labor from 2018 to now — and where you see it going in the future.

Nicholas Diakopoulos

The workforce component is a little bit of a tricky story. Management maybe likes to have something to point to when they make cuts, when they cut back staff. And AI doesn't get to really have an opinion. So you can just point the finger at AI and say, well, it's because of that. And it is more difficult for people to argue against that.

And of course, there's this hype narrative that everyone's going to lose their job. Even the essay circulating this week about how it's happening to software first, it's going to happen everywhere else next. Andrew Yang is still hyping this stuff. There's a whole hype economy now.

There is something to that. There's something changing, especially with software engineering right now. The degree to which that is going to extend to really truly undermining workforces in a domain like journalism, I think, is still somewhat yet to be seen. I don't want to say that there have never been any workforce cuts due to AI or technology in journalism. But I actually think that probably a vast majority of the reduction in the workforce that we've seen over the last decade has nothing to do with AI. It has everything to do with competitive dynamics, and competing against social media companies and the ability to capture attention in as effective a way as social media has. And to some degree, maybe just general mismanagement — it could be the case that these companies aren't keeping up in some of the same ways. They also have legacy costs that they're increasingly dropping, like the print editions, because those are expensive to produce. And so you can't necessarily compete against the nimbleness that you need in order to exist online.

Now, I will counter this narrative a little bit and say we've seen the rise of local media in the same period — something like Block Club Chicago. I don't know exactly how many journalists work for them, but it's more than a couple. It's maybe a few dozen.

Khullani Abdullahi

Yeah, it's a real organization.

Nicholas Diakopoulos

And so at the same time as we've seen cuts, we've seen things starting to bubble up in different ways. And it's bubbling up without the same kind of overhead structures of a traditional news organization, which had to own big expensive real estate and printing presses and all that kind of stuff.

Now, next to that — to bring up the management issue again — you've got private equity, you have big chains who own stuff, and they're really trying to squeeze out every last drop of value they can from news organizations. And they are using technology to optimize and make things efficient in certain ways, and that has negative implications for labor as well.

What I would say is, I am still optimistic about the augmentation vision that I talk about in the book. I don't want to be blind to the fact that there are some shifts in the workforce that are related to technology. I think the people who are doing well are able to learn and re-skill as they go, and learn how to use tools like AI tools to be more effective and more efficient in their work.

Teaching generative AI — and letting students write the policy

Nicholas Diakopoulos

To connect that over to my life as a professor — I've been teaching a course on generative AI in the media for, this is the third year I'm teaching it. And the first year or two I was struggling a little bit, because some students were clearly relying on the tools more than I perhaps wanted them to. Now, it's a course about generative AI in the media, so it's very lean-forward. I want them to use tools. I want them to learn responsible use of tools, when not to use it, where are the limitations and the gaps in capabilities. But it was clear that people were over-relying on it.

You can't usually use AI detectors to get a perfect read on something, but there are sometimes other little signals where it's pretty clear that someone's been using AI.

The shift I've seen this year is that the students — and maybe this also has to do with the way I framed it — on the first day of class, I bring them in and we basically workshop it. We say, let's spend half the class developing an AI policy, guidelines for use of generative AI in this course. And I trigger them to think about what are your learning goals. What do you even want to learn? And get them to really think about that and connect that to the tools that are in front of them.

The discussions are great, and people are clearly waking up to this idea that, huh, well, if we're actually going to learn something in this course and meet our learning objectives, we really can't over-rely on this technology. We really need to be critical of it. We really need to be skeptical. We need to be having full oversight. We need to be thinking about privacy and all of the ethical issues, bias that can emerge of tools.

And I've seen a real shift throughout the course after that. I see it in the reading responses that they write. I see it in the assignments they do. People are just being a lot more considerate in their use of the tools.

Maybe that's optimistic — that as people get to be more literate and critical of the tools, they can appropriately use them. Now, I do worry when these people go into organizations and the management and the leadership says, nope, we're going to optimize, we need to be more efficient. What happens to that individual sense of responsibility in that context?

What over-reliance actually looks like

Khullani Abdullahi

I would love for you to write and disseminate the way that you kick off your class, because I think that is such a phenomenal idea. It forces them to say, I'm investing in my education, I want these outcomes — am I going to get these outcomes if I use this technology in this way? I think that's such a great way of giving them ownership of helping shape the AI policy.

In my own work, I do a lot of governance, risk, compliance, policy development for corporations who are managing risks associated with adopting AI. But I hadn't thought about it in the classroom context. So I think that's something that needs to be circulated. We do see, in the public imagination, a lot about students not being able to read and synthesize, the loss of analytical thinking, the loss of being able to draft, to sit in front of a white page and be able to come up with a thesis and then support that thesis. So I think that is a bright spot, a silver lining as it were.

So that comment aside — something that I want to draw on is, in the real world, whether it's journalists or students, and we're thinking about how do we ensure that these technologies serve humans without deteriorating cognitive abilities: what does over-reliance look like? What are the factors?

I have developed something that works for me, which is that I like to be delighted by reading. And that's because I really enjoy reading. So I find content generated by generative AI to not typically be delightful. It's informative, could be well-researched, could be grammatically correct, but it's not delightful. So I subscribe to a number of literary journals and I read pre-2023 content. I subscribe to Foreign Affairs, and Foreign Affairs does a really good job of not having a lot of generative AI in it. But even then, I'll focus on things written before 2023, just for the purposes of being able to look at human-generated content. If it's pre-2023, I know that your book did not include any generative AI. You sat down and came up with every word and every paragraph and every sentence and every comma. That is not true for professors publishing today. Everybody drops it into ChatGPT or Claude or whatever and says, hey, critique the logical flow of this.

So what does over-reliance mean and look like, and what are the implications of that? And have you developed any safeguards in the way that you use these technologies in your own work that shape your desire to preserve the analytical skills that you've developed in the last 40 years since you started school?

Nicholas Diakopoulos

That's a great question. Over-reliance basically gets at this idea of trusting something maybe when we shouldn't, and getting comfortable just going with what's being suggested to us without being critical. So when I think about over-reliance, I think about building the muscle of critical thinking — I think you would call it metacognition, the ability to reflect on something and critique it from a step back.

So how do you do that? You try to take breaks and remind yourself that you need to do that. And it can be mesmerizing sometimes looking at the output of some of these tools. You can forget yourself, and you can forget that you should have that critical step back and reassess.

Now, sometimes it's more important than at other times. There are maybe low-stakes things where it just doesn't matter that much. But certainly for more high-stakes kinds of things, I think it does matter.

Why people get critical only when the stakes are high

Nicholas Diakopoulos

I will say, on an optimistic note, we've done some research where we interview people about their use of chatbots or AI Overviews in Google to search for news information. [Editor's note: This interview study, from Northwestern's Generative AI in the Newsroom initiative, was previewed publicly in February 2026 and had not been published at the time this transcript went online.] And you do find that people tend to over-rely when the stakes are low. They're willing to just go along with it and accept the credibility of it. To some extent they're aware that there might be hallucinations or issues, but they just accept it.

But when things really matter — like in the online content world, they would call it "your money or your life" — when things really matter to you, your finances, your health, people switch into deeper cognition. They start getting critical again. They start thinking, hmm, I need to click into this. I need to evaluate that piece of information. I'm not just going to rely on the synthesized summary of it. [Editor's note: "Your money or your life," or YMYL, is a category from Google's Search Quality Rater Guidelines covering topics that could significantly affect a person's health, financial stability, or safety.]

And so that in and of itself is kind of encouraging. People understand that when there are stakes, they do need to slow down and switch into that deeper level of thinking. So part of the challenge then, maybe, is how do we encourage people to think about the stakes? How do we get people to reflect on that? I don't necessarily have an answer to that, but I'm starting to think about it.

You cannot be the human in the loop without expertise

Nicholas Diakopoulos

In my own work, I don't know that I necessarily have good safeguards for preserving my analytical skills. I will say that I have enough metacognition to start to recognize when I am going a little soft.

I grew up as a kid writing computer code, and I did it in college and I did it in grad school. And as a professor, I don't really write a lot of computer code anymore — but I can. I know how to. I can write a lot more of it now and do a lot more data analysis using generative AI tools. But I know that I'm soft in those skills. I was already pretty soft in those skills. And so maybe I'm almost comfortable with not keeping that tool as sharp as it once was in my life.

I can still evaluate it. When I see the code that it produces, I still know how to read and interpret the code that it produces. And so maybe that's the skill that I need. And that's the skill that I'm still keeping sharp when I'm using the tools — because I have to, because I know I can't rely on it fully. And I know that I'm going to review the code that's generated.

Khullani Abdullahi

That's really insightful, because what it suggests is that you need a baseline level of expertise when using these technologies so that you can still play the judgment and intuition and evaluative function that the human plays. You can't be the human in the loop unless you know something that the tool you're using doesn't know — that there's something that it could get wrong. There's some degree of expertise and judgment which is only developed through time. If you didn't do the coding and the learning and the research and the writing before these technologies, I don't know if you would be in this position.

Which makes me a little concerned for the kids who've had ChatGPT since eighth grade. Is there going to be sufficient time for them to struggle through and develop those skills? Or does the skill formation get interrupted by these technologies? There's some research questions there, and some research projects there, I'm sure.

Nicholas Diakopoulos

Definitely. We had a workshop in our department yesterday, just talking amongst colleagues about how are we grappling with AI in the classroom. And this exact topic came up. Maybe we need a different approach for freshmen and sophomores who are still developing some of those core analytic skills, critical skills — whereas juniors and seniors are like, well, we need to really learn how to use the AI tools, because we know that wherever we go and work is going to want us to know how to use those. So how do you manage also that transition through an undergraduate's trajectory?

Khullani Abdullahi

That's a really great point, because those use cases and the AI literacy that is expected in the workforce — even when I was coming out of college, companies complained that we would come out of college unprepared to work. I went to a liberal arts college, my degree was in philosophy, no one expected me to show up and be able to use Excel. That wasn't happening unless it was an internship. So I think there's always been this sense of a gap between what industry needs and what academia provides.

But those core skills — being an analytical thinker, being a great writer, being able to synthesize data, being able to speak concisely, being able to do research and analyze data — as long as those core skills are formed, the AI tools change so rapidly that there's no one AI tool you could integrate into a student's education that you would be confident would still be relevant. But maybe the intuitions and the judgments that you've been talking about: when to feel like an output is sufficient, or when you need to really review its output and decide whether its reasoning was actually correct. I think there's something to be said there.

Measuring AI risk coverage across 27 countries

Khullani Abdullahi

I want to shift gears and talk about one of your papers that I thought was really interesting, which is the global perspectives on AI risks and harms. From my notes: in this paper, you looked at 27 or 28 countries to research and understand how AI risks are discussed and prioritized globally. You looked at rank-ordering the risks, and you found that societal risks dominated, then legal and rights, then content safety, cognitive, then existential risks — which I thought would have been number one. [Editor's note: The study covers 27 countries. Its results section reports Cognitive Risks at 14.2% ahead of Content Safety Risks at 8.3%, reversing the order given in the paper's abstract. Full citation: "Global Perspectives of AI Risks and Harms," Mowafak Allaham, Kimon Kieslich, and Nicholas Diakopoulos, arXiv:2501.14040.]

That's an overview, but the question that I really have is: you found that risk prioritization varied by the political bias of the news outlet, and you suggest that AI risk assessments and prioritization are inherently political. I'd love for you to tell us a little bit about the research and the paper, because in America, where we're talking about AI preemption and state AI regulation and federal preemption, this is a topic that is top of mind across the United States. So it's interesting that you have some data already on the role that political bias plays in how we even talk about AI risks and harms.

Nicholas Diakopoulos

This paper was motivated through an interest in understanding how we could see the impacts that were being covered in news media about AI. What are the harms that even get talked about in the media? And could that somehow inform risk assessment or risk management?

And of course, media is an imperfect system. It has its own biases. Our assumption going in wasn't that a policymaker would only look at news media for this, but that it could be another input in the stream of things that you're looking at. And that looking at it from a bird's-eye view could maybe help us understand some of the limitations or biases or tendencies of what got covered and what maybe didn't.

Starting from there, we wanted to look across different countries to see, internationally, whether or not there were differences. I think this is important when we're talking about AI especially, because a lot of it's produced in the US and exported around the world. Obviously there are AI companies in other parts of the world producing things, and some of them also do get imported into the US, but these are global technologies and they move around very easily.

So we wanted to understand: if the policymakers in the US are just paying attention to US media, could that lead to different emphasis on different kinds of risks and harms than maybe in other countries where these same tools might also be deployed?

How left- and right-leaning outlets report different harms

Nicholas Diakopoulos

And then we're also, within the US, interested in: how about even just left versus right? Are different kinds of harms and impacts discussed more on the left and on the right?

And it turns out that there are some differences. The left maybe focuses a little bit more on socioeconomic and environmental harms, whereas the right maybe focuses a little bit more on discrimination and toxicity. Interestingly enough, that is a right-leaning talking point — focusing on toxicity. Not in the same way as someone on the left would talk about this thing, but getting maybe at part of the same core idea, from a very different political point of view. Or folks on the right maybe talking more about malicious use or misuse of AI systems.

[Editor's note: The left/right findings described here are reported in a companion paper by the same authors — "Informing AI Risk Assessment with News Media: Analyzing National and Political Variation in the Coverage of AI Risks," Allaham, Kieslich and Diakopoulos, AAAI/ACM Conference on AI, Ethics, and Society 2025, arXiv:2507.23718 — rather than in the 27-country study referenced in the question.]

Khullani Abdullahi

Not the same definition, but yes.

Nicholas Diakopoulos

When you think about this through a policy lens, it's like, well, what are policymakers paying attention to? And if we go back to media effects theory — agenda setting and framing and so on — how are those biases maybe working their way into the kinds of debates and dialogues we're having about the technology?

Now, we're not able to connect those dots explicitly in this research. We're just more describing what we see there. But we do see some differences, which I think is helpful to know about.

The think tank to media to policymaker loop

Khullani Abdullahi

I think it would be helpful — not everyone is aware of why choosing media to look at this lens of AI risks and discussions is important. But one thing that I'll note before my next question is the interplay between experts and research, which I think goes to your next paper that we'll discuss. The interplay between the think tanks that do the research, and then the peer-reviewed research, and then the results that end up in the media — that pipeline almost creates this virtuous, or not-so-virtuous, feedback loop that then shapes the prioritizations of policymakers.

In light of this paper and your research in general: if I wanted to influence a state elected official on AI policy, how would my tactics change in light of this paper and this research?

Auditing a policymaker's information diet

Nicholas Diakopoulos

That's interesting — bringing it down to a tactical level. I would be thinking about what is the information diet of that policymaker. What are they looking at? And what is their office looking at? And do they have good information hygiene in terms of looking across the spectrum, or does it seem like they're really focused on one side or the other?

Media embeds biases in various ways. And so if you're, let's say, a policymaker with a particular political agenda, you might be predisposed to want to consume media which reflects that agenda in some way, or is focusing on the same frames of the problems, the same frames of the issue as you see.

So I'd be thinking about that, and then maybe thinking about how to inject a more balanced view to those policymakers or offices. I'm not sure the right mechanism necessarily to do that. It could be meeting with them to make sure that certain frames from the media are on their agenda, or if there's capacity, to send information sources into the office to ensure that they're looking at those information sources. That could be useful.

Expert bias and the case for participatory governance

Khullani Abdullahi

That's interesting. I think a lot about this because, to your point and to the researchers' point, we can't trust that our elected officials are looking at objective and a cross-section of data about AI risks and harms when they're developing their AI risk prioritization and policy frameworks. Highlighting that means there's an opportunity to address it up front. And it does, to your point, require engaging them and ensuring they have a diverse set of data about AI risks and harms.

In your second paper from early 2025, where you talk about envisioning stakeholder-action pairs to mitigate negative impacts of AI — in my mind, I read this paper as an attempt to democratize who is involved in shaping the conversations and policies around AI. So I think this is a great segue from the AI risk conversation into your central thesis and claim that there's a lot of missing nuance if we rely on an expert-driven and expert-based AI policy development framework.

First, I think that that is true and rare. What is it that you discovered in your research — that everyday lay individuals can meaningfully contribute to the development of AI policies that will impact their entire lives? And how can they go about participating? How do we bring participatory engagement from everyday lay individuals in what is probably the most consequential technology that our species has ever built, and the policies around which will determine everything from will there be jobs, to who has jobs, to who will eat? Where do lay individuals fit into that, based on your research?

Nicholas Diakopoulos

I love this. You're alluding to one of our papers from 2025, but we've been working on this line of research for, I don't know, half a decade at this point. [Editor's note: "Envisioning Stakeholder-Action Pairs to Mitigate Negative Impacts of AI: A Participatory Approach to Inform Policy Making," Julia Barnett, Kimon Kieslich, Natali Helberger and Nicholas Diakopoulos, ACM FAccT 2025, arXiv:2502.14869.]

Basically, we know that there's expert bias in this process. That's well established. Many people have studied that, and we know what some of those expert biases are — in terms of overconfidence in technology to some extent, in terms of focusing on certain issues which might not necessarily be the most important issues when you talk to lay stakeholders or a wider array of people from society.

And then you also have people of expertise in different ways. They could be domain experts who know a lot about the technology, or they could be domain experts who know a lot about law, or news, or healthcare, or something like that. And the point is not to dismiss expertise. We need people with domain expertise to help regulate these things. These are very complex systems. We need people who know how those systems fit together — where are the weak points of those systems, or the intervention and leverage points of those systems. So not to throw any of that away, but also to augment that with this more participatory input of information in the regulatory process.

And I think this is super important for legitimacy of the governance itself. If we go back to democratic theory — we've got to talk to people. These are the people who are being impacted by these technologies. We really ought to get some input from them in the process of developing those policies. Elections aren't the only way to get input from people. We can definitely leverage people's lived experience with the technologies, their individual identity and perspectives and context and how they interface with the technologies, and all that information can be input.

Stakeholder-action pairs: how the method works

Nicholas Diakopoulos

That's the premise — but how do we do that? In several of our research projects, we've been developing a method where we use scenarios, short written stories, as ways to convey how AI is being used or creating impacts or harms. And the great thing about stories is that they're really accessible. Everyone who's literate can read a short story and understand what's going on, and how the people are acting, and how the technology is interacting and impacting stuff.

And so these short scenarios — a few hundred words, maybe — have been a really powerful mechanism for us to then go solicit additional feedback from a broad array of people. We do these big surveys online where we can show people the scenario and ask them: how would you mitigate the harm that you see here? Who are the stakeholders that should be responsible for mitigating that? How do you think that this policy would change the outcomes in this scenario?

So we can start exploring those counterfactuals, and again leveraging people's diversity of thinking and perspectives. Different people bring different things to the table and help you understand what's acceptable and not. And also the normative dimension comes from that as well. This ties back to some of the political issues we were just talking about — different people have different senses of right and wrong, and what should be prioritized, and where the trade-offs should be between different things. So you can get all of that in these large surveys that we do, and start synthesizing that information together. That's the broader idea behind this particular paper, where we were trying to get people more involved in AI governance.

Two dozen AI bills in Illinois. Has anyone called?

Khullani Abdullahi

In Illinois, I think there's more than a dozen, almost two dozen AI bills being considered in this legislative session. [Editor's note: More than 50 AI-related bills were filed in the Illinois General Assembly in 2026, with nearly four dozen heard in Senate subject-matter hearings in April 2026.] Has any elected official reached out to you in the development of those bills, to talk about doing this work at scale among the Illinois public, so that that could inform and shape the policies we're developing?

Nicholas Diakopoulos

No, I have not received any requests for that. But I love the idea. I love the premise for it. I think it's something we could do.

Khullani Abdullahi

That's insane.

Nicholas Diakopoulos

I don't know that it would necessarily work for everything. I'm not familiar with every bill that's being considered right now, but I have to imagine there's some of them that would be accessible, that you could frame in certain ways where you could effectively gather input from.

Khullani Abdullahi

Child safety. Or at least in the consideration process. I do love the vignettes and stories. I think narrative is a very powerful tool, and finding a way to incorporate narrative into AI safety and risk and policy, and then being able to democratize who has input, I think is very, very compelling.

Ten years out: who figures out AI governance

Khullani Abdullahi

As we come to the end — any final thoughts on this. If you and I have this conversation again in 2036, and I'll be pointing to your 2018 book, which is almost 20 years old at that point: what do you expect to see from an AI-human augmentation, collaboration, interaction standpoint? And were we able to develop mechanisms to effectively govern AI?

Nicholas Diakopoulos

I love the question — ten years from now, where are we going to be? It's an impossible task. There's so many different directions the world can go.

I spent some time in the Netherlands in 2022, 2023, and I'm optimistic that someone's going to figure out this governance stuff. I don't know that it's going to be the US. Maybe we have some hope through state initiatives for getting some protection from some of the harms of AI. And I think over the next ten years we will see more state-level activity. I'm really hoping we see some kind of federal framework that can rein in some of the harms.

But politically, where are we even going to be in 2028 — I mean, 2029? A lot will depend on what Americans choose in the next election, and which direction that goes. Biden was already heading in some interesting directions in terms of developing responsible AI frameworks. The problem with executive decree is that it can go away when the executive changes. And so we somehow need to get our legislative bodies back on board with protecting people.

I do think that there are kind of slivers of leverage. Whether or not you're on the left or right, I think people seem to agree that protecting children is important. Maybe we'll see some action there.

Frontier AI laws, cybercrime, and the hiring Wild West

Nicholas Diakopoulos

Strangely enough, there's been action at the state level in New York and California around frontier AI. I'm sort of less convinced that that kind of policy activity is entirely necessary. I'm not thinking that we're looking at a biological or radiological event in the next ten years. I mean, it's hard to rule anything out, but I feel like people are very worried about some of the things — there's other things that we should be worrying about first.

[Editor's note: Illinois enacted its own frontier-AI statute after this conversation was recorded. The Artificial Intelligence Safety Measures Act, Public Act 104-0538, was signed July 6, 2026. It takes effect January 1, 2027, with its framework and third-party audit obligations applying from January 1, 2028.]

Khullani Abdullahi

The existential AI risks you would put on a lower tier of immediate — the lone terrorist in the basement who creates a new biological weapon. That you think is less likely over the next ten years.

Nicholas Diakopoulos

My sense is that that's probably less likely. But it's not to say that there aren't real risks as these systems gain more autonomy — especially, I think, cybersecurity risks. I would be more worried about AI tools that can generate code that can hack bank accounts. Stuff like that. But it wouldn't be autonomous. It would be that there's some criminal who figures out how to use the AI to do that thing.

So we should be worrying about those kinds of risks that are accelerated and exacerbated.

I also think we've got to do something like — I have friends who are on the job market now, and just the world that people are living in in terms of looking for jobs online, and the degree of AI on both sides, is really frightening. And it's the kind of thing that the framework in the EU prevents — hiring and job-related AI decisions are considered high risk, so they have an additional layer of scrutiny. And I just feel like we're unleashing this Wild West world of AI on people, and we've got to do something about it. So I think we will do something in the next ten years, and we'll see where we are then. [Editor's note: AI used in recruitment, screening, candidate evaluation, promotion and termination is classified as high-risk under Annex III of the EU AI Act. Those obligations were originally to apply from August 2, 2026; Regulation (EU) 2026/1744, in force since July 27, 2026, moved the date to December 2, 2027.]

Close

Khullani Abdullahi

Here's hoping we're still around in ten years. Thank you so much for sharing your thoughts, Nick. I really appreciate it. I will be sure to link to the two articles and the book that we discussed so that people get real context for the conversation.

Thank you to our audience. AI in Chicago — this will be on Spotify and everywhere else you get your podcasts. Until next time, chat soon.

Nicholas Diakopoulos

Thank you.

Transcript of AI in Chicago, hosted by Khullani M. Abdullahi, JD. Recorded February 19, 2026. Cite as: "AI in Chicago podcast, hosted by Khullani M. Abdullahi," with a link to this page.