Truth in the Age of Algorithms

CfD Conversations Fall 2025–2 | October 15, 2025

Written by Katherine Kim

The Center for Design hosted a conversation moderated by Sylke Meyer between John Wihbey, Jennifer Gradecki, and Derek Curry, who are all researchers working at the intersection of art, design, media, and information ethics, exploring how truth operates in this algorithmic age.

Meyer positioned this discussion around how new technologies enable new forms of deception faster than we can develop ways to verify truth. This discussion was grounded in both social epistemology and media theory, asking how individuals and communities learn, discern, and share information in changing and AI-mediated environments.

Misinformation spreads faster than ever before. Understanding how we navigate truth in these conditions has become crucial. This conversation examined these challenges from multiple perspectives.

Social Epistemology and the Paradoxes of Trust

John Wihbey approached the problem of truth and AI through the lens of content labeling. His research explored whether people notice warning labels on misinformation, how they interpret them, and whether labels actually change behavior and belief formation.

Wihbey explained how labeling systems shape user perception. He referenced research from the pandemic on content labels, asking whether people notice them, how they interpret them, and how they affect sharing behavior and belief formation. His research involved direct collaboration with major platforms, including Meta and Twitter (now X), presenting recommendations to their policy teams.

He proposed a social epistemological framework, meaning truth isn’t learned individually; rather, it is learned through social processes such as networks, communities, and shared belief systems.

“Most of what we learn isn’t from books — it’s socially learned. We have to think about the communities we’re engaging with, not just the content we’re consuming.” — John Wihbey

Wihbey identified several paradoxes of the current AI and misinformation landscape:

• AI saturation paradox: Platforms warn about AI-generated misinformation even as they promote AI creator tools.
• Trust paradox: Users report lots of distrust and anxiety towards AI, yet continue to use it.
• Transparency paradox: More transparency doesn’t always yield more trust. Too much exposure can erode confidence.

He warns of the implied truth effect, which is when misinformation is flagged, unflagged content may seem more trustworthy simply because it’s unmarked. He also describes the “liar’s dividend,” which is when an individual’s hyper-vigilance about truth leads them to start to doubt all of reality.

Given these challenges, Wihbey argued for thinking more systematically about social epistemological goals. He outlined three specific ways platforms could measure social epistemic health:

1. Improving the balance of accurate versus misleading content in users’ feeds.
2. Helping users recognize credible sources and expertise, improving users’ ability to curate their feeds by making better choices in who they follow or unfollow.
3. Encouraging users to seek out a wider range of viewpoints on disputed topics.

Wihbey proposed several tools for measuring these goals. First, the Technology Acceptance Model (TAM) can be used as a tool for understanding how users evaluate new technologies, measuring ease of use, trust, and perceived usefulness. He also suggested approaching users as learners and incorporating cognitive science education so people can understand how they process information from digital media. Finally, he emphasized the social epistemology theory and the importance of becoming aware of how the communities we engage with are helping or hurting us.

Disinformation, Echo Chambers, and Generative AI

Jennifer Gradecki and Derek Curry approach technology from a critical art practice, creating interactive installations that demonstrate how AI can be used for manipulation and disinformation. Their projects inform viewers about the capabilities of these technologies and create space for reflection.

Through their research, they found that disinformation typically takes two forms: information that is true but presented outside its original context to be weaponized, or value-laden judgments that can’t be proved true or false either way. Both types are difficult to debunk because they require critical thinking from the person receiving the information. This challenge is heightened by systems and algorithms that reward engagement and outrage, creating echo chambers that circulate on social media.

Gradecki and Curry’s artistic practice directly engages with these issues through two interactive installations. “Epic Sock Puppet Theater” uses animatronic sock puppets to present viewers with actual disinformation from real social media campaigns in a playful manner where viewers can reflect on what they’re seeing and hearing rather than just react to it. The project combines disinformation datasets from various sources, with the largest coming from the Russian Internet Research Agency’s 2016 campaign in the US, released as part of the Mueller report.

Epic Sock Puppet Theater

Their second project, “Generative Persuasion,” allows viewers to use generative AI to create disinformation targeting specific political orientations and personality traits from the OCEAN profile (openness, conscientiousness, extraversion, agreeableness, or neuroticism).

Gradecki explained that this project emerged from their research into how generative AI is being used in political disinformation campaigns. In August 2024, OpenAI reported that they had disrupted over 20 disinformation operations that were using ChatGPT, mostly carried out by foreign governments attempting to interfere with democratic elections. These campaigns employed tactics designed to divide and polarize people. For example, an Iranian operation generated content about the US presidential election, and Rwandan accounts generated partisan tweets ahead of elections in their own country. These campaigns often used both sides of an argument to further polarize people.

“Generative Persuasion” is presented as a fictional startup named “Psybernetica,” inspired by the Cambridge Analytica scandal. Cambridge Analytica was the British consulting firm that used five-factor personality quizzes to correlate social media activity with personality traits to psychographically microtarget the public during the 2016 Brexit and Trump campaigns. The project demonstrates how easy it is to generate disinformation and bypass content filtering and safety measures. It uses LM Studio to run models locally, including Meta’s Llama 3, which was trained on social media data and is particularly effective at producing social media content. The installation is presented in a military-style portable computing case to emphasize that it’s a weaponized LLM. The app is not and will not be made available online, so people can’t actually use it to make a disinformation campaign.

Generative Persuasion

Gradecki also addressed potential solutions and their limitations. She outlined two approaches: stopping disinformation from being posted (through negative labeling like flagging, demonetization, and deplatforming) and stopping it from spreading (through fact-checking and content provenance authentication). However, these methods have significant drawbacks. Negative labeling is criticized as undemocratic, and disinformation can still spread if it hasn’t been found. Provenance authentication can be hacked, and fact-checking relies on specific epistemologies that not all communities will accept.

Gradecki illustrated this point with research on Flat Earth conspiracy echo chambers. The study identified three different groups, each with distinct epistemological approaches:

1. Groups that value Bible verses as evidence, creating a polarizing logic between Christians and non-Christians.
2. Groups that trust only personal experiences (“I don’t feel like I’m spinning”), setting up an in-group and out-group of the public versus intellectuals.
3. Conspiracy theorists who felt powerful people were hiding information, positioning themselves as truth-tellers against journalists, scientists, and politicians.

Curry explained that there has been a push within the industry for technological solutions to distinguish truth from non-truth. The Coalition for Content Provenance and Authenticity, founded by Adobe, The New York Times, and Twitter in 2019, introduced a C2PA badging system. However, such solutions can be counterproductive. As cybersecurity professionals noted, it functions as an “honesty badge in an ecosystem dominated by bad actors who have no incentive to play by the rules.” Nikon recently revoked its C2PA image authenticity certificates after vulnerabilities were exposed.

Ultimately, Gradecki and Curry’s work underscores a central challenge: without changing the targeted advertising business model that promotes sensational content to hold attention, technological solutions will remain only minimally effective.

The Legal Perspective

Following the presentations, the discussion opened to audience questions, revealing how these issues resonate across different fields and disciplines.

Law professor Dan Jackson raised the issue of evidentiary reliability. In a legal context, before evidence can be presented to a jury, a judge must determine whether it’s sufficiently reliable. Thirty years ago, if a photo were used as evidence, establishing reliability was straightforward. Now with AI-generated images, this process is becoming increasingly challenging.

Jackson asked how close we are to losing the ability to “look under the hood” of technology to determine if content is real or fake. Gradecki responded that it’s an arms race. When they started working with AI-generated images in 2020, it was easy to tell when something was AI-generated. Now, tools are being developed to fix the smaller details, adding texture and grit, or to ensure the correct number of fingers, making generated images increasingly convincing and able to fool the eye. Curry added that although AI-generated images are based on probability rather than physics, unrealistic parts can be fixed with control nets, where an existing image provides the underlying structure. The technology exists today to create fully convincing false images.

Jackson acknowledged that the ability to create false evidence has existed for centuries, and courts have become adept at distinguishing true from false. In the past, a good lawyer could cross-examine a fake photograph by looking at the details. The concern now is reaching a place where those inquiries are no longer possible. Curry noted that artists have even created extra fingers people can wear to maintain plausible deniability if caught on camera committing crimes. The conversation turned to future implications for courts, where visual evidence might one day be deemed too unreliable, with courts relying only on first-person testimony.

Gradecki emphasized that this situation pushes us toward a different version of skepticism where we need to approach everything critically, making media and information literacy more important than ever.

Understanding Belief and Echo Chambers

The discussion then turned to why people believe conspiracies and how to get them out of echo chambers. Gradecki explained that there’s often a “kernel of truth” behind conspiracy theories, as she put it:

“There’s something behind it, meaningful and valuable — people just trying to understand the world.” — Jennifer Gradecki

The root of the issue is people wanting to feel like they understand the world without being exploited by powerful people. She also noted how it feels good to feel confirmed and feel right about something, which is why conspiracy theories can be so compelling.

Meyer raised what she called the paradox of truth: people want to believe that something is true because it gives them information and pleasure. It could also be that believers need a narrative that justifies their actions. The feeling of being right and knowledgeable is powerful.

Wihbey added nuance to this discussion, noting that the big headline issues, such as who won the 2020 election or whether climate change is real, present binary choices. But reality is much messier:

“Most claims online are kind of true and kind of not true… rumors are very important in terms of orienting us in our environment and our world and our community.” — John Wihbey

He suggested we should be cautious about thinking there would be any kind of big solution to the problem of truth online; it’s always going to be shades of gray.

Wihbey also wondered about the disconnect between how people behave online versus in their everyday lives. Online polarization is understood through data and content, but perhaps people are more grounded and truthful in their everyday experiences. The performative nature of online engagement might not reflect how people actually live.

Authority, Transparency, and Declining Trust

The discussion revealed deeper questions about trust and transparency. Dietmar Offenhuber drew parallels to C-SPAN: the idea that broadcasting congressional debates would improve democracy by allowing people to verify what their representatives said. Instead, it generated extreme polarization. As Meyer quipped, “Every dictator was super transparent — you don’t agree, and they kill you.” Transparency alone doesn’t guarantee accountability.

Wihbey added that when Meta first introduced AI labels on content, the creator community complained the most. Creators didn’t want people to think their content was suspicious because it diminished their economic value and audience trust. Well-intended transparency measures can backfire.

A question raised another critical dimension that authority figures, whom we normally trust, are circulating untruthful content. Trump’s weaponization of “fake news” became a way to get people to question what is true, particularly when it’s information someone in power doesn’t want people to know.

Wihbey contextualized this within broader trends. Survey data shows that the decline in trust in institutions (media, science, Congress) has been happening since before the internet, though it’s been accelerated by social media and generative AI. Much of this decline can be explained by political elites themselves criticizing institutions, from Nixon through Clinton and Trump.

Meyer suggested that data overflow might explain why misinformation and rumors are so popular. They feel like privileged information in an ocean of content. Wihbey wondered how younger generations perceive these AI-saturated platforms like TikTok and Instagram, where short-form video platforms have no way of verifying content, and whether this might create a backlash.

Accountability and Solutions

Questions of responsibility and potential solutions emerged as central concerns. One question addressed the ethics of training AI on existing artworks and protecting authorship. Gradecki pointed to Adobe’s recent policy of only using approved artwork in training their AI models as a positive development, while Curry noted that although artists have copied other artists for centuries, current copyright laws are likely behind the technology.

When asked how to hold companies like Meta accountable, Meyer reframed the question: Who is responsible? Companies, developers, users? Gradecki emphasized that responsibility is multilayered. Developers are responsible for how quickly they put technologies into the public. The public is responsible when they share and exacerbate misinformation. Curry raised the character.ai case, where a young boy was told to kill himself by an AI-generated character, asking where liability falls in such situations.

Jackson introduced the concept of strict liability, similar to how industrial revolution companies were held liable for dangerous machinery without having to prove causation. Wihbey mentioned colleague Laura Edelson’s theory that platforms should be regulated as products since many features are patented and could potentially be treated under product liability frameworks.

When asked about best-case scenarios for solving these problems, Gradecki argued that true solutions lie in literacy and critical thinking, especially given that different communities trust different authorities. More fundamentally, the targeted advertising business model must change. As long as platforms promote controversial content to hold attention, other solutions are just band-aids.

Curry suggested that younger generations might develop a healthy level of skepticism where people don’t feel they need to know everything or determine if everything is true. The question becomes: is this particular truth relevant, and if so, how do we decide?

The conversation ultimately reinforced a central tension: we live in an age that demands both greater skepticism and greater media literacy, yet the systems designed to help us navigate truth often create new problems. As the panelists made clear, there are no simple technological fixes to the challenges of misinformation and AI-generated content. The solutions require fundamental changes in business models, in education, and in how we think about responsibility and trust. Understanding how truth operates in this algorithmic age means recognizing that it’s not just about the technology itself, but about the social, economic, and political contexts in which these technologies exist.

Interested in learning more? Watch the event recording:

[embed]

CAMD Moderator:

Sylke Meyer: Professor, Theatre and Art + Design, CAMD; Co-Founder, Studio206

Speakers:

Derek Curry: Associate Professor, Art + Design, CAMD; Co-Founder, Bankster Games

Jennifer Gradecki: Associate Professor, Art + Design, CAMD; Co-Founder, Bankster Games

John Wihbey: Associate Professor, Journalism, CAMD; Director, AI-Media Strategies (AIMES) Lab; Co-Founder, Institute for Information, the Internet, and Democracy (IIID)

Join us! #CenterForDesign

️📩 centerfordesign@northeastern.edu 🔗LinkedIn 🔗Twitter 🔗Instagram 🔗Facebook 🔗Website 🔗Newsletter