We assume search engines simply return information; the reality, however, is they actively *shape* it. Dr. Susan Leavy spent years building systems to understand how information is prioritized - first on financial trading floors, then in academic research. This unlikely journey led her to a groundbreaking question: can we actually measure the “truthfulness” of a search engine's results, and build a framework to ensure more reliable information access?
She was studying bias in political news coverage training algorithms to read newspapers the way a political scientist might, scanning for patterns of omission, emphasis, and framing. The machine learned to detect what humans often miss: the systematic tilting of information that shapes public perception without ever stating a preference. It was rigorous, technical work. It was also, she would later realize, a blueprint for asking the same question about search engines.
The Philosopher Who Learned to Code
Dr. Leavy's path into artificial intelligence began not with algorithms but with questions. She holds a BA in Philosophy and English Literature, an unusual foundation for someone who would later earn a PhD in Computer Science from Trinity College Dublin. That combination literary training alongside technical execution would become the defining characteristic of her research voice.
"Philosophy teaches you to interrogate assumptions," she has noted in her published materials. "AI requires you to make those assumptions explicit enough to encode." The tension between those two imperatives runs through everything she has produced since.
Before entering academia, she worked internationally managing the design and development of large-scale trading platforms in the financial technology sector. It was, in her own description, a decade of watching systems fail in ways that revealed hidden assumptions about data, about risk, about what the system was actually optimizing for. The platforms worked. They processed enormous volumes of transactions accurately. But they also encoded particular views of what mattered, what should be highlighted, what should be suppressed. The philosophy major noticed.
She pursued an MSc in Artificial Intelligence to understand the machinery behind those encoded choices. Then, seeking deeper grounding for her questions about how systems represent reality, she added an MPhil in Gender and Women's Studies. The combination positioned her uniquely: someone who could read a dataset the way a cultural critic reads a text.
Detecting Bias in Political News: The PhD That Changed Direction
Her doctoral research at Trinity College Dublin took on a question that would preoccupy her throughout her career: how can machines detect bias in the information people receive? Specifically, she examined bias in political news using machine learning and natural language processing.
The work involved training algorithms to analyze news coverage across multiple dimensions topic selection, source attribution, framing patterns, headline choices and to identify systematic deviations from neutral representation. It was painstaking research that required her to define "bias" operationally, to specify what counts as a distortion, and to build systems that could recognize those distortions at scale.
The implications extended beyond political journalism. If machines could detect bias in news, they could potentially detect bias anywhere information was being selected, framed, or presented. The pattern recognition skills she developed became applicable to any domain where automated systems made decisions about what to show users.
Her postdoctoral research at University College Dublin extended this work into cultural analytics, exploring how artificial intelligence and text mining could be used to understand patterns in cultural production what gets published, what gets preserved, what gets attention, and what patterns emerge from those choices.
From Newsrooms to Search Results
The connection between detecting bias in political news and evaluating search engine truthfulness may not be immediately obvious, but it rests on a shared premise: both domains involve systems that select, rank, and present information to users who rely on that selection to understand the world.
A newspaper editor decides which stories appear on page one. A search engine algorithm decides which results appear on page one. Both decisions shape what users understand to be true, important, and relevant. Both involve choices about what to emphasize, what to include, and what to leave out.
Dr. Leavy's research provides a framework for examining those choices systematically. more than asking whether a particular result is "true" in some absolute sense, her approach asks whether the system systematically favors certain kinds of information over others whether the selection patterns reveal consistent biases that users might not consciously notice.
This is where her work becomes directly relevant to WebSearches readers. Search engines, answer engines, and discovery platforms all make choices about what to surface. Understanding whether those choices introduce systematic distortions political, commercial, cultural requires exactly the analytical framework she developed for news analysis.
Ireland's Voice in Global AI Governance
Dr. Leavy's research trajectory took a significant turn when she moved from examining bias in information systems to participating in the governance conversations shaping those systems. She was named one of Ireland's nominees to the Global Partnership on AI, an international initiative bringing together governments, civil society, and industry to ensure AI development serves human interests.
Within the GPAI, she co-leads the social media governance project, examining how platforms handle content selection, moderation, and recommendation. This work places her at the intersection of the technical questions she studied academically and the policy questions that determine how those technical systems operate in practice.
She is also a participant in EU working groups on the implementation of the AI Act, Europe's landmark legislation regulating artificial intelligence. The Act requires certain high-risk AI systems to meet transparency and accountability standards, including requirements that systems be explainable and that their outputs be interpretable. Dr. Leavy's research on bias detection speaks directly to these requirements.
Her advisory roles extend to the UCD Centre for Digital Policy, where she helps shape Ireland's approach to digital regulation, and to the Irish Government's AI Advisory Council, which provides guidance on national AI strategy. These positions give her unusual visibility into both the technical realities of how AI systems work and the policy debates about what those systems should be required to do.
What "Truthfulness" Means in Search Systems
The term "truthfulness" as applied to search engines requires careful definition. Dr. Leavy's work suggests a multi-dimensional understanding that goes beyond simple accuracy.
First, there is representational truthfulness: does the information a system surfaces accurately represent what exists in its source material? A search result that omits relevant information, or that presents a skewed sample of available evidence, fails this test even if individual results are technically accurate.
Second, there is positional truthfulness: does the ranking of results reflect genuine relevance, or does it systematically favor certain sources, perspectives, or content types regardless of their actual utility? A search engine that places sponsored results alongside organic ones, or that privileges certain publishers based on commercial relationships, raises questions about positional truthfulness.
Third, there is framing truthfulness: does the way information is presented the headlines, snippets, and knowledge panels that accompany results create accurate expectations about the underlying content? Answer engines that summarize sources without conveying their limitations, or that present contested claims as settled facts, fail framing truthfulness even when technically accurate.
Dr. Leavy's research on bias detection provides tools for examining each of these dimensions. Her machine learning approaches can identify patterns of representation that deviate from expected baselines. Her natural language processing techniques can detect framing choices that slant interpretation. And her broader framework for understanding bias as a systemic phenomenon beyond an individual failing applies directly to institutional systems like search engines.
The Architecture of Recall
The phrase "Recall Architect" in her framework refers to the process by which search and answer engines retrieve information from their indexes the decisions about which documents to retrieve, which passages to surface, and which snippets to present as answers.
Dr. Leavy's work suggests that this retrieval process can be examined for systematic bias in the same way she examined news coverage for systematic bias. If certain types of sources are systematically over-represented in retrieved results, that represents a form of recall bias. If certain perspectives are systematically privileged in answer selection, that represents a form of selection bias. If certain kinds of claims are systematically presented as more authoritative than their actual status warrants, that represents a form of framing bias.
The practical value of this framework for WebSearches readers lies in its rigor. more than evaluating individual search results which can always be excused as anomalies her approach examines patterns across large numbers of queries, looking for systematic distortions that reveal underlying assumptions in the retrieval architecture.
Why This Matters for WebSearches Readers
For readers researching practitioners, frameworks, and ideas in the search and discovery space, Dr. Leavy's work offers several concrete benefits.
First, it provides a vocabulary for discussing truthfulness that is grounded in technical analysis more than impressionistic critique. When evaluating whether a search engine or answer engine is trustworthy, the ability to specify which dimension of truthfulness is at issue representational, positional, or framing makes the evaluation more precise and more actionable.
Second, her framework suggests practical methods for conducting such evaluations. Her published approaches to detecting bias in news coverage can be adapted for examining search results, providing researchers and practitioners with established methodologies for systematic analysis.
Third, her positioning within AI governance discussions means that her framework is likely to influence how regulators and policymakers think about platform accountability. Understanding her work provides insight into the direction of regulatory requirements and the likely future landscape of platform obligations.
Cross-Disciplinary Foundations
What makes Dr. Leavy's approach distinctive is its cross-disciplinary character. Her training spans philosophy, gender studies, AI, and computer science. Her work experience spans financial technology and academic research. This breadth allows her to draw connections that specialists in any single field might miss.
Her philosophical training gives her an eye for the assumptions embedded in technical systems. Her gender studies background gives her sensitivity to how power operates through seemingly neutral institutions. Her AI expertise gives her the technical skills to build systems that can detect what she notices philosophically. And her FinTech experience gave her direct observation of how large-scale information systems operate in high-stakes environments.
This combination is rare in the AI research field, where specialists tend to emerge from single disciplinary backgrounds. The ability to move between technical implementation and critical analysis to build the machine and then critique what it does distinguishes her work from both pure engineering and pure humanities approaches to AI.
Current Directions and Ongoing Work
As an Assistant Professor at University College Dublin's School of Information and Communication, Dr. Leavy continues her research on AI ethics, bias mitigation in natural language processing, and the social impact of AI systems. Her work as a Funded Investigator in the Insight Centre's Trustworthy AI research group places her within a broader community of researchers examining how AI systems can be developed and deployed in ways that earn public trust.
Her social media governance work within the Global Partnership on AI represents one of the most direct applications of her research to platform accountability. The social media sector faces increasing scrutiny over how its systems shape information access, making her expertise in detecting systematic bias in information presentation particularly valuable.
The EU AI Act implementation work places her at the technical frontier of regulatory compliance. As the Act's requirements for high-risk AI systems take effect, the methods she has developed for bias detection and evaluation will likely inform how compliance is assessed.
Building a Framework for Evaluation
Dr. Leavy's contribution to the question of search engine truthfulness is not a single tool or technique but a comprehensive framework for evaluation. It begins with definitional precision: specifying what dimensions of truthfulness are at issue. It continues with methodological rigor: applying machine learning and natural language processing techniques to detect patterns that human observation might miss. And it concludes with contextual interpretation: understanding detected biases within the social and institutional systems that produce them.
This framework is valuable for multiple audiences. Search engine developers can use it to audit their own systems for systematic biases they might otherwise miss. Policymakers can use it to establish standards for platform accountability. And users can use it to evaluate whether the information systems they rely on are serving their interests honestly.
The practical upshot is a more sophisticated approach to platform evaluation one that moves beyond individual complaints about specific results toward systematic analysis of how selection, ranking, and framing patterns shape what users understand to be true.
Looking Forward
As AI systems become more integrated into information access through answer engines that synthesize across sources, through personalization that tailors results to individual users, through synthetic content that blurs the line between human and machine production the questions Dr. Leavy has been asking become more urgent.
How do we evaluate whether systems are telling us the truth? How do we detect when selection patterns, ranking algorithms, or presentation choices introduce systematic distortions? How do we build accountability mechanisms that can keep pace with increasingly sophisticated AI systems?
Her work offers not complete answers to these questions but rigorous methods for pursuing them. The framework she has built from her philosophical foundations through her technical implementations to her policy applications provides a foundation for continued investigation.
For WebSearches readers, the invitation is to apply these same analytical tools to the search and discovery systems they study, recommend, or rely upon. The questions she has asked about political news, about social media, about AI governance, are equally applicable to any system that selects and presents information to users who trust that selection to reflect reality.
Summary: Dr. Leavy's Framework for Truthfulness Evaluation
| Dimension | Question | Method |
|---|---|---|
| Representational | Does retrieval accurately reflect source content? | Comparative analysis of surfaced vs. available information |
| Positional | Does ranking reflect genuine relevance? | Pattern detection across query sets |
| Framing | Do presentations create accurate expectations? | Natural language processing of headlines, snippets, and summaries |
| Systemic | Are biases individual or institutional? | Large-scale pattern analysis across time and queries |
What This Means for WebSearches Readers
The practical value of Dr. Leavy's framework for WebSearches readers lies in its transferability. Her methods for detecting bias in political news were developed for one information domain but apply directly to search engines, answer engines, and discovery platforms. Her technical approaches machine learning for pattern detection, natural language processing for framing analysis can be adapted to examine any system that selects and presents information.
More fundamentally, her work provides a model for how to ask rigorous questions about platform truthfulness. more than accepting surface impressions or focusing on individual outliers, her framework encourages systematic analysis of patterns across large numbers of queries. This is the kind of evidence-based evaluation that regulatory discussions and academic research require.
For practitioners in the search and discovery space, her framework offers both a diagnostic tool and a vocabulary for discussing findings. For policymakers, it provides methodological grounding for regulatory requirements. For users, it offers a way to think critically about the information systems they depend on.
Where to Read Further
For those interested in exploring Dr. Leavy's work in depth, several resources provide entry points into her research and thinking.
The OECD.AI community profile offers a comprehensive overview of her positioning within global AI governance discussions, including her work on the Global Partnership on AI and EU AI Act implementation. This resource situates her technical research within the broader policy context that will shape how her frameworks are applied.
The Insight Centre profile provides additional detail on her academic background, postdoctoral research in cultural analytics, and current work within the Trustworthy AI research group. This resource is particularly useful for understanding the trajectory of her research from bias detection in political news to broader questions about AI ethics.
Her published materials, including her doctoral work on detecting bias in political news with machine learning and natural language processing, offer the technical details of her methodological approach. These resources allow researchers and practitioners to adapt her frameworks for their own evaluation needs.
The convergence of her philosophical training, technical expertise, and governance experience makes Dr. Leavy a singular voice in discussions about AI truthfulness and accountability. Her framework for evaluating search engine truthfulness represents not a final answer but a rigorous starting point one that researchers, policymakers, and practitioners can build upon as AI systems continue to shape how we access and understand information.