We live in a paradoxical era. On the one hand, data is growing exponentially, computational modeling and artificial intelligence are advancing at breathtaking speed, and knowledge is deeper and more specialized than ever before. Yet on the other hand, rates of transformative discovery appear to be slowing rather than accelerating across important domains, and crises of replication and generalizability seem to be intensifying. Across the world, research outputs remain stubbornly misaligned with the problems societies—and people—describe as most important to them. In short, we are living through an era of both epistemic abundance and epistemic crisis.
Several explanations are offered for this apparent gap. Misinformation is said to be eroding the foundations of shared knowledge. Cultural polarization is often cited as a barrier to the collaborative inquiry and pursuit of objectivity on which science depends. Outdated incentive structures—publish or perish, funding biases, information overload—are also mentioned. Each of these no doubt plays a role. But an equally important source of the problem lies further upstream, at a stage of knowledge production that receives surprisingly little systematic attention: the questions we ask.
Across sectors and fields, many of society’s most significant shortcomings today stem not from insufficient data or inadequate analytical tools but rather from the framing of inquiry itself. Scientific agendas are shaped by questions that are too narrow and overly method-driven. Policy is constrained by electoral cycles and available metrics. Even the private sector, for all its celebrated dynamism, asks questions bounded by existing business models and short-term returns.
The consequences of such misdirected inquiry are not abstract. It is now well established that women’s health remained systematically underfunded and understudied for decades--not because the tools to investigate it were lacking, but because the questions that shaped biomedical research agendas were framed around male bodies and male norms. Likewise, pandemic preparedness in the pre-COVID era was repeatedly optimized for speed of vaccine development while neglecting questions about distribution, trust, information, and misinformation. In both cases, the answers developed by experts were technically competent. The questions they asked were wrong, or at least incomplete.
All these shortcomings acquire extraordinary new urgency in the age of artificial intelligence. As AI systems become increasingly powerful “answering machines,” capable of generating immediate, plausible-sounding responses to almost any query, the bottleneck in knowledge creation is shifting: from generating answers to formulating the questions worth answering in the first place. The cheaper answers become, the more valuable questions are.
For all the sophistication of our analytical tools, for all the power of our new technologies, and for all our acute and pressing social needs, we still have no science of questions: no systematic infrastructure for studying how questions emerge, how they can be improved, or how they shape the knowledge that societies produce. This essay argues that we need such a science of questions. It traces the intellectual roots of a question-centered approach to knowledge (Section I), diagnoses the institutional failures that perpetuate what we call a “question deficit” (II), examines the politics of who gets to frame inquiry (III), and outlines the key elements of a science of questions (IV)—one made both possible and urgent by the rise of AI.
I. Questions Before Answers
The lack of a science of questions is somewhat surprising, because the idea that inquiry is cognitively primary--that questions come before answers, shaping how we think--has deep roots in philosophy and the history of science. Karl Popper, for instance, argued that scientific progress begins with the formulation of bold, falsifiable hypotheses, not with the accumulation of answers. Thomas Kuhn similarly—and famously--showed that paradigm shifts emerge precisely when existing questions and frameworks fail to capture the phenomena under study.
This work has been extended by a more recent generation of scholars, who have pushed the study of questioning into new disciplines. The philosopher Jane Friedman treats inquiry not merely as a precursor to belief but as a primary cognitive orientation: a way of engaging with the world that deserves study in its own right. Philipp Koralus, in his “erotetic theory” of reasoning, suggests that questions provide a “search space” or a “frame” that dictates what information is perceived as relevant and what we screen out. Change the question, in this account, and you change the reasoning process, and thus the ultimate conclusion.
Herbert Simon, the famous American psychologist, arrived at a related insight from a different direction. “Solving a problem,” he wrote, “simply means representing it so as to make the solution transparent.” For Simon, the architecture of a problem—the way it is framed and decomposed—is not merely a preliminary step before the real work begins; it is the real work. If a question is architected well, the solution often follows as a natural consequence of that representation. If it is architected poorly, no amount of data or computational brute force can compensate.
These are not merely philosophical or theoretical observations. They have tremendous practical consequences for how some of the most important research is evaluated and ultimately funded. The implication is stark: if the quality of discovery depends on the quality of questions, then it is essential to scrutinize the upstream processes of inquiry—the assumptions, the framing, the selection of what to investigate—with the same rigor we bring to methods and data. Institutions that invest heavily in methods for generating answers while neglecting the processes through which questions are sourced and prioritized are, ultimately, building on an unstable foundation.
Yet that, as we argue in the next section, is precisely what too many of our institutions do.
II. The Question Deficit and Why it Matters
Consider the typical life cycle of a research agenda. First, funding agencies issue calls structured around existing disciplinary categories. Researchers formulate proposals in response, often building incrementally on previous findings in the field. Peer reviewers then evaluate the proposals, again typically drawing on existing disciplinary conventions. At no point in this process does anyone systematically ask: Are these the right questions? Are we investigating the problems that matter most? Whose perspectives are shaping the direction of inquiry, and whose are being excluded?
The result is a knowledge ecosystem characterized by what might be called a question deficit: rich in analytical capacity and the data to support it, but impoverished in the upstream processes of inquiry design. Scientific communities ask questions shaped by disciplinary inertia and the path dependencies of prior research. Policymakers ask questions whose contours are dictated by bureaucratic mandates and available metrics. Philanthropies and development organizations ask questions framed by existing theories of change. Even the private sector, long celebrated as an engine of innovation and creative destruction, is constrained by groupthink and the gravitational pull of existing business models and theories. Across sectors, questions that would require new frameworks, a willingness to return to first principles, or the incorporation of outsider and non-expert perspectives are systematically underproduced--even ignored.
This pattern is visible across domains. As mentioned, women’s health has remained systematically underfunded for decades because research questions were framed around male bodies. Neglected tropical diseases affecting hundreds of millions of people receive a fraction of research investment because the questions driving pharmaceutical R&D reflect the disease burdens of wealthy countries. Similar patterns are apparent with regard to Indigenous knowledge, as well as systems and information relevant to global climate adaptation. The James Lind Alliance, a British non-profit, has documented striking divergences between the questions patients consider most urgent and the questions researchers actually pursue. All of these are evidence that the question deficit is not merely hypothetical but real and measurable.
At a high level, this deficit matters because questions are not neutral instruments. They are agenda-setting and even political devices, powerful mechanisms that determine what becomes visible, measurable, and fundable. The act of formulating a question defines the boundaries of investigation. It determines which variables are considered relevant and which are bracketed out, which populations are studied and which are ignored, and which outcomes are measured or go unrecorded. In this sense, as we argue in the conclusion, questions function as a form of epistemic infrastructure: they support and channel the direction of inquiry and knowledge itself.
III. Power, Politics, and Participation in Questions
Like all infrastructure, questions tend to become invisible once they are in place. We tend to scrutinize the quality of answers, but we treat the questions that generated those answers as given, as background noise. Despite the centrality of questions, the upstream moment at which inquiry is framed remains largely unexamined.
The invisibility of questions in our epistemic framework has a social and even political dimension. If questions set agendas, then the processes by which questions are formulated are really a form of power. They determine which questions matter—and, even more importantly, whose perspectives matter, and which needs or demands should be prioritized. The politics of questions also determines which issues remain invisible because they are never articulated in the first place.
It is important to understand how this exercise of power takes place in practice. In most fields, question-setting tends to occur within narrow expert communities. The questions that shape drug development are typically formulated by pharmaceutical researchers and clinicians, not by the patients whose lives depend on the outcomes or directly experience the drugs they are prescribed. The questions that drive urban planning are posed by planners and engineers, not the communities who inhabit the resulting environments. Likewise, the questions that frame AI safety research are defined by a small number of labs and academic groups, with limited input from the users and societies that will ultimately bear the consequences of how the technology is deployed.
None of this is to suggest that expertise is unimportant. The problem is not expertise per se but rather the exclusion of other forms of knowledge--or lived experience--from the question-setting process. When inquiry is shaped exclusively by disciplinary experts, then competing values, local knowledge, and marginalized perspectives are systematically screened out. The result is research that may be technically rigorous but socially exclusionary—and which, ultimately, ends up perpetuating or even exacerbating existing social, economic and political inequalities.
Some promising models for broadening participation in inquiry do exist. The James Lind Alliance, a UK-based initiative, has developed a methodology for “Priority Setting Partnerships” that bring patients and clinicians together to identify the research questions that matter most to both groups. The Open Innovation in Science (OIS) framework, pioneered by Marion Poetz and Susanne Beck, likewise suggests that scientific progress is often bottlenecked by a “closed” inquiry process and demonstrates that opening question formulation to a wider range of participants can yield fundamentally new lines of investigation.
These initiatives are valuable, but isolated. They represent experiments in participatory questioning rather than a systemic shift in how inquiry is designed. What is needed is a more comprehensive effort to make question-setting transparent, inclusive, and subject to the same rigor we apply to other stages of the research process. In short, as we argued above, we need a science of questions. We turn to the elements of that science in the following section.
IV. Toward a Science of Questions–an Intellectual and Practical Agenda
All of these dynamics are now being reshaped—and, in many cases, intensified—by the rapid advancement of artificial intelligence. AI is flooding the knowledge ecosystem with answers, making the question deficit more consequential than ever. It is concentrating the power to frame inquiry in the hands of a small number of technology companies and research labs. Finally, it is accelerating the pace at which poorly framed questions generate confident-sounding but misleading outputs, compounding the costs of getting the upstream framing wrong.
For most of human history, answers were scarce and expensive, while questions were abundant and cheap. Our institutions reflect that asymmetry: universities, laboratories, funding agencies, and peer review systems are all fundamentally answer-producing machines. AI is upending this arrangement. As the marginal cost of generating an answer approaches zero, the scarce and valuable resource is no longer analytical capacity but the ability to formulate questions that direct inquiry toward problems that actually matter. Our institutions have not yet adapted to this shift—which is why we need, now more than ever, a science of questions.
What would it mean to take questions seriously—not as a rhetorical flourish, but as a subject of rigorous study and institutional practice? This is a line of inquiry that we have been pursuing at the GovLab for some years now, through multiple initiatives and lines of work. We have identified six key conceptual strands of this nascent science, and several concrete actions that could advance the practice of the science.
Questions as a dynamic process. The most productive lines of inquiry do not begin with a perfectly formed question. They evolve through interaction with data, dialogue, experimentation, and reflection. A science of questions would develop methods and institutional frameworks for refining, architecting and reformulating inquiry over time, rather than treating the question as fixed at the outset.
Taxonomies of questions. Not all questions are equal. Some are marginal, generating incremental improvements within established paradigms, while others are foundational, opening new domains or challenging core assumptions. Still others are exploratory, emerging from observation and data, and some are normative, shaping how societal issues are framed. Better taxonomies would allow institutions to assess whether their inquiry portfolios are concentrated on incremental work or enabling transformative shifts.
Questions as agenda-setting devices. Questions determine what becomes visible, measurable, and fundable. A science of questions would study how questions function as governance mechanisms: how they allocate attention, shape priorities, and render certain problems legible while leaving others unexamined. Understanding this agenda-setting power is a prerequisite for making it more transparent and accountable.
Participatory question-setting. If the quality of inquiry depends on a diversity of perspectives, then the processes through which questions are formulated need to be opened up. This means developing rigorous methods for collective intelligence, priority-setting partnerships, and cross-disciplinary inquiry design. Promising avenues include citizen assemblies for research prioritization, structured collaborations between affected communities and domain experts, and asynchronous processes for public input on critical questions for specific domains.
The psychology of questioning. Emerging research suggests that effective questioners share certain traits: nonconformity, intellectual independence, an orientation toward impact. A science of questions would study these not to restrict questioning to certain personalities but to understand how institutions can cultivate environments where diverse questioning styles flourish.
AI and the ecology of questioning. Even as AI intensifies the need for better questions, it can also help meet that need– e.g., by identifying gaps, surfacing overlooked connections, and stress-testing the framing of inquiries. A science of questions also needs to be mindful of where AI falls short—for instance, areas in which human judgment is required to determine which questions are socially consequential and representative of diverse interests. The goal is not to replace human questioning but to develop a productive division of labor, and ultimately to augment inherently human capacities.
Practical Agenda: The preceding strands define an intellectual agenda. But a science of questions will only matter if it changes how institutions actually operate. Several concrete actions could contribute to this science and begin closing the question deficit:
Require “question statements” in grant proposals, subjected to the same scrutiny currently applied to methods sections.
Fund question-generation and priority-setting as standalone activities, not just as add-ons.
Establish question labs within research and policy organizations: dedicated spaces for the study and practice of questioning.
Evaluate research portfolios on the diversity and ambition of their questions, not only on publication counts and citation metrics.
Create public repositories of unanswered questions: open, searchable, and organized by domain.
Develop AI tools that audit and challenge dominant research framings, surfacing blind spots and identifying absent questions.
V. Questions as Scientific and Societal Infrastructure
As a society, we have invested enormously in our capacity to generate answers. We have built vast institutions for data collection, analysis, and dissemination. We have developed increasingly sophisticated tools for computation and modeling. What we have not done—what we have barely begun to do—is invest in the upstream processes through which the questions that direct all of this effort are identified, sustained, and governed.
Questions deserve to be understood as infrastructure. Just as physical infrastructure channels the flow of goods and digital infrastructure channels the flow of data, questions channel the direction of inquiry itself. They determine what societies are able to observe, investigate, and ultimately understand. When that infrastructure is poorly maintained—when questions go unexamined, when the processes that produce them are narrow and opaque—the consequences are not merely intellectual. They are social, political, and economic. We need to rethink our priorities: stop focusing on downstream uses, and build the supporting platforms that sustain inquiry.
As we have argued in this essay, the age of AI makes the need for a science of questions even more urgent. As our answer-generating machinery grows more powerful, the quality of the questions we feed it becomes the limiting factor for the quality of the knowledge it produces. Building a science of questions is not merely an academic exercise. It is a vital infrastructure project—one as consequential, in its own way, as the roads and digital networks on which modern societies already depend.