Beyond Boundaries: Rebuilding the Architecture of Knowledge

The dominant organizational unit of twentieth-century science was the discipline—a community of practitioners sharing a set of methods, questions, conceptual frameworks, and publication venues, with formal certification structures (doctoral programs, journal gatekeeping, tenure review) to manage membership. The discipline was an extraordinary instrument for the rapid accumulation of specialized knowledge, but it imposed characteristic costs: questions that fell between disciplines went unasked; methods from one field remained unknown in adjacent fields that needed them; and the integrative work of synthesis was systematically under-funded and under-rewarded.

Jack Oliver’s phenomenological account of scientific discovery, The Incomplete Guide to the Art of Discovery,  provides a useful framework for understanding what is at stake. Oliver argues that discovery has a characteristic cognitive structure: it requires a prepared mind with deep expertise in some domain combined with a willingness to look for anomalies in an adjacent domain where one’s existing conceptual apparatus generates an unusual angle of vision. The productive boundary-crosser is not a generalist without deep knowledge but a specialist who has deliberately expanded peripheral vision. This description captures something important: the value of interdisciplinary collision is not that all perspectives are equally valid, but that depth in one domain, combined with genuine exposure to another, generates hypotheses that neither domain could generate alone.

The Francis Crick Institute’s strategy of “Discovery Without Boundaries” represents an ambitious current institutional attempt to operationalize this insight at scale. The Institute was deliberately designed without academic departments, on the hypothesis that the organizational innovation would produce precisely this kind of disciplinary collision. Researchers from cell biology, genetics, immunology, neuroscience, and computational science share physical space and are required to participate in cross-disciplinary programs. The logic is that institutional design can serve as a partial substitute for the individual polymaths who historically performed the integrative function in science—that serendipitous collision can be engineered rather than merely hoped for.

The Derderian cilia paper illustrates both the promise and the complexity of this bet. The work draws on at least three distinct research traditions, each with its own methodological norms and standards of evidence. What makes the integration possible is precisely that the researchers have genuine depth in more than one tradition: the paper does not merely borrow concepts across disciplines but integrates methods at the level of experimental design. This is disciplinary collision at its most productive, but it required either unusually polymathic individuals or unusually effective collaborative structures. Neither is easy to scale.

The deeper tension the Crick strategy must navigate is that disciplinary specialization is not merely a bureaucratic artifact but a genuine epistemological necessity. The rigor, reproducibility, and cumulative character of scientific knowledge depend on communities that share sufficiently fine-grained methodological standards to evaluate each other’s work. Dissolving disciplinary boundaries entirely risks producing work that is creative but untestable—impressive in scope but unaccountable to the standards that make scientific knowledge reliable. The productive dissolution is not the abolition of disciplines but the creation of new integrative platforms—new question-framings and methodological vocabularies—that allow genuinely deep disciplinary knowledge to interact.

If the disciplinary boundary is primarily an epistemological problem, the institutional boundary is primarily an organizational one. The canonical institutions of twentieth-century discovery—the university laboratory, the corporate R&D division, and the national funding agency—were designed for a world of slower-moving, more legible knowledge production. Each has accumulated structural features that now impede as much as they enable: university tenure systems reward narrow depth over integrative synthesis; corporate R&D operates under quarterly reporting constraints incompatible with the time horizons of fundamental discovery; government funding agencies are structurally risk-averse, preferring incremental work in established areas to the high-variance bets that transformative discovery requires.
‍

A recent analysis by BCG of AI-powered R&D documents the institutional response emerging from the pharmaceutical sector: the rise of collaborative structures that explicitly attempt to combine the freedom and long time-horizon of academic research with the execution capacity and risk capital of industry. These are not merely licensing arrangements or sponsored research agreements but genuinely new organizational forms, in which academic scientists, computational biologists, and clinical researchers from multiple institutions share data, methods, and decision-making authority over extended periods. The organizational novelty matters: it represents an attempt to build institutional capacity for the kind of integrative, boundary-crossing work that no single sector can sustain alone.

Define Ventures’ survey of pharmaceutical C-suite executives reveals the other side of this institutional transformation: the rapid and widespread adoption of AI tools is forcing pharmaceutical companies to reconceive what they are institutionally. Companies organized around the deep chemical expertise required for small-molecule drug design are now asking whether their core institutional competence is chemistry, data science, or clinical operations—a question with profound implications for hiring, organization, and strategy. The institutional boundary between “pharmaceutical company” and “technology company” is dissolving in real time.

A recent report entitled US Resilience Resilient from Goldman Sach’s Investment Strategy Group maps the same dissolution across industries. The distinction between companies that are “in technology” and those that use technology is collapsing as digital capability becomes a precondition for operation in virtually every sector. The organizational implication is that the institutional home of innovation is dispersing—no longer concentrated in dedicated R&D divisions or specialist technology firms but distributed across the economy, embedded in operational functions that are themselves becoming sites of technical invention.

This dispersal creates a governance problem that existing institutional frameworks are poorly equipped to handle. When discovery happens within a recognizable institution—a university, a national laboratory, a pharmaceutical company—the institutional accountability structures (peer review, intellectual property law, regulatory oversight, shareholder reporting) provide, however imperfectly, a mechanism for public scrutiny and redress. When discovery is distributed across institutional boundaries or embedded in collaborative structures that span multiple organizations and jurisdictions, those accountability mechanisms weaken or disappear. This is a central challenge for global governance: how to maintain accountability for knowledge production that no longer fits within the organizational containers for which existing governance frameworks were designed.

‍

The third dissolution is in some respects the most socially consequential: the erosion of the boundary between the credentialed expert and the knowledgeable amateur. The history of science contains numerous examples of productive amateur contribution—Faraday began as a bookbinder, Darwin as a gentleman naturalist, Ramanujan as a self-taught mathematician—but the twentieth century saw a sharp professionalization of science that largely excluded non-credentialed participants from recognized scientific work. Digital infrastructure is now allowing for the reversal of this exclusion, and the question is whether the reversal is being governed well.
‍

Poetz and Sauermann in their 2024 book “How and When to Involve Crowds in Scientific Research” provide a systematic empirical account of what they call open science—the deliberate incorporation of non-expert crowds into scientific research. Their central contribution is a framework for predicting when crowd involvement adds value: it is most productive when research tasks are modular (capable of being decomposed into independently executable pieces), when evaluation criteria are explicit enough to allow non-experts to assess their own contributions, and when cognitive diversity (i.e., the value of perspectives from outside the established research community) outweighs the benefits of deep methodological expertise. On these criteria, crowd contributions are most valuable in observational data collection, hypothesis generation, and certain types of problem-solving optimization; they are least valuable in experimental design, data integration, and the high-context judgment calls that remain the core of scientific expertise.

What makes the Poetz and Sauermann framework important is not just its empirical content but its normative implication: the question is not whether to involve crowds in science but how to calibrate the boundary between expert and amateur contributions to match the cognitive demands of specific research tasks. This is precisely the kind of flexible, principled boundary management that is needed—not the rigid exclusion of non-experts, but a governance framework that identifies where their contributions genuinely add value and builds the quality-control mechanisms to make those contributions reliable.

The challenge is to hold two things simultaneously: a genuine openness to the productive contributions that non-experts can make to scientific work, and a robust defense of the epistemic standards — methodological rigor, falsifiability, peer accountability—that distinguish scientific knowledge from mere speculation. These are not in tension in principle, but operationalizing both at once requires governance frameworks that are considerably more sophisticated than either flat credentialism or uncritical democratization. The development of such frameworks is itself a form of intellectual and institutional work that the scientific community has been slow to undertake.

The most consequential and least well-governed of the four dissolutions is the erosion of the boundary between human cognition and machine intelligence in the production of scientific knowledge. This is not, strictly speaking, a new development as computation has been integral to experimental science since the mid-twentieth century, but the qualitative shift represented by large-scale machine learning, and particularly by foundation models capable of integrative reasoning across domains, represents a genuine discontinuity in the nature of the human/machine relationship in research.
‍

BCG and Define Ventures both document the rapid penetration of AI tools throughout the pharmaceutical R&D process, from target identification and molecule design through clinical trial optimization and safety signal detection. The productivity implications are significant: AI systems can explore molecular design spaces that would require decades of experimental iteration to map by conventional means and can integrate findings across the biomedical literature at a scale no human researcher could approach. Define Ventures’ survey of pharmaceutical executives finds broad anticipation that AI will compress drug development timelines substantially, though estimates of the magnitude vary considerably, reflecting genuine uncertainty about where the binding constraints on development speed actually lie.

What AI does well in the discovery context is, in essence, what disciplinary specialization has always done well: rapid and reliable pattern recognition within a well-defined domain of knowledge. What it does not yet do is what the best human scientists do uniquely: recognize when a pattern is surprising enough to be significant, ask questions that the existing data does not support, exercise the integrative judgment that connects empirical findings to theoretical implications, and maintain accountability to the communities whose problems the research is supposed to address. BCG notes that “curious generalists”—researchers capable of asking why a finding matters across multiple frames of reference—remain the limiting factor in the new AI-augmented research process, precisely because this integrative and evaluative function cannot yet be automated.

The governance challenge this creates is acute. In a piece entitled “Governing AI with Intelligence,” Urs Gasser argues that governing AI well requires precisely the capacities that AI is supposed to augment or replace: sound judgment, contextual understanding, and genuine value commitments. An AI system that identifies a promising drug candidate cannot evaluate whether that candidate should be pursued given the full social context of its development: the distribution of the disease burden, the likely affordability of the resulting therapy, the opportunity costs of the resources involved. Those judgments remain irreducibly human, and the governance question is how to ensure they are made deliberately and accountably rather than defaulting to the implicit values embedded in the AI system’s training data and architecture.

The IFTF Ethical Technology Governance Playbook approaches this challenge by arguing for the institutionalization of ethical deliberation within research organizations, rather than leaving it to external regulation. The playbook’s framework, which calls for structured processes of stakeholder engagement, values mapping, and anticipatory governance, is a useful starting point, though it leaves open the question of how to make such processes effective in organizations under strong competitive pressure to move quickly.

The argument so far has been largely diagnostic and, in tone, moderately optimistic: boundary dissolution, carefully governed, enables discovery that bounded disciplines and institutions cannot. But there are substantial risks that a celebratory account of boundary-crossing tends to understate.
‍

The first and most immediate is the integration problem. When multiple boundaries dissolve simultaneously, the cognitive and organizational task of synthesis  (making coherent knowledge out of contributions from multiple disciplines, institutions, expert types, and human-machine collaborations) becomes enormously more demanding. The Derderian cilia paper succeeded in this integration because the research team had genuine shared expertise and a single clear question to organize their collective efforts. At larger scales, integration is not guaranteed by proximity or collaborative intent; it requires deliberate investment in the people and processes that can hold complex, multi-provenance knowledge together. The Crick’s institutional design is an attempt to address this at the level of a single research institute, but it does not scale automatically to the distributed, multi-institutional collaborations that characterize much of the most ambitious contemporary science.

The second risk is equity. Boundary dissolution is not neutral in its distribution of benefits. The institutional innovations described in this Note are instruments available primarily to resource-rich actors in high-income countries. The persistent and, on current trajectories, widening gap between the production of scientific knowledge and its application to the most urgent health and development problems in the Global South are well documented. Global governance institutions are struggling to develop science-policy interfaces that give lower-income countries meaningful participation in the governance of knowledge whose benefits they need most. Boundary dissolution, unless explicitly designed with equity as a constraint, tends to accelerate advantage for those who are already advantaged, compounding existing inequities in the production and application of scientific knowledge.

The third risk is epistemological. If every boundary is dissolved—if the distinctions between disciplinary and non-disciplinary knowledge, between expert and amateur, between human and machine judgment, are collapsed without remainder—scientific knowledge loses the distinctive features that makes it valuable: reproducibility, accountability, falsifiability, and the capacity for cumulative self-correction. The productive dissolution is not the abolition of boundaries but their renegotiation on terms that preserve the epistemic values they were originally designed to protect. This is a more demanding requirement than it may appear: it requires the scientific community to articulate, with considerable explicitness, what the values at stake actually are—and to build governance frameworks capable of enforcing them in organizational environments that have not previously needed to be so explicit.