I wrote this essay in response to Seemay Chou's call for essays on systemic bottlenecks to science.
Structural change in scholarly communication practices is hamstrung by our recurring need to demonstrate our impact and research contributions through authorship position in narrative publications. Our obligation to our funders and the public demands that we share our research outputs widely. However, linear authorship position in a narrative is a lossy representation of an individual’s research contribution. This is especially true for the trainees doing the bulk of the actual labor of research, who often make core intellectual contributions. A structural change in research contribution and assessment is necessary to unshackle research communication from long-form narrative publication.
As a pre-tenure Assistant Professor in a biology department of an R1 research institution, I acutely feel this tension. I work in both cell biology and open science, and our lab’s research outputs take many forms: new schemas, individual results, hypotheses, models, and research tools. Too often, the old adage “wait til you’re safe with tenure!” gets waylaid by an attachment to the federal grant conveyor belt and fears that a trainee’s future success will depend on the same old metrics. This self-perpetuating loop back to the norm is hard or impossible to address within existing research assessment structures.
I decided that I couldn’t wait. COVID galvanized my conviction that now is the time for structural change in scientific research communication. As a postdoctoral cell biology researcher in 2020 at UC Berkeley, I learned that SARS-CoV-2 manipulates the very cellular processes we were studying in healthy cells. Curiosity and a desire to contribute to the global COVID knowledge base compelled us to organically form research collaborations investigating how SARS-CoV-2’s component proteins reshape our favorite subcellular compartments. This resulted in the voluntary contribution of 12 trainees from 4 major labs with expertise in microscopy, structural biology, and human cell biology. Everyone freely contributed on a given day according to their interest and capacity without obligation. By voluntarily choosing their work, each person maintained their own sense of agency and intrinsic motivation to contribute to the global COVID knowledge base. This generative stage was wildly successful: we collected 11 terabytes of cutting-edge, adaptive-optics-lattice-light-sheet-microscopy data of CRISPR-edited human stem cells expressing fluorescent SARS-CoV-2 proteins, using one of only 7 microscopes of its kind in the world while it was still being assembled. We saw some surprising morphological and dynamical changes to the cell’s mitochondria, nuclear membrane and endoplasmic reticulum induced by the viral proteins.
It was a dream project – a coronavirus utopia.
What happened? The project fell apart when it became time to narrativize and stake “first-author” credit on the grassroots collaboration. The rich grassroots network of researchers and contributions was pulled through the constrictive first-author pipeline, was pruned, withered, and died. The weight of responsibility shifted unsustainably: “lower-author” researchers became disincentivized to contribute, and I, as the putative first author, was crushed under the weight of 12 researchers’ branching data and findings. The data remain fallow on our hard drives. As millions of people died of COVID, it was our responsibility to freely share any useful information, as early as possible, whether or not the results conformed to a linear narrative. There must be some medium for scientific exchange larger than raw data and smaller than a narrative.
Around this time, I encountered the concepts of systemic and organizational design, wherein a different organizational structure fundamentally shifts the combinatorial, relational, and agentic outcomes of a system. I had a chance catalytic conversation with Kennan Salinero, who cofounded the nonprofit Reimagine Science. With Doug Kirkpatrick, author of “The No-Limits Enterprise,” Kennan described how more bottom-up organizational structures in companies and nonprofits transformed individuals’ sense of agency and motivation, enabling choice and organic collaboration over asymmetric responsibility, control, and oppression1.
Two texts fundamentally shifted my mental model from hierarchical control to grassroots self-organization and its capacity for systemic change. Dan Nicholson’s essay “Is the cell really a machine?” described how subcellular operation better fits the conceptual model of an organic, adaptable grassroots network than a stereotyped machine with interlocking parts. Peter Senge’s “The Fifth Discipline” taught me systems and design thinking. The structure of a game – its specified roles and rules for relationship between participants – determined the range of outcomes, miscommunications and failings more than any variation in who happened to inhabit a given role. Pulling the thread on a normative self-perpetuating loop is an elastic exercise: the harder you pull, the harder the normative pull pulls back. This is why a structural shift is necessary to escape the normative gravitational well into a new design space for scholarly communication.
I became captivated by the mission to apply these bottom-up “liberating structures” to my future interdisciplinary research lab. Bringing computer scientists, molecular biologists, physicists, and data scientists under one lab would require that they maintain context, iterative feedback, and a shared vision for our research projects. The natural medium for their shared understanding and collaboration surely was larger than lab notebook entries and smaller than a slideshow presentation.
The means to operationalize systems change in research collaboration
Through another chance encounter, I learned that the medium for scientific communication – new tools for research notetaking, sensemaking, and sharing – is a promising means to operationalize the systemic change that I had conceptualized. That’s when I met Human-Computer Interaction researcher Joel Chan on the Roam Research Slack workspace.
We sought a fundamental protocol for empirical research contribution. Like all protocols (HTTP, e-mail, shipping containers, Lego modules), a shared syntax allows for scalable interoperation of - in this case - knowledge assembly and synthesis, one flexible enough for expressivity while exhibiting the structure reflecting our design principles. Our design principles center empirical observations as the building blocks for evolving claims about the natural world. We hypothesize that a deliberately structured protocol for empirical research contribution can lead to earlier research sharing, deeper collaboration across farther disciplines, and more regular research contribution by less-established scientists.
Testing this hypothesis requires iterative cycles of designing, building, and testing both the protocol and associated user-facing applications with increasingly wider groups of collaborating researchers. In each iteration, we refine the protocol, improve the user experience of the applications, and change both according to researcher usage and needs.
Indeed, through 5.5 years of collaborative design-build-test cycles, Joel and I have experimented with these interoperable, grassroots protocols for collective knowledge assembly, coordination, and synthesis. We operationalize our designs through the iterative development and testing of open-source plugins in the graph-based (i.e. nonhierarchical, networked) notetaking apps Roam Research and Obsidian, with me, my lab, and ~10 pilot research labs throughout North America2.
We continue these explorations this June, at the inaugural workshop on Modular Interoperable Research Attribution in Dublin, Ireland. We created this workshop to bring together researchers, designers, publishers, and toolbuilders to prototype a means for researchers to share early-stage results and hypotheses with each other, via a protocol for attributable modular research3. It builds on recent Continuous Science Foundation workshops we attended on the future of science communication, standards for modular research reuse, and guidelines for incentivizing modular research reuse.
Defining and scaling modular research contribution
For such a system to scale, we must define: what is the fundamental unit of research contribution, and how is it assessed? At the Biohub 5th anniversary symposium, Ron Vale named the result – a single key research observation in an article – as a possible representation of a researcher’s contribution. I thought: that’s it! But for the system to work, it must be self-propagating, such that individuals following the lowest-friction path to make a strong research argument naturally cite others’ specific results, automatically generating the signal for attribution. This is the attributable protocol for modular research contribution.
While rates of research funding and job placement remain finite, a representation of our research contributions remains necessary to assess the reach and impact of our work4. With emerging technologies, this assessment can be richer, more accurate, and qualitative than previous attempts at distilling and representing research contributions5.
Benefits of attributable modular research contribution
Such a protocol also operates as a needed coordination mechanism for why and what we are sharing. What was your goal in sharing that research outcome? What question were you trying to address, and was it an open question? What is the new finding – what information are you introducing, and what’s the basis for your finding? What is your research argument, and what would make your argument stronger?
It also allows for different agencies – funding bodies, hiring committees, labs – to prioritize different selection criteria, and generate custom narratives or dashboards for their use cases. This accommodates diverse epistemic standards for acceptance, quality, and impact between research communities. A universal ranking is both unnecessary and harmful.
Outlook for modular research contribution
Why now? AI has exposed the gaps in scientific publishing as an effective collective sensemaking medium, increasing the urgency for us to experiment with forms of research communication and assessment. Indeed, AI needs such standards to effectively facilitate and make contributions to the social endeavor of science. Its coding and interpretive capabilities will lower the usability and technical barrier for using new scientific communication media.
This systems change is equal parts socio and technical. It’s discovering, through continual iterative trial and error, which parts feel, in practice, that they are necessary and immediately useful to the researcher, for which positive benefit to the community is an emergent outcome of the underlying structure as people operate within the structure.
Using this modular research protocol, one can retrofit existing and in-progress articles for dual use: the narrative and graph of contributions. Socially, modular attribution can operate as a bridge between current competitive societal norms (deciding who to hire or fund) and a liberating, open-source future for grassroots knowledge assembly.
Modular research attribution can guide us beyond capture of the artificially scarce, profit-driven Journal Fanciness Index. As is beginning to happen with “source code” in software, networks of modular research outputs can become the core specification for generating diverse narratives. These narrative stories – temporary, customized, and regularly updating – help individuals6 make sense of their field and new findings. The modular network, mutually interpretable by humans and AI, operates as a critical intermediary for collaborative knowledge assembly, no matter how powerful AIs become. Knowledge will be reassembled on demand as new evidence is introduced, revised, and strengthened.
In this network of modular research outputs, what is your research contribution? Where do your hypotheses and results sit in the “dependency graph” of knowledge undergirding future major discoveries (quantum gravity, a complex systems theory of life)? How strong is the claim of, say, the mechanism of CRISPR or the double helix model for DNA, absent the evidence from your research contributions?
Knowledge assembly is uniquely well suited to this type of liberating structure. Curiosity, ideation, and the desire to share discoveries are founded in the intrinsic motivation to make sense of the world and contribute to our shared understanding of it. Fair and accurate research attribution is our responsibility to less-established researchers so that they may represent their role in collective knowledge assembly and make the case to continue being supported in participating. We expect that the balance will tip in favor of these liberating structures once modular research attribution becomes a more useful and accurate signal for research contribution than journal article authorship position.
How does this all shake out for me and tenure? Well: I’m going up for tenure soon. I’m running the experiment – someone needs to. I’ll let you know how it turns out!