Process Note: Manifesto

This note outlines the intellectual process that produced OSC’s manifesto. More than providing historical background, it models a central tenet of OSC: documenting one’s steps from the seed of an idea to a public product, including twists and turns along the way.

Version 0.1 (May 2026)

Long before coalescing into a text, the ideas in the manifesto emerged piecemeal over the course of years. As a professor at a selective liberal arts college, I have regular exposure to three phenomena that would come to feature prominently in the manifesto: open science, youth achievement cultures, and AI debates. 

The open science movement reshaped—and strengthened—my home discipline of political science, as it did many other fields. While it largely targeted professional researchers, I began seeing possibilities to extend open science to younger researchers when I attended a 2016 workshop on “Teaching Integrity in Empirical Research” (also known as Project TIER) at Haverford College.

I also regularly encountered students in my classroom who, from an early age, were steeped in competitive pressures that can warp inquiry into performance. Today’s social media and job market disruptions undeniably intensify those pressures, yet I can still recall my own days as an “over-achieving” high school student who spent too many hours padding my résumé instead of cultivating my curiosity. Only in college did I finally stumble into a happy obsession with empirical social science and become an “accidental scientist.” As a professor, I started craving ways to set students on an inquisitive path before they ever enroll in college, as I wish someone had done for me. 

Later, I joined many other teachers in asking what AI means for education at an institutional level (what assignments to assign, whether and how to police student AI use, etc.) and what it means for learning at a deeper philosophical level (how to define original thought in collaboration with AI, how to expand the limits of our understanding, etc.). I have an unusual dual perspective on these issues because just as AI was bursting onto the scene, I became a student again myself by auditing the statistics major at the college where I work, completing assignments and exams alongside some of the same undergrads I taught. Studying calculus, machine learning, data science, and other technical subjects after two decades with no formal math training made me lean heavily on AI for tutoring, refreshers, and a safe space to ask “dumb” questions. When I merely stood at the head of a classroom, my own students’ AI use seemed abstract and alien to me. But sitting at a desk as a student wrestling with problem sets, projects, and quizzes, I came to experience firsthand the power of AI to democratize learning and fuel intellectual play—not just to help students cheat like people often fear. At the same time, I battled the temptation to take cognitive shortcuts by having AI think for me, shattering any preconceptions I previously harbored that AI could ever fall neatly into either a “good” or “bad” category as a tool for thought.

These three threads—open science, the inquiry-performance nexus, and AI—wove together at an April 2026 meeting of the Berkeley Initiative for Transparency in the Social Sciences (BITSS) on the theme of “Frontiers on AI and Open Science.” The closing discussion turned toward cultures of open science: What can shift norms around journals publishing mainly flashy results? What can free investigators from the pressure to confirm their hypotheses? What might incentivize funders to underwrite projects that have high prospective benefits but also high risks of failure? I suddenly recalled a remarkable article that I had recently read: “Rethinking High-School Science Fairs” by Leah Libresco Sargeant, which diagnoses the problem of turning scientific inquiry into “glorified internships” and then imagines an alternative system that would celebrate students for projects that yield null results, pilot studies, replications, or retractions of prior findings rather than splashy breakthroughs. I proposed to the group that cultures of open science—including curiosity, comfort with uncertainty, and intellectual humility—plausibly take root (or not) long before people become professional researchers, journal editors, or grant officers. A high school student might get their first taste of inquiry at a science fair where they internalize pressures to deliver impressive results in a tidy package. My comment prompted another professor to lament that his first-year undergrads seem to arrive at college already eager to p-hack or do whatever it takes to confirm their hypotheses because the possibility of being wrong terrifies them. 

Undergraduate and graduate students already have excellent resources to navigate these anxieties and challenge intellectual norms, including some that I already mentioned such as Project TIER and BITSS. However, I was unaware of any similar opportunity for high school students. 

I returned from the BITSS meeting inspired to build a new community to address several needs that felt increasingly urgent in the context of scientific mistrust, emergent AI, and ongoing policy challenges ranging from public health to public transportation. I conceived of OSC as an alternative to mainstream student achievement culture, a space for reflecting on the interplay between artificial and human intelligence, and a pipeline for aspiring researchers who might go on to produce trustworthy science powering progress in many forms.

I realized that OSC needed a founding document to guide its evolution—something more comprehensive than a typical mission statement or “about” page on a website. I wrote version 0.1 of the manifesto in close collaboration with ChatGPT (GPT-5.5), first by sharing some of the same anecdotes and thoughts recounted above, along with numerous others that did not survive the pruning process. Through hundreds of exchanges over several weeks, disjointed musings slowly converged on core principles now reflected in the manifesto. Some of these exchanges involved writing or editing prose, but many involved debating ideas, testing arguments, resolving inconsistencies, and exploring alternative directions. As the manifesto itself acknowledges, this back-and-forth blurred the lines between my human contributions and the AI’s contributions. Even after writing a section on intellectual ownership, I remain unsure of how to characterize my ownership over the resulting document. If someone asked, “Did AI write the manifesto, or did you?” I would honestly answer, “Both.” What I can say with confidence is that the resulting product feels true to my intentions. I stand by everything in it, with the caveat that the manifesto—and OSC as a whole—are bound to change as this experiment unfolds in collaboration with AI and humans alike. This is only the first draft.

When I sat down to write this, I realized that the process note fills a different purpose from the manifesto. The manifesto plays the mechanical role of introducing OSC as an institution, so using AI liberally to organize ideas and clean prose made sense for clearly conveying what OSC is all about. In contrast, this process note plays a more philosophical role of upholding OSC’s deeper values including transparency, the embrace of messiness, and intellectual humility (especially in the face of AI). I therefore wrote it in my own unrefined voice with minimal input from AI (e.g., asking how long and how autobiographical to make it). This is not because writing with AI is unethical, but because different occasions call for different tools for thought.

Version 0.2 (June 2026)

I circulated version 0.1 of the manifesto to two trusted human readers whose feedback centered on two suggestions: 1) shorten it; and 2) write it in your own voice. #1 sounded straightforward; many documents can use tightening in their early stages.

#2 prompted me to revise some of my thinking on AI-assisted writing. In May 2026, when I first "wrote" (with heavy AI involvement) version 0.1, I was in the midst of deeply negotiating the role that I wanted LLMs to play in my writing in various contexts. Although public debates had already been swirling around AI and writing for a long time, I had yet to seriously confront those debates in my own writing until working on the manifesto. As I recounted above in the process note on v0.1, I initially classified the manifesto as a type of institutional document where content matters more than provenance, thus concluding that AI would communicate information more clearly than I could. I still think this perspective has some merit, but I started to suspect that I misclassified the manifesto, which might fall farther to the "provenance matters" end of the spectrum than I initially assumed.

Several pieces of information led me to update my views. First, one human reader who is very familiar with my authorial voice strongly chafed against the AI-generated prose in v0.1 and argued that my brain would produce a better document. I also recognized that a manifesto is not a technical report; it expresses philosophical commitments as much as basic information. In addition, sharp commentary from Scott Cunningham, Jasmine Sun, and others alerted me to a cognitive bias I had overlooked, namely that AI prose generated in my own workflow might appeal to me more than it appeals to others.

For all of these reasons, I shortened the manifesto and rewrote it in my own voice, resulting in v0.2. I retained many of the core ideas from v0.1 and continued to use ChatGPT but mainly as a patient editor as opposed to a prose generator. Out of curiosity, I compared Pangram scores for both versions. Pangram fairly characterized v0.1 as mostly AI-generated and v0.2 as mostly human-generated and slightly AI-assisted. No AI detector is perfect, and I do not regard such tools as definitive measures of authorship or ethical adjudicators.

In the end, I hope that v0.2 reads as more concise and more distinctly human than v0.1. This process note will continue to document any future revisions, especially as OSC receives more input from diverse sources.