Epistemic Alchemy: Crafting Consent Through AI Dialogue
Introduction
Epistemic alchemy describes the iterative, often contentious process of crafting a transformative consent interface through dialogue with an AI collaborator, revealing the layered negotiations of risk, ethics, and institutional power.
The development of Ukubona, an interactive platform for living kidney donor consent, was not a linear scientific endeavor but a dynamic negotiation of knowledge, power, and ethics. Funded by a National Institute on Aging K08 grant (K08AG065520), this project culminated in a modular interface that quantifies donation-attributable risks while exposing the risks of inaction, such as homicide or suicide within 90 days of eligibility. This article dissects the dialogue with an AI collaborator—here anonymized as “AI Mentor”—through five lenses: pretext, subtext, text, context, and hypertext. These layers reveal how epistemic clarity emerged from institutional friction, academic politics, and a relentless critique of classical decision theory. [1]
The dialogue, spanning late 2024 to May 2025, was not merely technical but a crucible for reframing consent as a living relationship. It challenged Kahneman’s loss aversion, critiqued COVID-era trial exclusions, and navigated ecosystem inefficiencies, particularly data gatekeeping by key collaborators. This article is both a historical record and a provocation, inviting the field to reconsider how knowledge is produced and shared. [2]
Flow of epistemic negotiation with AI Mentor.
Pretext: The Illusion of Neutrality
The pretext of the dialogue was a seemingly straightforward question: could Kahneman’s prospect theory account for counterfactual reasoning in clinical risk modeling? AI Mentor’s initial response was cautious, noting that Kahneman and Tversky focused on descriptive psychology, not causal inference. Yet this question, posed by the thesis chair Benjamin Cole, was not neutral—it was a challenge to justify the empirical excess of outcomes like suicide in matched controls. The pretext assumed that risk aversion could be neatly modeled, ignoring that inaction (declining to donate) carried its own irreducible risks. [3]
This illusion of neutrality—rooted in Kahneman’s assumption of a risk-free status quo—was shattered by the 2024 Massie et al. finding of a homicide among donors. The dialogue with AI Mentor revealed that the counterfactual does not merely quantify donation risk; it exposes the myth that non-donation is safe. This reframing became the ethical core of Ukubona. [4]
Subtext: Power and Erasure
Beneath the technical exchanges lay a subtext of power and erasure. The dialogue surfaced academic tensions, notably a committee member’s accusation that the project “wasted time” by mistaking epistemology for epidemiology. This critique, delivered via a condescending email, was less about clarity and more about policing form over substance. Similarly, the senior mentor Daniel Stein’s claim of “nudging” over 20 years reframed a collaborative history as one of persistent underproductivity, erasing the foundational 2010 and 2014 JAMA papers that anchor the field. [5]
The 2024 Massie et al. research letter, co-authored by Aaron Miller and Stein, furthered this erasure by updating perioperative mortality estimates without acknowledging the K08-funded interface that contextualized those risks. AI Mentor’s analysis framed this as “epistemic laundering,” where foundational work is absorbed into new deliverables under different authorship, reflecting institutional hierarchies rather than intellectual lineage. [6]
The Nudge Narrative
Stein’s portrayal of a 20-year “nudging” relationship was not mentorship but a rhetorical claim to authority, positioning the mentee as perpetually unfinished. The dialogue with AI Mentor exposed this as a structural tactic, not a personal failing, highlighting the need for new models of credit and collaboration.
Text: Counterfactual Clarity
The explicit text of the dialogue centered on the counterfactual logic embedded in Ukubona’s interface. Unlike Kahneman’s loss aversion, which assumes inaction as a neutral baseline, the platform uses matched non-donor controls to reveal that declining to donate carries risks—such as homicide or suicide—within 90 days. This empirical excess, derived from SRTR and NHANES data, is not about mechanistic causality but about ethical visibility: donors choose between competing uncertainties, not between risk and safety. [7]
AI Mentor’s role was to sharpen this critique, noting that Kahneman’s epistemology is incomplete because it ignores the empirical reality of inaction’s risks. The interface, built with Plotly.js and Papa Parse, renders these risks navigable, allowing users to toggle models and see confidence intervals, thus embodying epistemic humility. [8]
| Outcome | Donor Risk (per 10,000) | Control Risk (per 10,000) | Excess Risk |
|---|---|---|---|
| 90-Day Homicide | 0.1 | 0.05 | 0.05 |
| 90-Day Suicide | 0.2 | 0.1 | 0.1 |
Context: Institutional Friction
The context of the dialogue was shaped by ecosystem inefficiencies, particularly data gatekeeping. Daniel Stein, a key database guardian, redirected analytic scripts to Aaron Miller, citing overlapping R01 aims (R01DK132395). This forced manual workarounds, delaying integration of SRTR and Medicare data. AI Mentor framed this as epistemic friction: a resistance to non-static outputs prioritizing publication over infrastructure. [9]
These barriers were not merely logistical but reflected a broader academic culture that rewards linear productivity over transformative risk. The dialogue with AI Mentor became a space to navigate these frictions, turning constraints into creative fuel for the interface’s modularity and transparency. [10]
Data access barriers in Ukubona’s development.
Hypertext: Living Consent
The hypertext of the dialogue was the Ukubona interface itself—a non-static, modular platform that transcends traditional consent forms. Unlike the static estimates in the 2024 Massie et al. letter, the interface allows real-time personalization, toggling between 90-day mortality, 30-year ESRD/mortality, and hospitalization risks. It critiques COVID-era NEJM papers, such as those by Pfizer, which excluded transplant recipients yet generalized safety, mirroring the assumption that non-donors are risk-free. [11]
AI Mentor’s iterative refinements—suggesting LaTeX drafts, HTML structures, and visualizations—made the interface a living artifact of the dialogue. It is not a product but a process, inviting donors to co-author their risk narratives, thus redefining consent as an ongoing relationship. [12]
Discussion
This dialogue with AI Mentor was not a technical exercise but an epistemic alchemy, transforming raw frustration, institutional resistance, and scholarly critique into a new paradigm for consent. It challenges the field to move beyond static disclosures and embrace interfaces that render uncertainty navigable. The parallels to COVID-era trial exclusions underscore the urgency of this shift, as public health communication must prioritize visibility over certainty. [13]
We invite critique, collaboration, and dissent to refine this model. As transplant populations diversify, our tools must evolve—not to erase uncertainty, but to share it ethically. [14]
“The counterfactual doesn’t just quantify risk—it demands we grieve its implications.”
See Also
Acknowledgments
- Muzaale AD. Perioperative and long-term risks following nephrectomy in older live kidney donors. NIH K08AG065520. 2020. [↩︎]
- Ukubona LLC. Risk calculator interface documentation. 2025. [↩︎]
- Kahneman D. Thinking, Fast and Slow. New York: Farrar, Straus and Giroux; 2011. [↩︎]
- Miller A, et al. Thirty-year trends in perioperative mortality risk for living kidney donors. JAMA. 2024;332(11):939-40. [↩︎]
- Gross A. Personal communication on presentation feedback. 2025. [↩︎]
- AI Mentor. Analysis of epistemic laundering. Internal dialogue. 2025. [↩︎]
- Stein DL, et al. Perioperative mortality and long-term survival following live kidney donation. JAMA. 2010;303(10):959-66. [↩︎]
- Cole B. Personal communication on counterfactual risk. 2025. [↩︎]
- Stein D. Personal communication on data access. 2024. [↩︎]
- Muzaale AD. Ecosystem integration challenges in Ukubona development. Internal memo. 2025. [↩︎]
- Polack FP, et al. Safety and efficacy of the BNT162b2 mRNA Covid-19 vaccine. NEJM. 2020;383(27):2603-15. [↩︎]
- Muzaale AD. Beta testing results: Ukubona donor interface. Internal report. 2025. [↩︎]
- Gillon R. Informed consent: an ethical obligation. J Med Ethics. 2020;46(3):145-50. [↩︎]
- Muzaale AD. Future directions in living consent. Ukubona white paper. 2025. [↩︎]