detrans.ai: A Counter Narrative | Peter James Steven
26 February 2026
With Peter James Steven
detransition
When artificial intelligence is trained on mainstream sources, it tends to reproduce mainstream assumptions — including the uncritical affirmation model that dominates gender medicine. Peter James Steven set out to build something different. His chatbot, detrans.ai, draws on a curated database of detransitioner experiences to offer what most AI systems do not: a grounded, evidence-informed counter-narrative. This episode examines what that project reveals about the information environment around gender transition, how technology encodes ideology, and why the stories detransitioners tell matter for any honest reckoning with gender medicine.
Artificial intelligence does not exist in a vacuum. The models that now answer questions about gender identity, medical transition, and detransition were trained on data that skews heavily toward affirmation — because the institutions, platforms, and publications that produce most of the text on the internet have, for the better part of a decade, skewed that way too. Peter James Steven recognised this and decided to do something about it. He built detrans.ai, a chatbot specifically designed to draw on the recorded experiences of detransitioners, the group whose stories most consistently fall outside the frame of dominant gender ideology. The episode opens a genuinely important window onto the relationship between technology and ideology. When someone asks a mainstream AI system about gender dysphoria or the risks of medical transition, the answer they receive reflects not neutral information but a particular ideological settlement — one reached by consensus among institutions that have largely declined to weight detransitioner experiences as evidence. Steven's project is a direct challenge to that settlement: an attempt to use the same technology to encode a different body of knowledge, one built from the ground up on the testimony of people who transitioned and later came to question or regret that path. The conversation covers the technical architecture of the chatbot — how it was built, which model underlies it, and how the database of detransitioner accounts was assembled and structured. The statistics gathered from this database form a substantial part of the discussion. Patterns in who transitions, who detransitions, and for what reasons begin to emerge when enough accounts are collected and analysed. These patterns do not simply confirm what advocates on either side might prefer; they reflect the complexity of real human experience in ways that ideologically shaped guidance rarely does. Steven also speaks to his own motivations — the personal connections that drew him to this work and the particular concern he feels about the trajectory of gender medicine. The conversation extends into questions about pronouns, identity, and how language itself shapes the terrain of this debate. The hosts draw him out on the social dynamics at play: how affirmation culture rewards certain narratives and makes others difficult to voice, and what it takes, personally and practically, to sustain a counter-narrative project in that environment. What the episode ultimately illuminates is a structural problem in the information ecosystem around gender. Detransitioners represent a living form of evidence — evidence about outcomes, about the adequacy of psychological assessment before medical intervention, about what happens when affirmation is treated as both the beginning and the end of clinical care. That evidence has been systematically underweighted by the institutions that should be integrating it: clinics, professional bodies, and now, by extension, the AI systems those institutions have indirectly shaped. The existence of detrans.ai is a symptom of that failure, and a small but meaningful attempt to correct it.
