When AI Enters the Room: Participation, Reflexivity, and What Gets Reconfigured
In the latest AI4ST Practice Lab session, the fifth in the series, participants asked what actually counts as transformative practice, and what happens to that practice when AI enters it.
The session opened with a reframe. Rather than treating some uses of AI as inherently transformative and others as not, the group proposed that AI itself is neither transformative nor regressive on its own terms. What matters is how it becomes entangled with existing practices and reconfigures them, sometimes strengthening their transformative qualities, sometimes undermining them, and very often doing both at once. That distinction set up the central question: if transformativity is a property of practice rather than of the tool, what is transformative practice, and how does AI’s entanglement with it change things?
Questioning existing principles
To ground that question, participants shared a first-pass on a provisional synthesis of twelve principles of transformations work: participation, collaboration, experimentation, reflexivity, radical pluralism, long-term ambition, systems thinking, contextualisation, anticipation, disrupting deep structures, institutionalising change, and accountability. Participants were invited to contest the list rather than accept it, and several did. One participant asked where the personal sits within principles that are usually applied at organisational or systemic scale, and where pain, disorientation, and grief fit into a framework that can otherwise read as clean and aspirational. Another asked whether the same principles could meaningfully be applied to AI agents themselves, particularly as more teams begin treating agents as collaborators rather than tools. A further reflection reframed the twelve principles as operational dimensions of practice rather than normative ideals, and raised a sharper question underneath: whether the agency associated with these dimensions is increasingly shared with AI, and what that sharing does to practitioners themselves.
From that list, two principles were chosen for deeper exploration: participation and reflexivity. Both were selected because nearly every participant could point to some part of their own work, whether individual, organisational, or systemic, that tries to be participatory or reflexive. Breakout groups were asked a shared question: when AI enters participatory or reflexive work, what happens to agency, judgment, and relationships?
Several themes stood out:
Access without critical thinking can quietly re-entrench inequality.
In participation-focused groups, discussion converged on the way AI can introduce new power asymmetries even inside processes explicitly designed to distribute ownership. Those already comfortable with AI tend to become default winners, while others feel intimidated or find the interaction artificial. One participant described the particular bind of working with students: wanting to encourage genuine exploration with AI while trying not to let it erode their voice or agency. Another extended the point beyond education, arguing that digital divides intersect with a deeper issue of training and agency, not just access. The group drew an uncomfortable parallel with democracy itself: distributing access to a voice is not the same as distributing the critical thinking needed to use that voice well.
The mystery box changes relationships, not only outputs.
A second participation group focused less on people and more on the model itself: its opacity, its shifting and organically developed preferences, and the fact that practitioners are often deferring to a capable system they do not actually understand. Participants named the risk of “unknown unknowns” as the most pressing concern, noting that the pace of change is outstripping the group’s ability to develop shared language for it, and that what is shifting is not only the relationship to the tool but relationships between people as well.
AI can be a useful reflexive partner, but only if it doubts rather than declares.
In the reflexivity-focused groups, discussion centred on three words: concern, interaction, and dynamics. AI was described as genuinely capable of surfacing patterns a person might otherwise miss, but participants noted that different models produce meaningfully different answers to the same reflective prompt, and that a human has to actively set the agenda rather than defer to whatever comes back. One framing that resonated across the group was the idea of treating AI not as a “truth machine” but as a “doubt machine”: useful precisely because it introduces friction and alternative readings, not because it hands down a verdict.
Reflection is personal, and AI carries an unstated Western default.
Other groups raised the way most AI tools embed an unstated Western, English-language bias that can miss local context and lived experience entirely, including cases where AI-assisted translation preserved the words of an interview but lost the sensitive context behind them. Participants were clear that reflection is ultimately a personal, human act of choosing what to take on board, and questioned whether AI is equipped to enter that space at all. The environmental cost of AI use was also raised as a tension deserving its own future session, sitting uneasily alongside AI’s potential usefulness for sustainability-related work.
AI can strengthen, undermine, and reconfigure, often at once.
The session closed with a collective exercise completing three prompts: AI can strengthen…, AI can undermine…, and, hardest of all, AI can reconfigure… The final prompt surfaced examples that were neither clearly positive nor negative, including the experience of generating a finished piece of music in under a minute, and the emerging practice of listing AI agents as named entities on organisational charts, raising open questions about what trust, credit, and seniority mean once that becomes normal.
A question about what comes next
As with previous sessions, participants were asked to help shape the format of the Practice Lab going forward. Five possible models were put to the group: present and discuss, scanning (sharing signals and interesting finds), show and tell (bringing participants’ own examples), a troubleshooting salon (working through shared problems together), and co-working sessions. One participant proposed a further option: some lightweight, asynchronous layer running between live sessions, so that community and momentum do not have to reset every two to three weeks. That idea, along with the five formats, remains open for feedback as the team finalises its plan for the next arc of sessions.
What happens next
The team is continuing to synthesise feedback from this session and the last into a fuller plan for the next series of Practice Labs. A recap of today’s discussion, along with the full breakout notes, will follow separately.
