In the latest AI4ST Practice Lab session, the fourth in the series, and the one that wraps up its first full arc, the participants moved from mapping possibilities to weighing whether those possibilities are worth pursuing at all.
The session turned to exploring the operational and cultural conditions that determine whether a promising AI use case actually survives contact with the real world. Building on the workflows, capabilities, and risks mapped in earlier sessions, participants worked in pairs through two linked prompts first identifying the conditions that would inhibit a plausible AI application in their own context, then, in a second round, the conditions that would enable it.
Several themes stood out:
Trust is not one thing it is many things stacked on top of each other. Discussions repeatedly pulled apart what “trust” in AI actually means: trust that the tool won’t hallucinate or produce biased output, trust that a knowledge system or worldview is genuinely represented rather than flattened, and trust that decisions made with AI’s help won’t quietly harm the people they’re meant to serve. Participants noted that AI sits inside existing systems of power, capital, and epistemology, and that a lack of trust often reflects legitimate, layered concerns rather than simple resistance to new tools.
Uneven literacy creates its own inhibiting conditions. Several groups converged on the idea that differences in AI literacy across a team or community can quietly undermine an otherwise good use case, producing shallow or inconsistent results that then get blamed on the technology itself rather than on the design of the process. The proposed fix was deliberately designed, participatory processes that bring people to a shared enough level of understanding and a shared enough set of values to decide together what AI should and shouldn’t be used for.
Poor first attempts can poison the well. When AI-assisted synthesis of community knowledge produces underwhelming, generic insights, it can convince an organisation that further investment isn’t worth it, even when the failure was really one of design rather than principle. Human judgement at every step of the process, several participants argued, is what prevents this collapse into blandness.
Agency and sense-making must stay human. A recurring worry was that AI could be used to “sense” across a large, dispersed community surveying, aggregating,summarising while the harder, more meaningful work of collective sense-making, interpretation, and disagreement gets displaced. Participants were clear that field-sensing and meaning-making are not the same activity, and that losing the second to automation would be a real loss, not a convenience.
Infrastructure and workflow clarity are enabling conditions, not afterthoughts. On the enabling side, groups pointed to concrete needs: good knowledge management and data architecture, tools simplified enough to lower the learning curve, and perhaps most importantly clear agreement within a team or organisation about which parts of a workflow are reserved for AI and which are reserved for humans. Without that clarity, participants suggested, people reasonably fear redundancy rather than augmentation.
A provocation on wisdom
The group turned briefly to a set of provocations on what it might mean to use AI wisely, distinct from using it responsibly, safely, or ethically. Drawing on a review of spiritual, decolonial, and indigenous knowledge traditions, three problematic tropes were offered as mirrors: AI as slave (a being beneath us to be extracted from without reciprocity), AI as vending machine (knowledge as transactional and universal rather than relational and situated), and AI as macho man (AI as an oracle of cold fact, sidelining more relational, caring, or embodied ways of knowing).
The group landed somewhere close to agreement: wisdom, unlike information, is situated rather than universal, and felt rather than checked off a list which means it can only ever be something we bring to our use of AI, not something the tool supplies on its own.
A question we did not fully answer
As with the previous session, time ran out before the group could fully close the loop on its own framework. Having named the inhibiting and enabling conditions, the harder next step deciding, case by case, whether a particular augmentation is worth pursuing at all once those conditions are accounted for remains open. So too does the wisdom question raised at the end: not just whether an AI use case is safe or responsible, but whether it is wise.
What happens next
This session closed the first series of three practice labs. The team is now turning survey feedback, breakout discussion, and chat contributions into a proposal for the next arc of sessions, likely continuing the cadence of roughly every three to four weeks, with a stronger throughline of continuity and shared output based on what participants asked for.
The conversation continues on Tuesday, 11 August 2026, with the next AI4ST Practice Lab:
Documenting and Sharing Emerging AI Practices
16:00–17:30 CEST | Online via Zoom
In this session, we will begin an exciting new series of Practice Labs, shaped by the rich participant feedback gathered throughout our initial round of experimentation.
If you have a workflow tension, an inhibiting or enabling condition, or a question about wisdom and AI that didn’t make it into today’s discussion, bring it with you this conversation is far from finished.
