The Practice of Worthy Amplification

In the AI Story

The first practice — self-interrogation — is the regular application of Glover's moral-identity question to one's own recent choices. Not the grandiose am I a good person? — unanswerable, useless — but the specific what kind of person am I becoming through the choices I've made this week, this month, in this project? The practice creates the external vantage point that the incremental slide cannot provide from inside. It asks whether the builder examined or reflexively approved, whether she considered affected persons or let the timeline eliminate the space for consideration, whether the self emerging from the week's work is a self she recognizes.

The second practice — constructed proximity — deliberately creates the encounters with affected persons that AI-mediated work eliminates by default. More than user research, which typically converts persons into insights for the production process. Genuine proximity: hearing the voice, seeing the face, registering the gap between intended and experienced effect, allowing the gap to produce the discomfort that is the human response in action. The encounter cannot be optimized or scaled. It requires the specific, irreducible friction of being in the presence of a person affected by what you have built.

The third practice — preserved friction — builds into workflow the structures that resist smoothness. Mandatory pauses between conception and deployment. Required reviews asking not does it work? but should it exist? Protected time for unstructured reflection. These structures will be experienced as inefficiencies by an efficiency-optimizing culture. They are not inefficiencies. They are the institutional equivalent of moral muscle — the resistance that prevents smoothness from eliminating the discomfort on which moral selfhood depends.

The fourth practice — expression-conviction alignment — addresses the specific erosion that Segal encountered at the coffee shop. Before approving any AI-generated output attributed to you, ask: do I believe this? Not is it plausible? — the tool's outputs are always plausible. Is it mine? Does it reflect what I actually think, or has the tool's fluency substituted for my own uncertain, effortful conviction? The discipline is expensive, slow, requires rejection of polished output in favor of rough belief. It is the discipline through which moral selfhood maintains its integrity in an environment that makes the substitution of plausibility for truth effortless.

The fifth practice — accepted obligation — addresses the diffusion of responsibility at the individual level. Understanding confers responsibility. The engineer who understands engagement mechanics bears responsibility for their deployment. The leader who understands what AI-assisted work does to her team's cognitive environment bears responsibility for the structures she builds or fails to build. The parent who understands what AI-saturated environments do to moral development bears responsibility for the conditions she creates at home. The understanding does not create obligation from nothing. It reveals the obligation that distance, diffusion, and the architecture of carelessness had rendered invisible.

Origin

The five-practice framework is On AI's synthesis, drawing on Glover's prescriptive insistence that moral formation is performed rather than held, on Vallor's virtue-theoretic treatment of technomoral practice, and on specific empirical findings from the Berkeley study, from Diane Vaughan's work on institutional discipline, and from the growing literature on AI-mediated cognitive effects. The practices are offered not as sufficient conditions for moral selfhood but as the minimum architecture without which the amplified self tends toward carelessness.

The structure mirrors Glover's own approach to moral prescription: specific, practicable, structural. He was skeptical of abstract ethical exhortation and consistently grounded his recommendations in observable practices that could be performed by ordinary agents in ordinary institutional conditions. The five practices preserve this skepticism and this groundedness while translating the framework into the specific conditions of the AI-assisted workplace.

Key Ideas

Performance, not principle. Moral identity is built through what one does, not what one believes. The practices are things to do, not things to think.

Specific, not grand. Each practice is a bounded activity that can be performed on an ordinary Tuesday. None requires heroism.

Expensive, not easy. Each practice costs time, attention, or comfort. The cost is the point — the resistance is what exercises the capacity.

Structural, not virtuous. The practices cannot depend on individual virtue alone. Institutional structures that require them — or at least do not punish them — are necessary for their sustainability.

Minimum, not sufficient. The practices do not guarantee moral formation. They constitute the minimum architecture without which formation is improbable in the AI-amplified environment.