The Fishbowl of the Powerful

In the AI Story

The concept builds on Segal's fishbowl metaphor from The Orange Pill while sharpening it into a specifically political claim. The fishbowl is the set of assumptions so familiar that the person inside it has stopped noticing them — the water the fish breathes, the glass that shapes what the fish can see. Everyone inhabits a fishbowl. The claim that deserves scrutiny is not the existence of the fishbowl but the specific ways in which the builder's fishbowl conceals the consequences of building from the person who builds. Segal describes the exhilaration of the Trivandrum training — twenty engineers, each operating with the leverage of a full team, a twenty-fold productivity multiplier achieved in days. Visible from inside the fishbowl: expanded capability, compressed timelines, the acceleration of the path from imagination to artifact. Not visible, or at least not foregrounded: the distributional consequences of a twenty-fold multiplier and the asymmetry of institutional protection between the builder and those whose capability he now directs.

Shklar's analysis of power and vulnerability sharpens the concept's political edge. The relationship between the builder and the engineers is not symmetrical. The builder possesses information they do not — about the company's financial position, about board expectations, about the competitive dynamics that shape strategic decisions. The builder possesses decision-making authority they do not — the authority to keep the team or reduce it, to invest productivity gains in expansion or in margin. And the builder possesses the epistemic privilege of the fishbowl — the set of assumptions about innovation, about building, about the inherent goodness of expanded capability — that makes the twenty-fold multiplier feel like a gift to the engineers rather than a threat. From inside, the multiplier is empowerment. From outside, it is a redistribution of power whose continued benevolence depends on the judgment of a single person.

Hannes Bajohr, simultaneously one of the foremost contemporary Shklar scholars and a leading researcher on AI and language, has extended the concept in a direction Shklar herself did not reach. The fishbowl of the powerful in the AI age is not merely a matter of what the powerful can see but of what can be articulated within the systems the powerful control. Large language models are vocabulary-imposing systems — they determine what can be said fluently and what cannot, what arguments are readily available and what arguments require effort to construct, what thoughts flow easily through the system and what thoughts encounter friction. The concentration of capacity to shape perception in the hands of system builders, combined with the distribution of consequences across entire populations, creates a structural asymmetry more durable than any based on information or resources alone.

The demand Shklar's framework imposes is not for better builders — though better builders would help — but for better institutions. Institutions that force the consequences of power into the field of vision of the people who exercise it. Transparency requirements that make distributional effects visible to workers affected by them. Accountability structures that create costs for harms the market would otherwise externalize. Mechanisms that give the vulnerable a voice in decisions that shape their lives — not as a gesture of corporate benevolence but as a structural feature of the political order. The glass of the fishbowl does not break by itself. It must be broken by institutions, deliberately, repeatedly, and against the persistent resistance of those who find the water inside it comfortable.

Origin

The concept synthesizes Shklar's analysis of power asymmetries with Segal's fishbowl metaphor from The Orange Pill and with Hannes Bajohr's 2023 and 2024 essays on AI, language, and liberal democracy. Bajohr's dual expertise — as a Shklar scholar and as a theorist of large language models — makes his extension of the concept particularly authoritative.

Key Ideas

Blindness is structural, not moral. The fishbowl is produced by the institutional environment of power-holders, not by the character of the individuals who hold power.

Information asymmetry grounds domination. The powerful possess information the affected do not, and this asymmetry is itself a form of power that standard accountability structures fail to address.

The vocabulary itself is controlled. AI systems constitute a new layer of fishbowl in which the articulable is shaped by commercial interests the user cannot inspect — deepening the asymmetry Shklar identified.

Individual effort is insufficient. The honest builder who attempts to see outside the fishbowl still operates within structural forces stronger than individual will; institutional remedies are required.

Demands for transparency are not optional. The liberalism of fear generates specific institutional requirements — transparency, accountability, voice for the affected — that are not courtesies but conditions of legitimate rule.