Representationalism

In the AI Story

Representationalism has roots deep in Western philosophy — in Plato's theory of forms, in Descartes' mind-body dualism, in the empiricist assumption that sense-data represent an external world. Its modern form, shaped by scientific practice in the seventeenth and eighteenth centuries, assumes a clean separation between the knowing subject and the known object, between language and the world it describes, between the instrument and the phenomenon it measures. Knowledge is accurate to the extent that the representation corresponds to the represented.

Barad's critique does not deny that representations can be more or less accurate — her framework is not relativism. The critique is that representationalism misdescribes the relationship between knowing and being. It treats the world as pre-given and the knower as neutral observer, concealing the constitutive role of the apparatus through which knowing occurs. The instrument, the cognitive framework, the institutional context, the theoretical assumptions — all participate in producing the phenomena they engage, not merely in representing them.

Applied to AI, representationalism structures most public discourse. Debates about whether AI will replace human workers assume that human worker is a pre-given category with determinate capabilities that can be compared with the capabilities of a pre-given category called AI system. The debate produces answers (yes, no, partially) that leave both categories intact. Barad's performative alternative reveals the categories as enacted through specific material-discursive practices — and shows that AI is not replacing a stable entity but participating in the reconstitution of what human worker will come to mean.

The critique has direct implications for how AI governance proceeds. If representationalism is correct, then AI policy is a matter of managing a pre-given technology's impact on pre-given populations — a technical problem admitting technical solutions. If Barad's performative alternative is correct, then AI policy is simultaneously ontological (determining what kinds of entities come into being), epistemological (shaping what can be known), and ethical (setting the stakes for what matters). The governance challenge is not managing a tool but configuring an apparatus, and the configuration carries consequences representationalism obscures.

Origin

The term representationalism has multiple philosophical uses — in philosophy of mind, philosophy of language, and political theory. Barad's specifically ontological use derives from Ian Hacking's distinction between representing and intervening, and from Barad's own development across her essays from the 1990s onward. Her most systematic treatment appears in Meeting the Universe Halfway (2007), especially in the chapters on the ontological presuppositions of scientific practice.

Key Ideas

Representationalism assumes the pre-given. Entities exist with determinate properties before any knowing occurs.

Knowing is mirroring. The task of knowledge is to produce accurate representations of a pre-existing world.

The knower is separable from the known. Observation can be neutral; instruments can be transparent.

Performativity inverts the picture. Entities are produced through practice; knowing is participation; the apparatus is constitutive.

The shift matters. Different assumptions produce different policies, different ethics, different forms of life.