The cycle’s central challenge is to see large language models clearly—neither with the narcotic of hype nor the paralysis of fear. The Myth of the Given is the most important single obstacle to that clarity on the side of the builders. It generates the persistent belief that machine learning achieves objectivity because it learns from data, as though data were theory-neutral ground. Sellars’s demolition of the myth does not make machine learning useless; it makes it honest. A model trained on a dataset does not see the world as it is. It sees the world as the people who built the dataset saw it, compressed and averaged across millions of judgments. That is a real and often valuable thing. It is not the view from nowhere.
The Myth of the Given also underlies the popular notion that a model’s outputs are more objective than a human’s precisely because they are statistical—the machine has no axe to grind. But the statistical process is applied to data that embodies the axes of those who ground it. Averaging over bias does not remove bias; it distributes and conceals it. The cycle asks readers to hold this clearly not as a counsel of despair but as a condition of honest engagement: you can build fair, useful systems only by attending to the conceptual framing of the data, not by pretending the framing isn’t there.
The classical empiricist tradition from Locke and Hume through the sense-datum theorists of the early twentieth century held that knowledge has a foundation in episodes of pure sensory experience—the mind opened to the world and the world poured in, unmediated. This was a genuine achievement of intellectual discipline: it rejected innate ideas and metaphysical speculation, insisting that the court of appeal for any claim was experience. But Sellars saw that the discipline rested on a confusion. The empiricists had ruled out the Given as a theological or rationalist posit and then quietly reinstated it as the deliverances of the senses, without noticing that sense experience, to do the epistemic work required of it, had to be already conceptually structured.
The target of Sellars’s attack in “Empiricism and the Philosophy of Mind” was the then-dominant sense-datum theory, according to which perception delivers atomic mental items—red patches, hard surfaces, loud tones—that are simply present to the mind, self-authenticating, and available to justify beliefs without themselves requiring justification. Sellars’s reply: a red patch that plays a justificatory role must be reported as red, and reporting it as red requires already possessing the concept, and possessing the concept requires standing in a web of normative relations that makes the bare episode already a complex achievement. There is no foundation below the web. The web goes all the way down.
Conceptual saturation. Every act of observation or registration that can play a role in knowledge is already saturated with conceptual structure. This is not idealism—Sellars was a realist who believed the world exists independently of us. It is a claim about the structure of knowledge: there is no direct readout of uninterpreted reality, only a structured engagement with it that is possible only because the observer already possesses the relevant concepts.
Reliability without foundations. Rejecting the Given does not mean rejecting the possibility of reliable knowledge. Sellars held a coherentist-cum-reliabilist picture: knowledge is a self-supporting web in which observations play an evidential role, but that role is sustained by the rest of the web rather than by independent, uninterpreted access to bare fact. The web is reliable not because it is grounded in the Given but because its components mutually support each other and the whole system tracks the world reliably over time.
The AI objectivity illusion. A large language model trained on “raw data” inherits the conceptual framings of those who produced the data without making those framings visible. The Myth of the Given generates the illusion that this process is neutral; Sellars’s demolition of the myth converts the illusion into an empirical question: what framings are embedded in this data, and whose concepts are they? This is not a counsel against machine learning. It is the condition for honest machine learning.