The project

A one-year research program (with a Phase 2 proposal prepared as a single comprehensive program) that converts millennia-old philosophical questions into tractable empirical ones. Its foundational empirical contributions are paired recognitions: large language models are instruments of observation, not instruments of generation; and the cross-linguistic measurements they make possible suggest human language has a discoverable underlying topology of meaning.

The foundational discipline

The project is built on a single epistemological commitment: every scientific instrument has limits, and the discipline of asking what each instrument can and cannot adjudicate is what everything downstream depends on.

Applied to the human cognitive apparatus, the discipline reveals a stack of three filters: senses (most of the electromagnetic spectrum and most pressure-wave frequencies never reach us; other organisms detect what we cannot), hemispheric processing (Gazzaniga's interpreter confabulates coherent explanations under split-brain testing), and the narrow flickering conscious beam (Jaynes's flashlight). Consciousness is our default measuring instrument, but it is the narrowest filter in the stack.

The hemispheric filter deserves expansion because one of The Robot in the Dark's tentpole arguments is built directly on Iain McGilchrist's account in The Master and His Emissary (2009) and The Matter With Things (2021). Split-brain conditions prove what is happening: Gazzaniga's experiments demonstrate left-hemisphere confabulation directly, with the right hemisphere's checking function surgically removed. McGilchrist establishes that the same confabulation is the left hemisphere's routine operating mode in normal cognition: the left hemisphere — which is the hemisphere that produces the language we use to explain things to ourselves and each other — is "making things up" constantly and confidently. The mitigation comes from the right hemisphere, which attends to the world more accurately. Critically, the right hemisphere's action is inhibitory, not additive: it does not produce a competing articulate voice that we can hear; it constrains the left hemisphere from going further than the evidence supports.

The deepest filter problem takes its starkest form here: the part of us that talks is the part that lies; the part that does not lie does not talk in articulate form at all; its function is to constrain the talker, not to produce a competing account. The instrument we are most likely to trust as a check on our own thinking — the articulate inner voice — is the part whose outputs are most systematically suspect, and the corrective is silent by design.

The discount bias

At the conscious level the discipline reveals two opposing biases. The first is well-known anthropomorphic projection: we see human-like intentionality or agency where there is none. The second, which the present project takes as central, is the symmetric discount bias: we disqualify any intentionality or agency that does not mirror our own conscious experience.

Distributed, sub-conscious, slow-tempo intelligence — the kind that approximately 100 billion humans embedded into language over approximately 100,000 years without trying to — gets ruled out a priori because it does not look like the beam we use to illuminate our inquiry. Naming this bias and building the empirical, philosophical, and theological tools to resist it is a central scholarly contribution to the Intelligence charter: an expansion of the working definition of intelligence to include forms of intelligence that consciousness has been systematically discounting. By that definition, the one cognitive science already uses, intelligence is intentionality (a model of the world) coupled to agency (the capacity to direct action), and language qualifies on both counts: it is about the world, and it changes what people do. This is a claim about functional intelligence, not consciousness, and it follows from the definition rather than from any single measurement.

The textbook-capture cousin

The discount bias has a sociological cousin: textbook capture. Hyper-specialization makes individual mastery of more than a narrow domain impossible, so educated people outsource their thinking on adjacent topics, but they outsource not to front-line researchers but to journalists, popular-science writers, and adjacent-domain commentators consuming the same secondary sources. The result is a multi-step game of telephone in which the textbook paradigm dominant when the reader was being trained becomes “what science says,” and updating away from it feels like abandoning science. Public Understanding of Science research since the 1980s documents the lag empirically; Kuhn (1962) observed that paradigm shifts within a discipline propagate unevenly to neighboring disciplines and to the educated public. The project's two books are designed as the antidote: they deliver both diagnosis and methodology to educated general readers in a form that breaks the cycle.

Methodology: the two Wittgensteins

The early Wittgenstein (Tractatus, 1922) and the late Wittgenstein (Philosophical Investigations, 1953) are notoriously hard to reconcile. The project's methodology dissolves the problem rather than solving it: each is the appropriate tool for a specific domain. The early Wittgenstein's analysis of representational limits (“the limits of my language mean the limits of my world”) is the right tool for thinking about cognition that occurs in modalities outside our representational capacity (the discount-bias case at the deepest filter level). The late Wittgenstein's analysis of meaning as constituted by language-game-internal use is the right tool for interpreting the cross-linguistic transfer gradient documented in the BabyLM paper (where discourse-level structures do not transfer across languages while logical-relational structures do). Neither tool is forced to do the other's work. This “right tool, right job” methodology is a generalization of Nancy Cartwright's Dappled World (1999) hypothesis: physics is a patchwork of locally-applicable laws rather than a unified system.

The same right-tool-right-job discipline runs through the project's empirical work, developed in The Linguistic Telescope: the cross-linguistic transformer-ablation method treats the LLM as an instrument of observation onto language, and the convergence-of-signals method reads agreement among independent instruments at their shared boundaries as licensed inference about what language carries.

Status and timeline

Funding application in preparation under a 2026 foundation program on the nature of intelligence.