Cranes, Not Skyhooks

In the AI Story

The distinction has been Dennett's sharpest rhetorical instrument for four decades. He used it to dispatch vitalism, Cartesian dualism, Searle's Chinese Room, Penrose's quantum consciousness, and every other proposal he considered an attempt to smuggle unearned magic into the explanation of mind. The move was always the same: identify the skyhook, propose a plausible crane, and show that the skyhook added nothing the crane did not already deliver.

The AI debate has revived every ancient skyhook. Consciousness is said to be required for 'real' understanding. Intentionality is said to be impossible in silicon. Creativity is said to require something beyond pattern completion. Dennett's response, extended from his earlier work, is that these claims name capacities without explaining them, and that the correct methodological move is to ask what cranes could build the capacity — and then notice that gradient descent, at sufficient scale and with sufficient data, is a serious candidate crane for a large class of cognitive functions that had previously seemed to require skyhooks.

This does not settle the question of whether current AI systems have the relevant capacities. It reframes the question. Instead of 'does the machine really understand,' the question becomes 'what would a plausible crane for understanding look like, and how much of it has the system been built?' The first question invites endless metaphysical escalation. The second invites empirical work.

The framework connects directly to The Orange Pill's river of intelligence. Segal's image of intelligence as a 13.8-billion-year process that flows through progressively more sophisticated channels is, in Dennett's vocabulary, the crane-built history of mind. AI is a new channel in the river. It was built by cranes — statistical learning, gradient descent, enormous training corpora — all the way down.

Origin

The crane/skyhook distinction was introduced in Darwin's Dangerous Idea (1995) and remained Dennett's signature rhetorical move through the rest of his career. Its application to AI became explicit in the 2010s as deep learning began to exhibit capacities that skyhook-seekers had declared impossible.

Dennett's final interventions — in interviews, essays, and his 2023 memoir I've Been Thinking — insisted that the AI debate was being miscast by people who kept demanding skyhook authentication of machine capacities that had obviously been built by cranes.

Key Ideas

No unearned capacities. Every genuine capacity in nature was built by some process from simpler components; the explanation is the process, not an authenticating miracle.

Selection as universal crane. Darwinian selection, suitably generalized to cultural and computational domains, is the crane of cranes — a mechanism that can build arbitrary complexity given enough time and variation.

The skyhook move is diagnostic. When a critic insists that some capacity requires X, and X has no causal story, the critic is calling in a skyhook — and the correct response is to ask what crane they have in mind.

Cranes for cognition are under construction. Large neural networks are not the final crane for mind, but they are serious cranes for significant subsets of cognitive function, and that is enough to shift the burden of proof.