Could Cruise Be the Theranos of AI?

Newsletter post, November 4, 2023. By Gary Marcus.

Key Takeaways

• The core allegation: Cruise presented vehicles as autonomous while depending heavily on remote intervention to mask what the AI could not do.
• The numbers, via New York Times reporting: remote operator intervention every few miles, and roughly 1.5 workers per vehicle.
• The generalization is the sharper claim: “remote operators could well be the dark secret of ALL self-driving” — with minimal disclosure anywhere in the sector.
• The Theranos parallel is specific, not rhetorical: a goal that may be theoretically achievable, oversold as presently achieved, with no realistic delivery timeline.
• Marcus calls for California regulators to compel disclosure of remote-operator usage, on the grounds that lives are at stake.

Connections

• The pessimistic pole of The Driverless Dream. Against How a Driverless Car Sees the Road, it forms the draft’s central tension: the 2015 talk made the machine’s competence legible; this piece argues the legibility was partly staged.
• The transferable idea beyond AVs: a demo that hides its human scaffolding is a claim about cost structure, not just capability — which connects to Bubble Architecture and the subsidized-unit-economics argument.