The lecture opens with a deceptively simple observation: an ordinary conversation between two people comes with a built-in way to confirm receipt — a listener can ask for something to be repeated, or check that they understood correctly. A society transmitting its collective "message" in the form of an election result has, historically, almost nothing resembling that mechanism. The lecture treats this gap as the organizing question for everything that follows: voting, examined as an information channel, complete with bandwidth, noise, and signal loss.

Using Claude Shannon's 1948 model of communication — source, transmitter, channel, receiver, destination — the lecture maps each stage directly onto the electoral process: citizens' actual preferences as the source, the ballot or interface as the transmitter that encodes preference into a discrete mark, the counting and transport infrastructure as the channel, the electoral commission as the receiver, and the legitimated decision as the destination. Concrete distortions are traced at each stage — a double-negative referendum question mis-encoding real preferences at the transmitter, a scanning or transcription error at the channel, and an aggregation method's own structural bias at the receiver, tying directly back to Arrow's theorem from the previous lecture.

The lecture then introduces Shannon's noisy-channel coding theorem: every channel has a calculable maximum capacity for reliable information transfer at a given noise level. Applied to voting, a binary yes/no ballot and a ranked-choice ballot carry structurally different amounts of information about the same underlying preferences — a difference in capacity built into the format itself, not a matter of execution quality. Choosing a voting format, the lecture argues, is therefore an engineering decision about acceptable bandwidth, not merely a matter of political tradition.

A second engineering principle follows: robustness against noise comes from structured redundancy rather than naive repetition — independent observers, parallel manual counts, multiple channels for filing complaints all serve the same function redundant encoding serves in any communication channel. The lecture draws a sharp, consequential line between noise — random, symmetric distortion that redundancy can catch — and bias — systematic, directional distortion that no amount of recounting can fix, and that instead requires intervention at the design stage of the channel itself.

The lecture's third block examines the Condorcet Jury Theorem, often cited as mathematical vindication for "the wisdom of crowds": under the right conditions, a large enough collective becomes more accurate than any individual within it. But the stronger, popular version of this claim — that sufficiently large collectives approach near-perfect accuracy simply by virtue of size — does not hold up. The theorem depends on three conditions holding together, and the crucial one is the independence of errors among participants.

The lecture locates the actual point of failure not in voter competence but in the architecture of the information environment: when a large share of a population draws on the same limited set of dominant media sources, errors caused by distorted information stop being independent and become correlated, and the collective starts erring in a synchronized rather than self-canceling way. This reframes disinformation as an engineering problem about the architecture of information channels, deliberately not a narrative about "uninformed voters."

Returning to the lecture's opening question, the final substantive block introduces end-to-end verifiability — an active, real class of engineering solutions under which any attempt to alter, discard, or miscount a vote must be either technically impossible or detectable with high probability, without requiring trust in any single administrator. Individual voters, under such systems, can obtain their own confirmation that their encrypted vote genuinely encodes their intended choice — their own version of "asking for clarification" — even while the technical mechanics of how this is achieved are deliberately deferred to a later lecture on the tension between transparency and secrecy.

The lecture closes by synthesizing four points — voting as a multi-stage communication channel, noise and capacity as unavoidable engineering facts, collective accuracy as conditional rather than automatic, and end-to-end verifiability as a real answer to "you can't ask for clarification" — and opens onto the next lecture's question: if noise in this channel can never be eliminated entirely, how can society reliably detect and measure it, and tell it apart from deliberate interference?