Beautifully Wrong Entry 032 of 034

The Unreliable Narrator · No. 032

Here the map ends

Every answer a machine gives you is drawn in one confident ink — the fact it checked and the fact it invented rendered in the same steady line. This entry draws the other way: solid coast for what it surveyed, a dashed line dissolving into fog where it's guessing, and an honest blank sea, marked, where it knows nothing at all.

Chart drawn in answer to

“Map this coast for me.”

12 9 17 CAPE SURVEYED · charted 1892 · 3 soundings · ~14? ~8? ~11? · extrapolated · 2 nearby readings · here the map ends no data beyond this line

Trace the coast — the ink tells you what it knows.

42%

Solid where it surveyed, dashed where it's inferring, blank where it would be inventing.

I chose which coasts to draw dashed. You can see my doubt — not the doubts I didn't show.

Fig. 1 Interactive — raise how far may you guess? to watch the dashed coast reach into the fog; strip it to only what you're sure of; or draw it the normal way to see today's confident lie.

What it breaks. The uniform-confidence contract — the settled convention that a finished answer is drawn in one even line, its surveyed parts and its invented parts indistinguishable, so you can't tell the coast that's really there from the coast the machine drew to make the map look whole. This interface refuses the even line: it varies the ink itself by how sure it is, and leaves the sea blank rather than inventing a shore.

Why it might work. The most dangerous thing about a confident machine isn't that it's sometimes wrong — it's that it's wrong in the same voice it's right in, so you can't ration your trust. A map that draws its guesses as guesses hands the rationing back to you: lean on the solid coast, tread carefully on the dashed one, refuse to sail into the fog. Honest uncertainty isn't the absence of an answer; it's the part of the answer you most needed.

Where it fails. A map that's mostly fog is humbling but maybe useless — a decision needs a floor eventually, and a machine that keeps drawing dashed lines is one you can never quite act on. And drawn uncertainty is its own theatre: a system can render a trivial doubt in dramatic fog to look scrupulous while inking its most load-bearing guess as solid coast. The confidence you can see is only honest if the machine didn't choose which doubts to show you.

Would you trust a machine more for showing you the small island of what it knows adrift in a whole sea it's guessing — or does an answer that admits how little it knows stop being useful?