The Eye Test by Chris Jones

The Eye Test by Chris Jones

Author:Chris Jones
Language: eng
Format: epub
Publisher: Grand Central Publishing
Published: 2022-01-11T00:00:00+00:00


During 2020’s apocalyptic fire season along the American Pacific coast, smartphone cameras weren’t capturing how red the sky looked to the people standing under it.19 Some users cried conspiracy—tech giants didn’t want the world to know how fearsome the situation had become—but in truth the problem was an algorithm that was struggling to make sense of things it wasn’t designed to understand. Sony builds nearly half of all smartphone cameras and trains them to see by feeding them hundreds of millions of images; the machine then uses that knowledge to reproduce accurate images of similar things. But the algorithm wasn’t fed pictures of fire skies, leaving it confused by the amount of red it was registering in otherwise ordinary-seeming scenes of houses, cars, and trees. These weren’t volcanoes, after all. Some people’s smartphones corrected what they decided must have been an error, making light conditions seem more typical than they were and washing out a horribly vibrant reality—because that reality was so extreme, their cameras doubted their own lenses. They didn’t believe they were seeing what they were seeing.

After decades of dominating our political conversations, polls, too, appear increasingly unable to capture the truth about us. In 2020, when President Joe Biden beat Donald Trump, national polls missed the mark by even larger margins than they had in 2016, when they had overwhelmingly predicted that Hillary Clinton would win. Certain state polls were also catastrophically wrong. The day before the 2020 presidential election, Biden was expected to win Wisconsin by more than eight points; he won by a whisper, 0.7 percent. Ohio, Florida, and even Texas were thought to be statistical dead heats and maybe even in play for the Democrats; Trump won all three by comfortable margins.

The misfires extended beyond Trump’s weirdly inelastic popularity, influencing down-ticket races: Polls overestimated the potential of Democratic candidates by an average of four points—a margin of error that made those polls effectively useless in close races. Republican senator Susan Collins of Maine didn’t lead in a single pre-election poll over Sara Gideon, her Democratic opponent. Collins won the actual race by more than eight points. Democrat Jaime Harrison seemed in the running in South Carolina, where he faced off against Senator Lindsey Graham, the Republican stalwart. Outside money started pouring into the apparently competitive state, and Harrison raised more cash in the final quarter than any Senate candidate in U.S. political history. He lost by more than ten points.

Fairly or not, Nate Silver’s vaunted FiveThirtyEight.com took much of the post-election heat. Silver’s models, which weigh and aggregate a number of outside polls, had become famous in 2012, when he had correctly forecast every state’s preference for either President Barack Obama or Mitt Romney, including nine notoriously difficult-to-predict swing states. Analytics had once again proved unbeatable. In 2016, however, Silver had fared less well. He was less wrong than most; three days before the election, the Princeton Election Consortium gave Clinton a 99 percent chance of victory, while Silver gave her a 71.4 percent chance.



Download



Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.