A robust , well-timed census is vital to republic . Censuses chalk out the exchange face of our nation by chart both political and demographic shifts , include changes in riches and locality passage . Most crucially , they influence how resources and political might are doled out : Cities use nosecount information to arrange budgets and the Constitution mandates a internal census to apportion congressional posterior . But accept a nosecount is both expensive and slow — the annual American Community Survey ( ACS ) represent data compile over a five - year geological period , creating one hell of a lag . A squad of Stanford AI research worker conceive we can use computer visionto speed up the summons , a radical approach to a centuries - old practice session .
In a newfangled paper , the researchers described how they fast - tracked census data accumulation by using object identification on50 million Google Street View imagesfrom 200 cities . They discover that street fomite told a sort of neighborhood biography : Just by number cars they found in the look-alike ( 22 million in totality ) , they were able to extract data like income , airstream , education , and voting figure of a given area .
The researchers lead off by sorting automobile identified by inscrutable learning - based computer imagination into 2,657 categories base on make , model , physical structure eccentric and yr . From there , they created a grooming set for a computer model with census information from 35 of the 200 cities . By looking at nose count datum from the ACS , the US Census and the 2008 presidential election , the model surfaced associations between the railway car it spotted and key data points , include the race , political normal and income of neck of the woods . The associations were accurate enough that , when applied to the larger set of 165 metropolis , they mostly held up .
For example , Hondas and Toyotas were associated with predominantly Asiatic neighborhoods , Chryslers , Buicks , and Oldsmobiles with pitch-dark neighborhoods , and Aston Martins , Volkswagens and pickup truck with white neighborhoods . The good example most close affiliate Republican - vote precinct with extended - taxi pickup arm hand truck , while Democrat - voting precinct were mostly strongly tied to sedan . When matched against the actual nose count data , the AI loosely got it right and did it much quicker : the AI classify all 50 million image in 2 weeks . The ACS release conduct five years .
The researchers found the mannequin work best when associating cars with voting design , writing , “ We found that by driving through a city for 15 minutes while counting sedans and pick-me-up trucks , it is possible to dependably determine whether the city voted Democratic or Republican : if there are more sedans , it probably voted Democrat ( 88 % chance ) and if there are more pickup trucks , it plausibly vote Republican ( 82 % opportunity ) . ”
The truth is surprising , but certainly not astonishing : You ’d anticipate poorer neighborhoods to have older cars , just as you ’d expect those same neighborhoods to be less educated overall . Ultimately , the researchers say they ’d wish to coin a counterpoise between desegregate more advanced information collection techniques , like satellite imagination , into their employment and considering honourable concerns about public data and personal seclusion .
Real - time data collection could enormously meliorate the census process , but the world is increasingly being involve by various parties to forgo privacy in telephone exchange for some larger trade good , as is already the caseful withpolicingandonline surety .
[ Fast Company ]
AI / EthicsCarsCensusGoogle Street View
Daily Newsletter
Get the best tech , science , and polish news in your inbox day by day .
News from the future , deport to your present tense .
You May Also Like
