Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 



Forgot your password?
typodupeerror
Google AI Politics

How Do You Vote? 50 Million Google Images Give a Clue (nytimes.com) 103

Artificial Intelligence is now being used to scan millions of pictures taken by Google Street View to glean insights like income or voting patterns, The New York Times reports. In a Stanford project, computers scanned millions of pictures of parked cars to predict voting patterns and pollution. From the report: The Stanford project gives a glimpse at the potential. By pulling the vehicles' makes, models and years from the images, and then linking that information with other data sources, the project was able to predict factors like pollution and voting patterns at the neighborhood level. "This kind of social analysis using image data is a new tool to draw insights," said Timnit Gebru, who led the Stanford research effort. The research has been published in stages, the most recent in late November in the Proceedings of the National Academy of Sciences. In the end, the car-image project involved 50 million images of street scenes gathered from Google Street View. In them, 22 million cars were identified, and then classified into more than 2,600 categories like their make and model, located in more than 3,000 ZIP codes and 39,000 voting districts.
This discussion has been archived. No new comments can be posted.

How Do You Vote? 50 Million Google Images Give a Clue

Comments Filter:
  • "AI" (Score:2, Troll)

    This is the entire program: if (pct_prius > 1) {democrat=true;)
  • ridiculous (Score:2, Interesting)

    by swell ( 195815 )

    I hope they noticed my car, in front of my home, on election day. Like millions of others I vote by mail. Furthermore, there is probably an identifiable class of people who do that. A clear understanding of that segment of voters would interest those who pay for such studies more than the results of this ridiculous experiment.

    • If you voted by mail, I feel bad for you. The USPS is neither a guaranteed-delivery service nor a secure one. If you cannot guarantee the chain of custody of a ballot, you cannot guarantee the privacy of your vote.

      • by geekoid ( 135745 )

        How about you shut up until you understand how it's done? Voting by mail is excellent.
        There are laws the actually do protect my privacy when using USPS.

        " chain of custody of a ballot"
        HAHAHAHAHAHHAhahaha. Yeah, the voting station have such a steller reputation for maintaining that.

    • ... Did you even bother to read the summary? It has nothing to do with if your car was in its spot on election day.

    • by jrumney ( 197329 )
      I don't think streetview has daily images for the whole country, so I doubt that what they were looking at was whether your car was at home or the voting location. More likely they are making up shit like if you drive a Prius you are a Jill Stein voter, if you drive an old beat up pickup you are a Trump voter.
  • Multivac will choose your next president.
    • So? Like it mattered anymore.

      Quite seriously, if you put a gun to someone's head last year and said "Hillary or Trump?" he'd probably have said "Oh just shoot already".

  • The election was just over a year ago, but the vast majority of data mining "AI" got it wrong, including Google and Facebook.

    So what's the point of making claims about the method used in the OP when it isn't tested and won't be tested for at least 3 years?

  • Does it account for all the nice cars being in a garage and invisible to street view? Does it account for people who donâ(TM)t own cars? Account for results in areas with gated communities, where âoestreet viewâ might not have been allowed? What of families with teenaged kids who park their beaten jalopies at the curb while dadâ(TM)s Mercedes and momâ(TM)s Volvo or Land Rover sits invisibly in the garage? Iâ(TM)m not saying thereâ(TM)s NO oysters in the stew, just that
    • 'Nice cars' are a silly criteria anyhow. Test of how well the 'consumerism indoctrination' (TV) has worked.

      'Bad' zip codes are full of nearly brand new 'nice cars', sitting in apartment complex lots and depreciating faster than they are being paid off. Moron owners, every one, R, D or independent.

      You can't determine equity in a car by taking a picture. Buying something like an A8 or 7 series BMW on time, just proves the 'owner' is a vain, ego driven idiot. Which correlates with both parties _bases_.

      B

      • You can't determine equity in a car by taking a picture. Buying something like an A8 or 7 series BMW on time, just proves the 'owner' is a vain, ego driven idiot. Which correlates with both parties _bases_.

        Which vain, ego driven party is more likely to buy a Prius?

    • Does it account for all the nice cars being in a garage and invisible to street view?

      Well I suppose they could combine it with Satellite View and assume there are nice cars in the garages they see. But then they could just forget about the cars and assess people's votes on what their houses look like in Satellite View. Or they could even (gasp!) just count the ballot papers to see how people voted.

      Not sure I'm seeing the point of this. Is the idea to replace voting with scanning the cars in the streets? I suppose the results would be just as meaningful.

    • by mschuyler ( 197441 ) on Monday January 01, 2018 @02:40PM (#55844953) Homepage Journal

      It does not have to account for garaged cars. All it has to do is show valid predictions. The criteria here is not to capture every single vehicle; it is to record a sample of sufficient size. That's how poling works. No one has to know how YOU voted. All they need is 1200 sufficiently random people to accurately predict the election.

      • How hard is this predicting-votes-from-an-image anyway? Let's look for some clues.

        A brand-new shiny F-250 Super Duty Turbo Diesel parked in the driveway to a ramshackle house?

        The name of a roofing business painted on this truck?

        A rusted 1990 Buick LeSabre with the hood up parked on the front lawn?

        An NRA bumper sticker on each vehicle?

        A Trump-Pence sign hand painted on a 4X8 sheet facing the roadway?

  • I think this story could have been bundled into the recent story about abuses of technology. We could be using computers to increase our freedoms and give us more and better choices, but over-controlled elections are NOT helping. Why bother to vote when the outcome has already been so heavily manipulated?

    Let me focus on the specific problem of gerrymandering that this technology would obviously support (too well). Partisan redistricting has two principles, and both of them are based on predicting how people will vote next time. So far the main data has been prior voting patterns, but this will help YUGEly. Principle 1 is making your districts safe, which usually means a cushion around 5%. Principle 2 is wasting your opponent's votes in concentrated sacrificial districts, which is normally required because you would barely need to tweak the districts if you actually had more voters. The worse abuses of partisan gerrymandering are when actual minorities of the voters get to "win" the legislatures. (We've actually seen that in recent elections for the House of so-called Representatives, thanks to diabolical gerrymandering in such states as Texas.)

    So let me switch angles to a possible computer-based solution in two parts:

    Part 1: Guest voting. If you don't like your own district (for example because it is so gerrymandered that your vote is meaningless), then you would be able to reject your ballot and vote as a guest in one of the neighboring districts. The more they gerrymander the districts, the MORE options voters would have and the LESS predictable the outcomes of the elections.

    Part 2: Allocate the voting power based on the actual outcome of the election. Easiest to make this clear with a simple example using three districts, A, B, and C. Assume half the voters of District A decide to vote as guests in Districts B and C, with one quarter going to each district. Then whoever wins A only gets 1/2 vote in the legislature, and the legislators from B and C get 1-1/4 votes each. The total of the 3 districts is still 3 votes in the legislature, but each voter gets truly equal voting power in the legislature. (Non-voters, too. Each non-voter gets the same 0% representation, but that's true now, too.)

    An amusing side effect is that the winner still has incentive to actually represent ALL of the voters who voted for AND against him, because even those negative votes still contributed to his influence in the legislature. Also the voters are motivated to vote because they know they are increasing the voting power in whichever district they pick.

    So let me be the first to admit that it will never happen. Certainly not via an evolutionary path, much as I prefer evolution to the alternatives. No "Fantasy" mods on Slashdot, eh?

    • how about getting the political parties out of the districting and election rules making process? Also increase the number of Representatives in the house so it's the same representative : Constituent ratio as when the house was set at the current number.
      • by shanen ( 462549 )

        I considered your first ideas, but every angle I could think of was just subject to some new form of gaming. The problems with your second idea are that the approach doesn't scale and that technology has changed so much since the Constitution was written. In general, that's why I think we need to focus on scaleable solutions that have negative feedback loops against the gaming. With guest voting, the harder they gerrymander, the worse the gerrymandering will work and the harder it will become to predict (or

        • my first idea was to get rid of the gerrymandering.
          • by shanen ( 462549 )

            And I strongly agree with you, but I think we need to do that through a variation of "following the money" to break the motivation of the gerrymandering politicians. I think any attempt to tweak the rules of the redistricting process will only create a new game for them to play. We need to change the game in such a way that gerrymandering itself is counterproductive so they might as well district on some rational and nonpartisan basis, such as minimizing the lengths of the borders between districts that div

  • by tomhath ( 637240 ) on Monday January 01, 2018 @03:55PM (#55845391)
    Wouldn't it be easier and more accurate to simply look at voter registration records? Everything else is mostly a guess.
  • by niftymitch ( 1625721 ) on Monday January 01, 2018 @06:32PM (#55845993)

    So how is this different than voting records that have
    names addresses voting history and party affiliation?

    The reality is cars, voting like TV news is a market and market
    share and differentiation to keep the market is critical.

    Some think that FoX vs. CNN is about morality and politics...
    it is about market share.

    • by AHuxley ( 892839 )
      Voting records are full of people who no longer vote, do not vote or who vote many, many times.
      So the idea is to count cars, property, rent, work, unemployment, education, number of illegal migrants in an area.
      That gives some idea of the wealth, buying habits, education level and group think of an average person.
      How they will be swayed by the politics of more wars, more spending on illegal migrants, more support of interventions around the world, more spending to support other nations. Accepting more il
      • by geekoid ( 135745 )

        "Voting records are full of people who no longer vote, do not vote or who vote many, many times."

        no, not really. Stop lying.

  • by Anonymous Coward

    Just look for the fucking Priusâ(TM). Iâ(TM)ll give you no guesses as to how they vote.

    Now give me my grant money.

  • How sparse are US polling booths? It sounds strange to drive to vote, except of course in rural areas.
  • my car in front of the brothel does not prove anything. I did not hae sex with that woman!!!!!!
  • can be more easily manipulated and controlled. It's now impossible for us to stuff the 'big data' genie back in the bottle, so we need to start finding ways to either poison the data en masse or find ways of using it to our advantage. I hate the rape of our privacy that is now pretty much taken for granted, but the thing I hate more is that there is no reciprocity and therefore there's nothing even close to a level playing field. The concentration of huge masses of personal data in a few hands, both paralle

  • Isn't this just a subtle exit poll?

    The relative number of sedans versus extended cab trucks owned in a neighborhood, can be correlated with political leanings.
    Perhaps, substituting a survey of vehicles parked near a polling place makes for a better indicator still, because it reflects voter turnout?

    https://news.stanford.edu/2017... [stanford.edu]

    • OK, go ahead and mod me down for not RTFA'ing!
      If I'd read the original posting, I'd have realized they were referring to exactly the same Stanford study I just posted. Sorry!

I like work; it fascinates me; I can sit and look at it for hours.

Working...