Slashdot Mirror
Google Turns To Users To Improve Its AI Chops Outside the US (wired.com)
Google is betting that algorithms that understand images and text will draw business to its cloud services, make augmented reality popular, and prompt us to search using our smartphone cameras. From a report: The search company's machine learning systems work best on material from a few rich parts of the world, like the US. They stumble more frequently on data from less affluent countries -- particularly emerging economies like India that Google is counting on to maintain its growth. "We have a very sparse training data set from parts of the world that are not the United States and Western Europe," says Anurag Batra, a researcher at Google.
When Batra travels to his native Delhi, he says Google's AI systems become less smart. Now, he leads a project trying to change that. "We can understand pasta very well, but if you ask about pesarattu dosa, or anything from Korea or Vietnam, we're not very good," Batra says. To fix the problem, Batra is tapping the brains and phones of some of Google's billions of users. His team built an app called Crowdsource that asks people to perform quick tasks like checking the accuracy of Google's image-recognition and translation algorithms. Starting this week, the Crowdsource app also asks users to take and upload photos of nearby objects.
When Batra travels to his native Delhi, he says Google's AI systems become less smart. Now, he leads a project trying to change that. "We can understand pasta very well, but if you ask about pesarattu dosa, or anything from Korea or Vietnam, we're not very good," Batra says. To fix the problem, Batra is tapping the brains and phones of some of Google's billions of users. His team built an app called Crowdsource that asks people to perform quick tasks like checking the accuracy of Google's image-recognition and translation algorithms. Starting this week, the Crowdsource app also asks users to take and upload photos of nearby objects.
Yes, AI gets pretty dumb when it sees stuff that's even slightly different than what it's used to. It adds sheep to pictures if it sees a pasture: https://www.theguardian.com/te...
Or, any yellow and black image must be a school bus: https://www.wired.com/2015/01/...
http://www.geoffreylandis.com
Batra is tapping the brains and phones of some of Google's billions of users. His team built an app called Crowdsource that asks people to perform quick tasks like checking the accuracy of Google's image-recognition and translation algorithms. Starting this week, the Crowdsource app also asks users to take and upload photos of nearby objects. ...
Google isnâ(TM)t offering to share that potential bounty with people contributing data to Crowdsource. The app rewards contributors with a system of points, badges, and certificates. Collect enough and youâ(TM)ll be invited to join online chats with other top contributors via Googleâ(TM)s Hangouts service.
Hey. Look at that. American ingenuity at it's finest. Get the people to work for free or as close to it as you can. I can see the pitch now!
"The wisdom of the crowds at your fingertips, free to us but let me talk to you about the down payment you need to engage the greatest engineering feet designed by man. Hopefully you've not already a second mortgage on that house, my friend, your small business is going to need what we can do for you. Without it, you'll never see the top of the search heap. Nosiree. You'll. Never. See. The. Top."
I'm curious as to how Google plans to prevent trolling on the part of people like 4chan that just love to abuse AI systems. I can easily see someone uploading photos of dicks in response to every photo request or adding racist translations among other mischief.