Workers can spend anywhere from 5 minutes to 40 hours a week going through simple tasks on a smartphone, making AI all over the world more intelligent and efficient. By building platforms and crowd sourcing a large supply of workers, the companies that train AI were smart to realize the data needed could be processed by anyone with steady access to the internet.
Read more: Will AI change the future of music?
From this developed companies like Clickwork, Neurala, and Alegion Inc., which utilize "human taggers" to push AI education. With over a million "clickworkers" signed up, there's an estimated 100,000 people logged on at this very moment, sifting through droves of information and lending a human touch to robot training.
"Staring at security cameras or an airport scanner, remote-controlling a robot, driving trucks up and down a mine did not exist as jobs until recently," says Massimiliano Versace, CEO of Neurala, "and will probably not exist as jobs occupying a full-time human as they are today. Humans were created to do much more complicated and elevated tasks than watching videos 24/7."
Teach a machine to see a sign
Although the engineers have the hardest tasks — the actual creation and building of AI — the load of information needed to teach a robot to work properly is overwhelmingly large and tedious to sort. For instance, to train a car to drive itself safely it must learn to recognize pedestrians, stop signs, and other traffic signals. To teach it how to recognize these objects, hundreds of thousands of photos are individually sifted through and labeled to teach the machine what is and what is not a stop sign.
Companies like Alegion and Clickwork send out thousands or even millions of photos to their custom-built platforms, and workers on the other end go through and process each photo, creating the bite-sized pieces of data that will sculpt the AI's perception of the world, eventually enabling it to drive.
"To recognize these events, you have to harvest hundreds of thousands of images," says Vito Vishnepolsky, the head of North American business development for Clickworker, "then feed them to the systems through algorithms so they can recognize these patterns. You need to train those systems."
The categories that need the most clicking include image definition for autonomous driving, language and speech transcription, and categorizing media. For some tasks the compensation may be less than a dollar, and for more complicated tasks like identifying x-rays the compensation is sometimes $1,000 a photo.
The money continues to flow into AI
A recent report by consultancy PricewaterhouseCoopers estimates AI may contribute as much as $15.7 trillion (13.3 trillion euros) to the world economy by 2030. An estimated $6 trillion is predicted to come from increased production and job automation, while another $9 trillion will come from "consumption side-effects" like changes in shopping habits as a result of improvements in technology and manufacturing.
Venture capitalists have been throwing money at AI-based startups for years, and the list of major clients continues to grow with companies like Facebook, Amazon.com, Apple, Microsoft, and Google searching for ways to streamline business with new technology. The major contributing factor Clickwork and others bring are the quality-control methods and the software interfaces used to catalogue the data.
Is clickworking the new video surveillance?
While human taggers may decrease in demand, there will unlikely come a time when humans are not at least working in tandem with the AI we have created, as well as maintaining it.
There is still disagreement within the industry as to how robust the jobs economy supporting AI will end up being. Versace thinks crowd-sourcing will diminish, "The need for humans to teach AI will reduce greatly over time, as companies figure out ways to reduce the workload on human taggers to get data ingested by AI."
But Vishnepolsky of Clickworker feels otherwise, saying AI is taking jobs from security companies and creating jobs in high technology, artificial intelligence software houses.
"This world is constantly evolving and you will always need engineers to keep the systems up to date and you will always need crowd source workers to create the data sets to train the systems," he says.