A truism of all work: If you have lots of people providing tiny pieces of information, you need a human manager who sees the big picture to assemble it into a coherent project.
Funneling everything through a human supervisor has been one of the few things to slow down the rise of crowdwork, a digital assembly line of sorts in which online workers perform so-called microtasks — such as verifying links work, assigning labels to photos, gathering small bits of information to answer a question.
But researchers at Carnegie Mellon University have shattered that truism in a report released Tuesday.
In a collaboration with technology giant Robert Bosch LLC, CMU researchers created a prototype computer system that, aided by machine learning, can assemble mammoth amounts of human-gathered information into a final product.
The system has wide implications, said Niki Kittur, associate professor in CMU’s Human-Computer Interaction Institute and one of the report’s co-authors.
A common example of crowdwork is Wikipedia, the online encyclopedia in which a group of editors stitch together articles from millions of contributors.
“In many cases, it’s too much to expect any one person to maintain the big picture in his head,” Mr. Kittur said.
Researchers used their system, called the Knowledge Accelerator, to solicit workers to create articles that answer 11 open-ended questions.
These included, “How do I unclog my bathroom drain?” and “What are the best attractions in LA if I have two little kids?” and “How do I deal with the arthritis in my knee as a 28 year old?”
As Mr. Kittur described it, the computer program allowed human workers to continue to perform the deep-thinking research, which includes editing and finding multimedia.
Though the articles were written by non-experts hired through Amazon Mechanical Turk, one of the most common platforms for crowdwork, researchers found they were rated significantly higher by paid reviewers than many expert Web pages on the topics.
Mr. Kittur has an altruistic vision for improving how information is assembled online.
Whether it’s a patient trying to learn about symptoms, a consumer looking for a new car, or a voter trying to decide which candidate aligns with their beliefs, the best answer can be difficult to track down.
What if Knowledge Accelerator could collect thousands of points of view from Internet forums, blend it into neutral information, and speed up that process?
“There are a lot of places online where there’s a lot of messy data, but that data has interesting value in it,” Mr. Kittur said. “We don’t know what to do with it because there are thousands of points of view.
“What we really want to do with this is be able to take all of that work people are doing in their heads and capture it and make it useful for others.”
The National Science Foundation and Google also supported the research, a CMU press release said.
Daniel Moore: dmoore@post-gazette.com, 412-263-2743 and Twitter @PGdanielmoore.
First Published: May 11, 2016, 4:00 a.m.