January 6, 2016

Melding Human, Machine Computing to Solve Big Problems

George Leopold
Screen Shot 2016-01-06 at 2.21.50 PM

Machine learning and intelligence attract a lot of attention in the big data community. Now researchers are looking for ways to combine the emerging technology with human computation skills in areas like “creative abstraction” to tackle some of the planet’s toughest problems.

Researchers at the Human Computation Institute and Cornell University stressed in a recent article published in the journal Science that they seek to combine the raw horsepower of high-performance computing with the subtle cognitive skills of humans. The combination of machine intelligence and human computation—defined as the “science of crowd-powered systems”—would allow humans and machines to ” accomplish tasks that neither can do alone,” the researchers insist.

Examples of human computation range from Wikipedia to the web widget used to transcribe distorted text to prove a user is human. The researchers said they want to build on those early examples of human-machine interaction to tackle intractable, or “wicked,” global problems like climate change and pandemics that have so far defied traditional problem-solving methods.

The research institute seeks to forge partnerships among human computation specialists, domain experts, on-the-ground relief agencies like the Red Cross and potential funding sources like government agencies. That approach is combined with crowd-sourcing techniques, citizen scientists and “distributed knowledge collection” that can make more precise observations than machines alone and do so at massive scale.

Once data is gathered, machines with learning abilities can chew on the data to help come up with possible solutions to “wicked” problems.

Along with projects focused on environmental and geopolitical concerns, human computation researchers are applying the approach to treating diseases like Alzheimer’s. The institute’s collaborators at Cornell University have used new imaging techniques to study a symptom association with Alzheimer’s Disease and other forms of dementia: reduced blood flow to the brain. Understanding the mechanisms behind this reduced blood flow has pointed to a potential treatment that could reduce cognitive symptoms and slow disease progression, researchers said.

The catch is that additional research is extremely labor intensive, requiring several weeks of data recording by laboratory personnel to collect a hour’s worth of relevant data. “Indeed, the curation aspect of the analysis is so time consuming that to complete the studies necessary for identifying a drug target could take decades,” researchers warned.

Enter human computation, which proponents assert could be used to handle “perceptual tasks” that are beyond the capability of machines but relatively easy for humans. “We aim to address the analytic bottleneck via crowdsourcing using a divide-and-conquer strategy,” researchers said.

Along with Cornell, the effort to crowd source a treatment for Alzheimer’s Disease includes researchers from Princeton University, University of California at Berkeley and the citizen science project coordinators SciStarter and WiredDifferently. (The latter group helped map the neurons on the human retina to help researchers better understand how vision works.)

Those interested in participating in the Alzheimer’s research project can register here.

Recent items:

Startup to Offer Neural Network Training in the Cloud

Machine Learning Tool Seeks to Automate Data Science

Share This