Computer-human hybrids could be best at scanning for danger

Humans are great at picking up on important information in images – but can be swamped in data. Time to team up with computers

Humans are good at spotting anomalies (Image: Dado Ruvic / Reuters)

IN A world of algorithms, there are still a few places where humans reign supreme.

At a US government lab in Albuquerque, New Mexico, researchers aren’t interested in replacing our brains with fancy neural networks or machine learning software. Instead, they are using eye-tracking and brain analysis to create a system that lets our natural intelligence shine.

“It’s a human and machine data system that collectively makes everything better,” says Laura McNamara, an organisational anthropologist at Sandia National Laboratories. “Human beings are supremely good at pattern recognition, but what overwhelms that is having way too much data.”

What we’re so good at is finding the signal in the noise. Intelligence analysts, for example, can comb through troves of satellite imagery of enemy land to spot a tank or a weapons cache.

Their expertise is tricky for a computer to mimic, in part because what they are looking for can be a little different every time. Even the experts have trouble explaining how they do it. In a series of experiments, the Sandia team studied how specialists and novices evaluate images – when and where they zoom, pan or toggle buttons on the computer, or how their eyes move as they scan the screen.

They found that each expert had a distinct style, often eschewing the techniques they had been trained in.