Computers and Humans ‘See’ Differently. Does It Matter? | Abstractions blog - Quanta Magazine

In some ways, machine vision is superior to human vision. In other ways, it may never catch up by Kevin Hartnett, senior writer at Quanta Magazine covering mathematics and computer science. 

Photo: Abstractions blog

When engineers first endeavored to teach computers to see, they took it for granted that computers would see like humans. The first proposals for computer vision in the 1960s were “clearly motivated by characteristics of human vision,” said John Tsotsos, a computer scientist at York University.

Things have changed a lot since then.

Computer vision has grown from a pie-in-the-sky idea into a sprawling field. Computers can now outperform human beings in some vision tasks, like classifying pictures — dog or wolf? — and detecting anomalies in medical images. And the way artificial “neural networks” process visual data looks increasingly dissimilar from the way humans do.
Computers are beating us at our own game by playing by different rules...

There is a lot we don’t know about human vision, but we know it doesn’t work like that. In our recent story, “A Mathematical Model Unlocks the Secrets of Vision,” Quanta described a new mathematical model that tries to explain the central mystery of human vision: how the visual cortex in the brain creates vivid, accurate representations of the world based on the scant information it receives from the retina...

(In an interview with Quanta Magazine last year, the artificial intelligence pioneer Judea Pearl made this point more generally when he argued that correlation training won’t get AI systems very far in the long run.)
Read more

Source: Quanta Magazine