An AI with vision superior to that of the human eye

Researchers at the University of Central Florida have unveiled a new Artificial Intelligence capable of seeing, recognizing shapes and identifying objects. This technology could be used in robotics or to improve autonomous car systems. Would sight then be a new sense soon to be unlocked for Artificial Intelligences?

It is in any case the project of its researchers from the University of Central Florida. Thanks to a device capable of reproducing the retina of one eye human, their creation could lead to a new, more efficient AI with new capabilities. Concretely, technology could allow this artificial intelligence to instantly understand what she really has in front of her. The most obvious use case seems to be the autonomous car or robotics.

Another benefit is the technology, described in a recent study published in review ACS Nanoalso performs better than the eye regarding the range of wavelengthsthat is, it can perceive as well theultraviolet that the light normal. For self-driving vehicles, the device’s versatility should provide safer driving in a variety of conditions, as explained by Molla Manjurul Islam, the study’s lead author.

A versatile device

” If you are in your autonomous vehicle the night and the car’s imaging system only works on a particular wavelength, say the visible wavelength, it won’t see what’s in front of itdescribes Molla Islam in the study. But in our case, with our device, he can actually see in totality “. According to the researchers, there is currently no device of this type capable of operating equally well in the ultraviolet range, that of visible waves and even infrared. These skills give this device a unique character.

The intelligent imaging technologies currently in place operate in several stages, from the detection of information to its processing, including storage. ” We had devices, which behaved like the synapses from brain human, but we didn’t directly provide them with the imageexplains Tania Roy, professor at the University of Central Florida. Now, adding the image sensing capability to them, we have synapse-like devices that act like “smart pixels” in a camera by simultaneously detecting, processing and recognizing images “.

In addition, the device does not need to be multiplied, and the chip, very compact, is the same size as a thumb. At the moment the accuracy rate is around 70-80% for each wavelength and it should continue to improve before it can actually be used. The researchers estimate that the technology could be ready within the next five years.