Reproducing Human Vision Is Beyond AI Capabilities!

Computers can be faster than humans in identifying familiar faces or objects but their accuracy is questionable. Despite their power and promise, deep neural networks (a type of artificial intelligence) that resemble the human brain have yet to master visual recognition, according to a recent study published on 19th March and led by Marieke Mur, a neuroimaging expert at Western University in Canada.

The study looked at deep neural networks that processes incoming data through interconnected nodes or neurons. While these networks are designed to resemble the human brain, Mur's study found they still have major flaws when replicating neural responses measured in human observers viewing photos of objects like faces and animals. "While promising, deep neural networks are far from being perfect computational models of human vision," Mur said.

The implications of this research are significant, particularly for real-world applications like self-driving vehicles that rely on accurate visual recognition. If deep neural networks cannot fully understand the visual features indicative of their environment and categorize objects, applications will not be fully functional. In that case, they may make critical errors in identifying objects and making decisions based on that information.
 

It is also a fact that neural networks have a wide range of applications in various fields, including image and speech recognition, natural language processing, autonomous vehicles, robotics, fraud detection, recommendation systems and many more. They can also be used for predictive analytics, forecasting and optimization. With their ability to learn and adapt, neural networks are becoming increasingly popular in developing artificial intelligence and machine learning systems.

For instance, ImageNet Neural Network has achieved a better performance than humans. It classified the 1.2 million high-resolution images into 1000 different classes faster and more accurately than the average human!

However, according to Mur, "We suggest that neural networks can be improved as models of the brain by giving them a more human-like learning experience, like a training regime that more strongly emphasizes behavioral pressures that humans are subjected to during development."
 

 

While deep learning has made significant strides in recent years, this study shows that there is still a long way to go before computers can replicate the accuracy and complexity of human visual recognition.

Do you think by incorporating a human-like learning experience, we will be able to improve the performance of visual recognition systems? Pitch your thoughts in the comment section below!