Artificial intelligence research often mirrors human brain traits to find the best way to develop the technology, and IBM is more of a company doing it.
At a conference held in Melbourne, Australia, researchers from IBM presented two scientific articles that go this route: one seeks to create a “period of attention,” while the other examines how to apply the process of birth and death of neurons to systems of machine learning.
The attention algorithm mirrors how this works in the human brain. “Attention is a reward-driven mechanism,” researchers Irina Rish of IBM told Engadget. Thus, it informs IBM’s neural network which entries offer the best reward – the higher the reward, the more attention it takes.
“We know that when we see an image, the human eye has a very narrow visual field,” continued the researcher. “So depending on the resolution, you can only see a few pixels of the image in detail while the rest is blurred. Then you quickly move your eyes so that the mechanism of association of the different parts of the image allows you to recognize what it is.”
This system will work in tandem with the other technology developed by IBM researchers, which serves to make the AI memorize images. Based on neurogenesis – the process of birth and death of neurons in the human brain – the system uses new data that is presented to it to form new connections with old data, creating a memory for it.
So while one system uses the attention span of the human brain as the basis for making artificial intelligence focus on one image, the other system based on neurogenesis is about making it learn and memorize the image.
The systems are quite complex and therefore should not find practical application any time soon. “I would say this should take a few decades – but it’s a guess,” Rish explained. “We’re just scratching the surface.”