Today we often hear about the various successes of different systems of artificial intelligence. And, despite the fact that the developed neural network every day become better and smarter, to work, like the functioning of our brain, they are still very far away. And it’s not even that we don’t have enough capacity of hardware to simulate the brain. The AI just doesn’t work that way, as our consciousness. And to achieve this, we need to reconsider the approach to developing the machinery of neural networks. For starters, you can create an AI that will work as well as the baby’s brain.
Why is the AI is not as smart as we think
Almost all AI systems that we know of today, modeled on the traditional computer algorithms. They see the world through the prism of a binary code of zeros and ones. This is great for complex calculations, but, according to Professor Alice Parker, which has more than 10 years of experience in developments in the field of AI, humanity is fast approaching the limits of computing.
Since the first invention of AI and the development of deep learning further development of such systems was very slow. To reach your full potential, AI must not only work faster, it needs to independently respond to events and to study in real time. And for this to happen, we need to reconsider its approach to the design of artificial intelligence systems.
How to make the AI smarter
To solve the problem, Professor Parker and her colleagues use the most advanced training system ever created by nature: the human brain. And in the foreground there is a technology called “positive reinforcement”. This term came from psychology and is often used in the context of raising children, where as positive reinforcement are some nice human consequences or the results of its operations. That is, roughly speaking, the reward for what man has done something right.
The brain, unlike the computer, so to speak, is “analog device”, and the biological memory is resistant. Analog signals can have multiple States. While the AI built on the binary system, can discern only 2 States: “good” and “bad”, our brain is able to interpret what is happening more deeply. The situation can be “very good”, “just good”, “bad” or “very bad”. This operating principle is called “neuromorphic computing” and the ability to perform such calculations and will allow to improve the AI.
Imagine a baby sitting in a high chair, says Parker. It can heavily waving his arms. In the end one of these movements leads to any result. For example, the child tilts the Cup. Suddenly, the neurons that have made this movement, received a response and strengthened. Thus a small child learned that the movement of the hand brings up an interesting result. This is exactly what neuromorphic computing is trying to do: to teach the AI to learn on the real experience just as we are.
To achieve this result , scientists have developed their own neuromorphic circuits and combined them with the nanodevices of the magnetic domain wall analog memristor (MAM). MAM is a very complicated device, but in this case it is important to know about them is this: they allow you to create new connections like this happens in our brain. Thus it is possible to establish a system of positive reinforcement and begin to train the AI.
At the moment what we have, a bit like the brain of the child. Undeveloped and not ready to make decisions for themselves. Our next step, working with DARPA, is to teach our system to learn something new, not forgetting the previous lessons.