Brains and Supercomputers

The brain operates on the next order higher.

Although it is impossible to calculate precisely, it is postulated that the human brain operates at one exaFLOP (10¹⁸), equivalent to billions of billions of calculations per second.

When we discuss supercomputers or even computers in general, we refer to meticulously designed machines based on logic, reproducibility, predictability, and math over data. 

On the other hand, the human brain is a tangled, seemingly random mess of neurons that do not behave predictably. Still, the same architecture can be ported into a computer system.

The brain is both hardware and software, and the computer is the same; even if the architecture and the functionalities are quite different, both execute operation over data.

One of the most crucial difference is that inside the brain, the same interconnected areas, linked by billions of neurons and perhaps trillions of glial cells, can perceive, interpret, store, analyze, and redistribute at the same time. By their very definition and fundamental design, computers have some parts for processing and others for memory; the brain does not make that separation, making it hugely efficient.

The computing power of the brain is achieved in an extraordinary efficiency compared with supercomputers. The brain's power consumption in the order of 10-20 watts for 10¹⁸ operations per second. In a conventional supercomputer with the same computational power (based on CPUs and accelerators), the power consumption is in the order of 50 Megawatts!

The brain is millions of times more efficient than today computers!

The same calculations and processes that might take a computer a few million steps can be achieved by a few hundred neuron transmissions, requiring far less energy and performing far greater efficiency. What is interesting for the future is that the right virtual architecture computers can emulate the brain.

One of the things that genuinely sets brains apart, aside from their clear advantage in raw computing power efficiency, is the flexibility that it displays. Essentially, the human brain can rewire itself, a feat more formally known as neuroplasticity. Neurons can reconnect with others and even change their basic features, something that a carefully constructed computer cannot do at the hardware level. In any case, the computer can emulate the brain architecture using a sort of architectural virtualization or better using mathematical algorithms organized like in the brain, where structures can be dynamically adjusted to perform different tasks, analyze data, and learn from experience.

We call these mathematical structures "neural networks." 

We are still at the beginning of these structures' investigation, even if these structures were invented more than 30 years ago.

We did not have at disposition enough computing power in the past, but today things are changing. 

With neural network and their variants, we can emulate functions typical of a brain like, prediction, learning, intuitions, and creativity. The still limiting factor is the computational power and the energy efficiency that is still far from the biological brain, but we will solve this gap in the near future.

A3Cube, with its GRIFO™, can reach 10¹⁶ operation per second, sufficient to create the next-gen of AI today, with "only" 30KW of power consumption; we can think to get the power to emulate the human brainpower at an affordable price by 2023.