Consider a Humanoid Robot that will not require daily charging. It toils in days, learning how to do it. This isn't science fiction. It is the promise of neuromorphic computing, an extreme new take on AI that is transforming the future of Robotics.
The current generation robots are power hungry. Their brilliant minds which are in most cases run on gigantic GPUs burn down batteries quickly. True freedom is a far-off fantasy. However, what would happen with a complete redesign of the brain of the robot? Suppose we were able to imitate the most productive computer in the universe--ours?
The Power Wall: The Existing AI Robotics is Stunning on a Ceiling
There is a significant bottleneck in the modern AI Robotics. The conventional processors are extremely inefficient concerning real-life tasks. They are continuously moving data in and out between distinct memory and processing units. This creates a traffic jam. It wastes impressive volumes of energy to do simple perception and make a decision.
One of the professionals in the Neuromorphic Computing Lab at Intel simply said: We are trying to create autonomous systems with brains running on full blast. It is a move like using a car engine to charge a smartphone. This inherent incompatibility is restrictive of possibilities. It maintains our highest ambitions in Robotics either by a power cord or by very short mission duration.
Nature teaches us: The Design of a Neuromorphic Chip
So, what's the alternative? Look to biology. Your brain is only powered by approximately 20 watts which is equivalent to a weak light bulb. Nevertheless it works better than any supercomputer. Neuromorphic engineering aims at replicating this efficiency on silicon. The point is that two fundamental changes are the key.
To begin with, it applies Spiking Neural Networks (SNNs). Such networks do not operate on data at all times. Instead, they only give information where and when necessary, in short electrical impulses. There is no stuff, no energy is used. Second, it brings memory and processing together. This does away with the energy consuming data shuttle. It is an entire renovation of the architecture.
Loihi 2: A Chip That Smells Like Intel
Let's get concrete. One such example is the Loihi 2 research chip by Intel. In another interesting case study, scientists applied it in order to design a fake nose. A small number of samples could be used in this system to identify hazardous chemicals. Amazingly, it did it during the power consumption of the milliwatts.
The chip acquired the new smells in one trial. It is a type of learning in a moment very efficient indeed.
This has enormous effects to AI robotics. One rescue robot would be able to maneuver around a disaster area. It would be able to detect the smell of a gas leak instantly. It would not have to link to the cloud. The brain is there, on a chip sucking energy.
Real-Life Effect: The Robot Revolution
So what does it imply on the future? The uses are disruptive. We would be able to observe environmental monitoring sensors scattered in a rainforest. They would operate for years. They would trace the changes in the ecosystem with little intervention of humans.
Another game-changer would be search-and-rescue robots. Rather than the flight time of 30 minutes, drones would cover days of disaster zones. They would go round and round seeking to find survivors. The basic idea is that this technology makes it possible to create a new breed of intelligent and persistent machines.
An Expert Leaves his Opinion: the Machine Nervous System Construction
I have recently interviewed an systems architect who was working on IBM NorthPole chip. She presented a different point of view. This to her is beyond chip design. We are creating a new nervous system of machines.
We are not simply coming up with higher-speed processors. Embodied intelligence is an engineered substrate. The boundary between the senses, processing and action is becoming unclear.
This is a profound shift. The AI does not only represent a program that runs on a computer. It is something that is inherent in the physical existence of the machine. This produces more fluid and adaptive behaviors of robots. It makes them seem more alive.
The Road Ahead: Crises and Discoveries
Naturally, the way is not easy. They have large obstacles. These spiking neural networks are complicated to program. It poses a complete challenge to developers and needs new tools. The ecosystem is still young.
Moreover, it is still challenging to increase production. What is the cost-effective way of mass-producing these new brain-inspired chips? The industry is yet to determine this. Nonetheless, there is no denying the progress. Technology giant investments and DARPA investments are a clear indication that this future has a high level of confidence.
A Last Speech: Efficiency as the Secret of Real AI
When we refer to artificial intelligence we tend to equate it to uncooked computational power. We think.--we think bigger and faster and hotter. But what is it the real machine intelligence that is not the brute force? What supposeth it be graceful, smart, and brainlike design?
One of the fundamental assumptions in AI is violated in neuromorphic computing. It is by means of smarter, more efficient brains in the robot itself. It may be the breakthrough that the age of truly autonomous, long-lasting AI Robotics relates finally to our daily life. The future of the intelligent machines does not necessarily require more power, but a better brain.
