Authored by Paul Dlugosch, President & CEO
Nature and evolution are never wrong. Evolution can’t be wrong – the survival of entire species is dependent upon evolution working. Nature and evolution are also patient. While evolution may not be fast, the results it produces are truly remarkable in efficiency and elegance. The slow nature of natural evolution should produce outcomes that are the best quality possible. Imagine your boss assigning you a task to which you respond, “Sure boss, I’ll have that done in about 10 million years.” Your boss would be entitled to expect you to deliver something truly spectacular.
Humans are far less patient than evolution. In our quest to be first and to succeed, we often ignore the most valuable examples that can lead to the solutions we are searching for. Evolution doesn’t hold office hours and we can’t schedule appointments with evolution to discuss the merits of one design choice versus another – but we can and should strive to study and exploit the examples that evolution has left all around us.
Although evolution can be an excellent teacher, there is peril in taking the results of evolution too literally. Our ability to fabricate machines is, in many ways, quite limited compared to what nature can produce. Still, evolution leaves subtle clues that we must look for in order to construct man-made machines that will perform similar functions.
Nothing highlights the consequences of ignoring evolution or the futility of trying to copy nature too closely than humans’ quest to conquer the challenge of flight. Some of the examples discussed below can be seen in the video, ‘The Wrong Brothers’. It is entertaining and enlightening to watch these early attempts at flight.
Ignore Nature at Your Own Peril
Imagine trying to fly by quickly moving an umbrella up and down. Yup, we tried that. It’s amazing the test pilot didn’t suffer a concussion in the attempt. It’s hard to be too critical, though. On one hand, it seems like it might work. If you could just move that umbrella fast enough… In this attempt at flight, you will see an incredible amount of energy poured into the effort. This misuse of energy is often a signature trademark of ignoring nature. Nature is incredibly efficient because it has to be. Survival often requires conserving energy. For example, migratory birds can’t afford to misuse energy as they fly cross-country twice a year. Evolution has ensured that they are incredibly efficient in flight. Watch a flock of Canadian Geese flying south in formation – a true demonstration of elegance and efficiency thanks to evolution.
Equally spectacular failures occur when we try to take nature too literally. Nature can do things that human’s machines just can’t. Consider the early flying machines that included flapping wings; even to this day, it’s extremely difficult to build flying machines that behave like birds flapping their wings.
Several of the early attempts at wing flapping resulted in destruction of the aircraft while it was still on the ground. Here, again, the attempt is understandable and the inventors get credit for at least acknowledging nature and trying to leverage the principles that were visible to them. Examples like these remind us that we must recognize the limits of what we can achieve. The trick isn’t to exactly copy nature, but to learn the fundamental principles that allowed nature to succeed.
Of course, we all know how this epic quest for flight turned out. Eventually, the wonderful elegance of the airfoil was discovered. Thereafter, motors were added to generate the thrust necessary to pull the airfoil through the air. While we couldn’t build reliable flapping wings to generate thrust, we could build motors and propellers. The key was to recognize the need to push or pull the airfoil through the atmosphere. If we couldn’t flap our wings to generate thrust, then we had to find another means to achieve the same motion. Humans are really good at these kinds of workarounds.
What does this have to do with AI?
Like the challenge of conquering flight, the quest to develop intelligent machines is one of the most challenging and vexing problems facing mankind today. And, true to the adage, “those who do not learn from history are destined to repeat it.” While perhaps not as entertaining as videos of early flying machines, we can see the same kinds of futile attempts as we toil to develop intelligent machines.
In some cases, like the early aviation pioneers, we are entirely ignoring nature. Instead, we have chosen to work harder and waste tremendous energy like the reciprocating umbrella of our predecessors. Understand that traditional computing is an entirely man-made concept. There are simply no examples in nature that will show us how to calculate a spreadsheet faster or with more precision. We are compelled to use the machines we’ve developed for this unique purpose in our effort to build a kind of machine capable of exhibiting natural intelligence. Should we really expect that to work? Today’s processing technology was not based in any way, whatsoever on the natural processing models we see in nature, and the inefficiencies current processing technology demonstrates provide ample evidence of such. Consider that the human brain operates on approximately 20 watts of energy. Humans can easily perform tasks that today’s computers cannot, yet these computers consume hundreds or even thousands of watts of energy in their efforts to exhibit ‘intelligence’. If we could just move the umbrella faster, maybe we will fly…
It is also true that we have erred on the side of trying to exactly mimic nature. Just as we found it difficult to build reliable flapping wings, we also find it very difficult to build high fidelity neurons like the type seen in the human neocortex. But, that hasn’t stopped us from trying. The human neuron is an incredibly complicated electro-chemical cell. Attempts to produce an equivalent to this cell consume incredible energy and capital. The SpiNNaker project is a prime example of the complexity of attempting to create a network of high-fidelity spiking neurons. Projects like these can require one million processors to begin to achieve only one percent of the neural capacity of the human brain. While noble and useful for scientific exploration, these kinds of projects are hardly efficient or practical.
All of this begs the question: does artificial intelligence have its own version of the airfoil? Can we learn from nature to build an intelligent machine, even if we cannot copy nature exactly? Can we achieve better results with less energy than our current processor architectures are capable of? At Natural Intelligence, we believe the answer to this question is a resounding yes.
Current processing architectures aligned with the man-made invention of sequential instruction processors are wonderful inventions and truly remarkable in their intended field of use. They are also poor choices to implement the natural efficiency of the human brain. The team at Natural Intelligence is not compelled to force-fit a legacy architecture into doing a job it was never designed to do. Instead, we took the opportunity to start from scratch, watch the birds in flight and think from a fresh perspective about how processors could be built. Get ready for take-off…