I have been in tech for a long time, and at a place where I used to work, we constantly embedded computers in the equipment that we built (it was a military systems joint). So consequently the birth of a computer was a fairly standardized thing. You designed and built the hardware with it processors, registers, buses and drivers, memory, flipflops and latches, power supply, serial stuff and such. It was just a dumb piece of doped silicon. To make it behave like a computer to do things, you put the BIOS (Basic Input/Output System) into flash memory. That got kicked whenever you powered up. The BIOS did the checks, started up the sub-BIOS of things like video cards and serial interfaces, it started the peripherals and it booted the operating system. It essentially contained all of the instructions to make the dopey silicon appear lifelike and able to act intelligently. This entire process was on my mind as I was eating some watermelon on the balcony and the flies showed up. And then the mosquitoes joined the picnic. The flies wanted in on the sweet watermelon juices and the mosquitoes, all female, wanted a blood meal from my skin.
The arrival of the bugs caused me to think about the embedded knowledge in the brains of these insects. They hatched from an egg left to its own devices. They had no opportunity to learn behaviors. They live for a short period of time. And they are capable of some incredibly complex actions that sometimes defy understanding. For example, how does the only mosquito in your backyard find you from what would be miles away in terms of mosquito scale?
Sure they have chemo-receptors that detect carbon dioxide and they have faceted eyes that detect shapes and they can detect odors such as the fragrant esters that fruits produce and the musky odors that sweat glands produce. But if you think about it, it needs an algorithm to transfer that nerve excitation from the sensor into into complex action. You need to do some homing from the scent-source. You need to plot course to it. You need to do complex aviation maneuvers to reach the source.
Scientists have figured out how a female mosquito finds a human to get a blood meal so that she can reproduce. It is an amazing evolutionary adaption. She finds a host by using a combination of carbon dioxide detection, movement detected by faceted eyes at close range, smell for skin odors and bare skin, home in with a flight path and drill for blood all the while injecting an anti-coagulant. Then she metabolizes the blood to create eggs. Then she has to find a male to mate with. Afterwards she needs a puddle of water to lay eggs in. Then she dies. It is quite a bizarre sexual cycle with very specific needs. If you don't find a blood donor, then you don't have sex and you don't reproduce. It's a very weird path to sexual reproduction.
Imagine if a human female had to find a body fluid of another type of animal before having sex. Imagine if it were wombat sweat, which is just as weird as the idea of a very small insect getting 3 millionths of a liter of blood from a human. I think that if humans had evolved that way where the females needed wombat sweat to make their eggs, men would soon find a way to farm wombats just to have sex. But I digress.
So I am thinking about this and I start to wonder how many lines of code that it would take to convert the antenna smelling either fruit esters or carbon dioxide from the breath of humans. From that initial good whiff, the insect must translate that into a decision tree to pursue that smell. So it must calculate and extrapolate the scent stream and lay in a course for the juicy human. As it is, flying must have it scent radar on continuously in real time, and make course corrections to follow the strongest trail. It must also monitor flight conditions, such as a head wind, and must turn on the jets to overcome the spurious natural forces against it.
Once the insect is in close, the visual cortex takes over, and heads for the large object movement, all the while monitoring the scent trail. Once it gets close to the target, it switches from carbon dioxide to skin scent smells to find a bare patch of skin instead of one that is covered in clothing. It lands on the skin, and then deploys proboscis drill to find a pore in the skin. It injects the anti-coagulant and puts its phasers on suck to get blood. All the while, it must have its radar on to avoid the oncoming Big Swat from the human that will crush it to smithereens.
So the geek in me wonders how many lines of code it would take to program that action. For example in the silicon world of intelligence, the Ford GT has 10 million lines of code, which is about eight million more than what Lockheed used in the F-22 Raptor fighter jet, and at least three million more than Boeing needed for its 787 Dreamliner. And if you think about it, the mosquito or fruit fly has to sense, navigate and fly to the source of its attractions, not unlike the functions that you would find in these machines.
I immediately thought of how this "software" was created, stored and executed in a "dumb" insect who hatched from a puddle in an old tire and wasn't taught a thing as an adult, and yet knew how to do these complex things. After all, a baby is pretty much born with a blank slate and can do nothing but suckle, cry, throw up and poop its pants after it hatches. What makes insects so smart?
It only makes sense that the "software" was somehow stored in the tiny egg and copied to the brain of the insect when it hatched. Was it coded into the DNA? Was the "junk DNA" the source code or compiled code repository for these behaviors? Moreover, how many lines of code were there? Obviously the code had to be in some sort of compiled form. It was a complete puzzle to me -- until the Eureka moment hit.
I was reading an article that said that scientists have identified 50 distinct parts of insect brains. Furthermore, insects can live without their heads if when decapitated, they don't lose too much of their equivalent of cerebral-spinal fluid. The reason is that they offload neural processing such as walking to little brainules called ganglion distributed throughout the body. That accounts for the movement and locomotion of those multi-legged insects that move like robots. That means that their computer control systems are distributed. But that didn't tell me where the software was stored to account for complex behavior. After all, all of these systems require computation in the time-domain and other calculations like frequency-domain performance measures, degree of stability, Nyquist criterion (determining if feedback is faulty), frequency-domain of flying, and compensation techniques. It was a real puzzler as to how these advanced functions were passed on from the parent, to the egg and ultimately to the next generation. I didn't even have a theory.
Then came my dabbling in AI or artificial intelligence. It is an amazing process to create a machine with intelligence. You start by whipping up a network of artificial neurons. You initialize them with junk or random threshold values. Then you feed data into them over and over and over again, and you keep adjusting thresholds until they get it right. After all of that, you have a network that can do fairly sophisticated things like identifying cat pictures without writing a line of code to teach it. That was the Eureka moment. AI networks do not execute code to do their work. They have a built-in cascading network of layers of artificial neurons that sum inputs and decide whether to fire or not, based on a threshold value. They do sophisticated things with a multitude of simple, unsophisticated things. The insect Eureka moment for me, was that these complex behaviors do not need to be encoded in a layer of abstraction like BIOS.
For an insect, the brain and distributed ganglia throughout the body are made at hatching time from the DNA with a neural network, already pre-formed that does the sophisticated actions. There is no need for any learning, because the neural networks for what they need to do, has evolved over several million years. It is almost like making a computer with the software already loaded and running. In the case of humans, it is like being born with a pre-learned adult brain. It is as if walking was enabled in your brain the same way that the hand pulls away from pain or a burning flame.
So here is where the biomimicry comes in. If we now have artificial intelligence on chips, why can't I get a computer that doesn't need to have software loaded? Why can't I have a laptop that already knows what to do with a .docx file? Why do I need a special program like Microsoft Word? Why can't a computer just be taught to do stuff and then replicate the learning in the computer chips like the learning is embedded in an insect egg? Is this the future of computing?
I can see it now. There will be no more software packages in a new computer -- just a whole bunch of AI. Some of the neural nets are pre-trained, and others can be trained to do things yourself. And if you train some neural nets to do specific functions, you should be able to download a representation of those neural pathways and sell them to other users. It certainly could be a brave new world of computer dynamics if the development of AI took a turn like this.
No comments:
Post a Comment