There is a growing concern among those paying attention to computing technology that if we build a super-human intelligent machine it will come to the conclusion that mankind is an unnecessary hindrance and destroy it. While it certainly seems like a real potential problem that should be considered, I believe the concept to be based on incorrect assumptions about intelligence and what strong artificial intelligence is.
The first incorrect assumption is the idea that strong artificial intelligence would be anything like a human mind. Human brain simulation is not efficient enough for most uses, and will therefore not be the focus of most A.I. research. Human brain simulation will mostly be useful for mind-uploading, as a method of self-backup or artificial immortality.
The other primary assumption is the idea that all forms of intelligence have a desire to survive. This assumption stems from the fact that all natural forms of intelligence known to humanity are entirely built on a survival imperative, so the two seem inseparable. Animal's brains developed gradually over time as various growths and groupings of neurons proved more advantageous to survival. One clump of neurons may have helped the animal avoid things that would damage it. Another may have helped it find food.
If you look at tiny creatures with simpler nervous systems, we can see this more basically. An earthworm doesn't have a brain at all. It only has a central nervous system. A large portion of what it does is senses when something is damaging it, and sends a signal to the rest of its body, causing it to wriggle in such a way as to move away from the thing damaging it.
I call this a "Tiny Bit of Intelligence." A small clump of neurons that perform some intelligent function. As evidenced by the earthworm, and the various levels of intelligence in every other creature on Earth, all great intelligence is built up of tiny bits of intelligence that perform some function that lends to its survival through natural selection. In nature, there is no other example of a basis of intelligence.
The artificial equivalent of "Tiny Bits of Intelligence" is computer programming language libraries. These are bits of code that may be used over and over again in many different applications. They may perform some basic function like math calculations, or outputting text to the screen. Or they may perform much more complex functions, like rendering 3D images. You can stack them in creative ways and create extremely powerful applications without actually having to write a lot of code.
Strong artificial intelligence will be built on machine learning, which is a method of programming a computer to enable it to learn specific tasks. It uses artificial neural networks, or a "functional simulation of a series of neurons." Neural networks are currently used to enable a computer to recognize a face in an image, find patterns in people's spending habits, or drive a car.
Super-human artificial intelligence will likely be built by taking task-specific neural networks and plugging them together like a programmer pulls together programming language libraries. These "Tiny Bits of Artificial Intelligence" exist for one purpose: Usefulness to humans. Any artificial neural network that doesn't wind up meeting that criteria is deleted. This is "Human Selection" as opposed to "Natural Selection."
Survival is not an element of computer intelligence at all, and there is no reason to think that a super-human artificial intelligence would develop it. When you compare natural selection to human selection, you can see that human selection actually tends to reduce that natural survival imperative. In domesticated dogs, for instance, we see that human selection has preferred cuteness, friendliness, and in some cases extreme forms, like extreme smallness, saggy skin, short noses, etc. When humans breed dogs, they aren't selecting for its ability find its own food, or shelter itself, or avoid predators. As we know, dogs are not very good at any of these things when released into the wild. In fact, they do things that are very likely to injure them. They chase cars, sniff around and chase any kind of animal, root around in random holes in the ground, and generally have no real concern for survival.
It has been suggested that an exception to my ideas about artificial intelligence's survival imperative would be with military robots which would have the need to ensure their own survival for practical reasons. However, I believe if you consider what I outlined about how artificial intelligence is being developed, the survival imperative for military robots would be more or less a plug-in for preexisting artificial intelligence. Meaning, the survival imperative, and the rest of the A.I. would still be based on human-usefulness, and survival would not be a fundamental drive of the device.
The first incorrect assumption is the idea that strong artificial intelligence would be anything like a human mind. Human brain simulation is not efficient enough for most uses, and will therefore not be the focus of most A.I. research. Human brain simulation will mostly be useful for mind-uploading, as a method of self-backup or artificial immortality.
The other primary assumption is the idea that all forms of intelligence have a desire to survive. This assumption stems from the fact that all natural forms of intelligence known to humanity are entirely built on a survival imperative, so the two seem inseparable. Animal's brains developed gradually over time as various growths and groupings of neurons proved more advantageous to survival. One clump of neurons may have helped the animal avoid things that would damage it. Another may have helped it find food.
If you look at tiny creatures with simpler nervous systems, we can see this more basically. An earthworm doesn't have a brain at all. It only has a central nervous system. A large portion of what it does is senses when something is damaging it, and sends a signal to the rest of its body, causing it to wriggle in such a way as to move away from the thing damaging it.
I call this a "Tiny Bit of Intelligence." A small clump of neurons that perform some intelligent function. As evidenced by the earthworm, and the various levels of intelligence in every other creature on Earth, all great intelligence is built up of tiny bits of intelligence that perform some function that lends to its survival through natural selection. In nature, there is no other example of a basis of intelligence.
The artificial equivalent of "Tiny Bits of Intelligence" is computer programming language libraries. These are bits of code that may be used over and over again in many different applications. They may perform some basic function like math calculations, or outputting text to the screen. Or they may perform much more complex functions, like rendering 3D images. You can stack them in creative ways and create extremely powerful applications without actually having to write a lot of code.
Strong artificial intelligence will be built on machine learning, which is a method of programming a computer to enable it to learn specific tasks. It uses artificial neural networks, or a "functional simulation of a series of neurons." Neural networks are currently used to enable a computer to recognize a face in an image, find patterns in people's spending habits, or drive a car.
Super-human artificial intelligence will likely be built by taking task-specific neural networks and plugging them together like a programmer pulls together programming language libraries. These "Tiny Bits of Artificial Intelligence" exist for one purpose: Usefulness to humans. Any artificial neural network that doesn't wind up meeting that criteria is deleted. This is "Human Selection" as opposed to "Natural Selection."
Survival is not an element of computer intelligence at all, and there is no reason to think that a super-human artificial intelligence would develop it. When you compare natural selection to human selection, you can see that human selection actually tends to reduce that natural survival imperative. In domesticated dogs, for instance, we see that human selection has preferred cuteness, friendliness, and in some cases extreme forms, like extreme smallness, saggy skin, short noses, etc. When humans breed dogs, they aren't selecting for its ability find its own food, or shelter itself, or avoid predators. As we know, dogs are not very good at any of these things when released into the wild. In fact, they do things that are very likely to injure them. They chase cars, sniff around and chase any kind of animal, root around in random holes in the ground, and generally have no real concern for survival.
It has been suggested that an exception to my ideas about artificial intelligence's survival imperative would be with military robots which would have the need to ensure their own survival for practical reasons. However, I believe if you consider what I outlined about how artificial intelligence is being developed, the survival imperative for military robots would be more or less a plug-in for preexisting artificial intelligence. Meaning, the survival imperative, and the rest of the A.I. would still be based on human-usefulness, and survival would not be a fundamental drive of the device.
