Guest
New member
- Sep 6, 2024
- 128
- 0
Alas poor human...
I'd say it is. Not primarily because of what it could eventually do, but because of the kind of people who are now in control of it. Nuclear power could destroy us all. But it is heavily regulated and nuclear nonproliferation is taken seriously. But AI is still regarded as a sort of plaything for autistic types like Mark Zukerberg. People like that simply don't have the judgement necessary. They will keep trying these things because artificial intelligence feels less threatening to their autistic minds than human intelligence does.
It seems to me that there are two developments in AI that pose the greatest threat. One is the use of evolutionary algorithms. These are algorithms that go through an evolutionary process, modeled on biological evolution. The key point here is that nobody knows how any of these algorithms will finish up. Random mutations are added with each generation until the program has come to an end. At present, these algos are being used to generate relatively simple and harmless solutions. But, you see the problem. It is accepted that the algo can mutate into something that the programmer had not planned or foreseen. Indeed, this is regarded as the strong point of evolutionary computation.
The other threat that I see is the swarming of a multitude of AI generating computers, which again produce something more than the sum of the parts and produce effects that cannot be foreseen. Swarming is used to get small AIs to work together as bees do. Each bee has low intelligence, but thousands of bees together produce work of great complexity and reach accurate solutions to complex problems. It may seem far away now, but what if an AI is created, generated over a great multitude of computers in many different places, and this AI begins to demonstrate illogical or emotional reactions? Then we are in very serious trouble. And the politicians are paying no attention to any of this. It all seems too far away to bother about. But, really it isn't. Something like this could be happening within ten years.
I'd say it is. Not primarily because of what it could eventually do, but because of the kind of people who are now in control of it. Nuclear power could destroy us all. But it is heavily regulated and nuclear nonproliferation is taken seriously. But AI is still regarded as a sort of plaything for autistic types like Mark Zukerberg. People like that simply don't have the judgement necessary. They will keep trying these things because artificial intelligence feels less threatening to their autistic minds than human intelligence does.
It seems to me that there are two developments in AI that pose the greatest threat. One is the use of evolutionary algorithms. These are algorithms that go through an evolutionary process, modeled on biological evolution. The key point here is that nobody knows how any of these algorithms will finish up. Random mutations are added with each generation until the program has come to an end. At present, these algos are being used to generate relatively simple and harmless solutions. But, you see the problem. It is accepted that the algo can mutate into something that the programmer had not planned or foreseen. Indeed, this is regarded as the strong point of evolutionary computation.
The other threat that I see is the swarming of a multitude of AI generating computers, which again produce something more than the sum of the parts and produce effects that cannot be foreseen. Swarming is used to get small AIs to work together as bees do. Each bee has low intelligence, but thousands of bees together produce work of great complexity and reach accurate solutions to complex problems. It may seem far away now, but what if an AI is created, generated over a great multitude of computers in many different places, and this AI begins to demonstrate illogical or emotional reactions? Then we are in very serious trouble. And the politicians are paying no attention to any of this. It all seems too far away to bother about. But, really it isn't. Something like this could be happening within ten years.
