/sci/ - Science & Math

>>2222985
>(cont)

Imagine we finally manage to create an AI based on artifical neural networks. We have no idea how to program those things, the only way to use them is train them. Like a child.
So you train a supercomputer like a child for 15 years. Teach it how it works. Let's assume it has the potential to be smarter than a normal human... it still has to understand all things to make use of its raw processing speed. This will take time.
Now, let's say after 25 years (optimistically) it can finally start improving itself. But it's only a single smarter-than-human entity. It's not godlike, it's just very smart. So it'll have to work for years, together with the assistance of humans.

So, let's say after 30 years it finally has made a breakthrough. Ok... but this breakthrough is incompatible with its own system. We have to implement it on a new supercomputer and begin teaching a new AI, from scratch.

The singularity may come, but it will consume itself by making things better at the cost of increased complexity