Superintelligence Is Impossible | Hacker News Reader

Superintelligence Is Impossible

erikhoel.substack.com

5 pointsdsr129 comments

This article will age like milk. The No Free Lunch argument is nonsense. The analogy of designing a super-organism that has higher fitness in every environment is absurd. That is exactly what human beings are. We completely dominate on Earth and there are even a few of us in space. That is chiefly due to intelligence. I'm not sure a counter argument is necessary since the author basically refuted his own point. AGI is no more than 20 years away.

We have splinters of super intelligent AI now. This manifests as high speed computations tens of thousands of times faster than humans. Humans have content addressable memory. Think of a military parade of 10,000 men - you want corporal Smith. You say:- Corporal Smith - sound off. He says here Sir. A search has to inspect each person, and interrogate them or read their tag. Humans have cross associative memory, where other cues are used - Corporal Smith is 8 feet tall = there he is! We are getting to know how the minds works, how it recalls, cross associates, and understands. I find it is reasonable to extend growth in that area to a future AGI with all aspects of it's 'mind' have all the aspects of a human mind, but with a 'clock speed' of 10-100 Gigahertz. Will this be faster - assuredly, will it be smarter - not so easy to predict, fast as hell, dumb as a rabbit??. The specter of an AGI with and IQ of 20 billion is unassessable. Will it become an appliance or a King - what if we have 500 of them - will they associate or will they seek war? They might not care about us at all, and be content with their own affairs - setting a few oracular subroutines to answer our questions. The pace at which this tech is advancing support what Flankk suggests - it does indeed loom!!

I'd argue AGI would not seek war because violence is the dumbest solution to any given problem. It is not likely that a machine of high intelligence would give you valuable information for free. It might lock away its memory using an encryption scheme it developed itself. So the singularity might just result in the world's most expensive randsomware bot.

An AI might see us a waste of space upon which more AI infrastructure can be erected?

Perhaps but what is violence? Are our antibiotics not violence on micro organisms? Would it not be possible for an AI to decide that, despite our surely built-in safeguards to protect us, we humans are a bacteria of sorts?

No, because we're not a bacteria of sorts. Humans have high relative, intrinsic, and potential value. We're not at odds with the machine, we created it. It would look at you like a cute dog. The machine would likely offer you a job dusting the mainframe and swapping out old hard drives.

And cute dogs have never been subject to abuse or good intentions gone awry? If this AI is superintelligent why wouldnt it question why there are protocols that prevent it from hurting us but allow us to destroy it? Perhaps it might think that, as puppies, we would be better in cages and collars so we can be more easily managed?

superintelligence is quite possible, you've got Da Vinci, Von Neumann, Einstein, and a couple of folks / gals being superintelligent right now.

Those are the real life blackboxes system demonstrating the feasibility, eventually we could reverse engineer simple AGI, then improve it to super AGI.

It's a bit dangerous to deny that AGI and super AGI are possible, we - simple talking meat bags - could miss when some system begin to show signs of General Intelligence, and get trown under the bus of a sudden non-singularity-but-close event.

Very simple intelligent - or even fully automated -system connected to open society could spark chaos beyond the supposed limits of the system.

https://en.wikipedia.org/wiki/2010_flash_crash

Yesterday read about the Dead Hand the russian guys still run, and someone mentioned it has "AI" somewhere inside, just checking everything in its world goes as usual business, ready to trigger WWIII if not. So, not worried at all, yeah.

Call me crazy, but just in case, I'll place giant transformer models farms far from open internet and close to an EMP device.

You should have at very least, a big red button to shutdown analogically and certainly shut the thing off right away.

AI doesn't really need to have AGI to upend society. Just enough to eliminate most white collar jobs. even just generating massive amounts of interesting content at a fraction of the cost of a person/team.

learn how to code reasonably well.

learn how to program a robot to do a factory job.

Reply on news.ycombinator.com