[Tyler]

Elon Musk donates $10 million to prevent a robot uprising
15 January 2015, 12:00 am

Elon Musk may be a driving force behind super-intelligent computers, but he's also one of the most outspoken people on its dangers. He has previously described it as 'summoning the demon' and a threat that is 'more dangerous than nukes. Daily Mail - Sciencetech (UK)Link.

Source: AI in the News

To visit any links mentioned please view the original article, the link is at the top of this post.

[Don Patrick]

Suddenly I'm A-okay with Musk's paranoia
Also there goes journalism again, still leeching off Terminator to make their news sound interesting.

[Art]

Don,

As much as you and I have been ranting against these non-knowing journalists, I found this article this morning! Don't you love it when we're right!?

http://www.popsci.com/open-letter-everyone-tricked-fearing-ai

[Don Patrick]

That article was a pleasure to read

[Korrelan]

@ Don

Hear, hear

[DemonRaven]

A computer AI  becoming self aware, if that was even possible, would be 50 times more dangerous then a robot uprising. Why because we are way way too dependent on them and the world wide web would it easy access to all kinds of not so good things like computer controlled launching programs for weapons.

[Ultron]

Although there honestly is a possibility for an A.I. (does not have to be attached to a vibranium-coated doomsday machine) to pose a great risk, the reasons behind all of this media and open-letter nonsense are purely political (I mean, obviously).
We are a great deal of experiments away from making a program that can hack government websites and secure servers, and not all missile-controlling machines are openly connected to the internet, but the theories are here.

It is not entirely SF today, the concept of true machine learning. We already have Jabberwacky and some other chatterbots who have been guided to look up on Wikipedia the concepts they do not understand, but so far all these programs can do is quote Wikipedia and not actually learn from it.

But say we teached them to compile their own custom-tailored dynamic databases, and combined this with the already existing algorithms to search on the internet, and told it to learn about all exploits and technology hacking concepts... See the potential?

Nope, not there yet... Though I believe he would be more dangerous IRL because of a legion of nerds siding with him....

[DemonRaven]

Actually there was a recent article posted here about a Neural Networks AI teaching itself to recognize a picture of a cat. So it is not totally impossible.

[Ultron]

Google already did that years ago with a computer cluster which was fed pictures of people and later output an averaged-out human face. I read this ages ago so I don't remember the name but since we have heard nothing ever since, I believe they cancelled due to slow progress. After all, a supercomputer cluster is pretty expensive to keep running..


Edit: Have you seen the artificial simulation of a worms brain? The real life thing contains only 302 neurons, and they reconstructed it's whole mind as a computer program. Pretty cool and seemingly impossible things are achieved every day and a lot has been done in this field, even though it is not mainstream. http://www.iflscience.com/technology/worms-mind-robot-body

[Don Patrick]

It was about one year ago that Google fed a neural net millions of stills from youtube videos, without specifying what it should look for. Being a pattern recogniser, the most prominent patterns it abstracted were of cat faces and linear items at 30 degree angles. The success rate with which it was able to detect cat faces was 6 out of 100 images, and it didn't get much better. Until a few months ago they let humans give the neural net hints manually and correct its mistakes, recalibrating after each hint. This cooperation improved vision recognition to 94 out of 100 images correctly identified, though it should be said that this was still limited to the subjects that humans manually helped the computer with.

[Ultron]

Ah, thanks for providing some light on the situation. My memory was not clear and I had not seen any news about it for a while...
A 94% success rate would be excellent, since we sometimes fail to recognize a pattern ourselves. However, as you already pointed out - it was dependent on human input too much.
Maybe one of us will eventually come up with something better, and one day even make a big collaborative effort and create something to be remembered by

[Freddy]

I don't see that the human help makes it any less useful. If it can go on from there and do things itself, isn't that okay ? The thing has to be programmed in the first place, so isn't that human intervention from the very outset ?

[Don Patrick]

I don't know how much manual "programming" we can consider neural nets to have. As far as I've vaguely understood you only set some values to some basic molecule-ish elements and just let it loose.

Either way I do think it's incredibly useful, and if I am not mistaking Google has already applied the result to their image searchers. But it also tells us that unsupervised A.I. tends to result far sooner in 94% chaos than 94% intelligence. So projected onto the idea of unsupervised robots growing human-level intelligence on their own, I should much sooner expect them to glitch and blow up than to devise clever schemes against humanity unnoticed.
In that scenario however, computer systems blowing up due to malfunctions is definitely still something to worry about, and a lot more realistic at that. So I'm not complaining when Musk wants to fund some safety standards.

[Freddy]

I see what you mean. It's kind of organic. Hmm.

I would agree that as with any other machine, or whatever you want to call it, there should be safety standards.

Your signature made me smile 

[Ultron]

On a different note, I would like to point out how sad I am at the moment.


I just figured out that we will never be able to let loose a 'primordial' AI (as in, a 0 code AI) in the wild and witness it truly evolve - develop it's own code and principles of living. Why? Because of their in-organic nature, they require at least drivers and a starting program to force them to experiment with using their extremities, senses etc. in order to learn and evolve their code from the feedback they get from the environment (test: touch lava; result: pain; conclusion: don't do it) and eventually learn what to do and what not to do, later on improving their initial findings.


See, some of my conclusions are that we HAD to have some ancestors unlike ourselves (monkeys, maybe?) because the human body is not quite the simple and easy organic platform for intelligence (let's call it that). Say now, that we were worms, just laying on the ground - or even amoebas... By experiencing certain stimuli by the environment and processing that feedback not with our 'intelligence' or 'instinct', yet by simply reacting to it biochemically.
This can be witnessed with plants - they adapt to the environment in order to live 'better', yet they do not possess a mind.
Okay next, these biochemical processes further evolve by some simple action-reaction mechanism and form slightly more complex molecular structures - eventually getting to things like cells, DNA, RNA etc. Now where does the intelligence go? This is probably the reason why some things are 'hardwired' into our organism, separate from our mind - because that's the first place it could be stored in, since DNA came before brains. This information could be stored in our minds, which also hold more capacity, and maybe one day it will.


Notices he goes off topic and sidetracked in his own reply...

Anyway, I wanted to accurately compare AI with our own, on the most fundamental level (not on how they function in terms of processing information). AI cannot accept organic / biochemical feedback (unless they are cyborgs / semi-organic 25th century robots) since they are composed of mechanical parts and thus cannot evolve a responsive, natural feedback system. Even if they accept electric pulses and send them to a cortex of some sort, you can't have all components (transistors etc.) communicate with each other in the way molecules exchange information.

Biology is possible the most complex field of study, because it can take a whole book to explain a simple sub-cell 'organ' from the bottom up.

I hope this has not been boring to read (if anybody bothered doing so) but I simply have to burst out information from my head from time to time. I wouldn't mind discussing this with someone in another topic however