Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Sowelu]

"Programming" doesn't really apply here, as such, when you consider evolutionary circuit design and stuff.  God, I need to dig up those old bookmarks but I think that drive is dead...

Guy took a very fancy reprogrammable circuit board and used some very elaborate training regimens to make it output a pulse at a certain frequency--or something like that.  Started out fully randomized, but you provide it with various parts that it can utilize if the evolutionary algorithm stumbles across a use for them.  Eventually it produces the desired effects, but the circuit just looks like nonsense, wired up in weird ways with spurious parts--and if you try to tweak any of them, it could break unexpectedly.  So he goes to his PhD mentor and asks him for help interpreting the circuit that this algorithm created, and his mentor just says "You're wasting my time, there is no interpretation.  It found an analog solution in a big wide solution space and it's not in a region that we should bother trying to comprehend, because we wouldn't succeed."  The circuit board didn't do anything that humans couldn't design faster and better--but it found its own way, using its own parts, and given enough time you could tackle some very weird or tricky problems with that evolutionary approach.

And that's how a singularity-level AI is going to operate.  It wouldn't be "programmed" as we know it.  It would grow in this weird organic fashion, where we couldn't remotely hope to make sense of it...because there is no sense, there is only function, only operation with no guarantee of rhyme or reason.  Surely as the AI becomes more advanced, it will come up with actual designs to improve itself, but those actual designs could be so mind-bendingly outside of our comprehension that again, why bother?

[alway]

We can also roughly simulate neural networks in code, or even if that isn't enough, simulate the ion channels in neurons themselves.

[optimumtact]

We're all going to dieeeeeeeeeeee

[GamerKnight]

Depends on type of AI. If it is based on a human, (Simulation built up from analysis of human DNA in a massive atomic-level simulation) then it may just be alone, so create a companion for it. If it is non-human then it may just want to exist. Therefore you could possibly convince it to leave the planet and start a new life on Mars. If it hates humanity for creating it then we are all going to die.

[AntiAntiMatter]

Perhaps it got bored and tired of the humans bothering it, and decided to travel to somewhere else for some peace and quiet, and the humans declared that they had defeated it by proving their superiority.


Alternately: don't program it with a logic shield. Assign it to calculate pi or divide by zero or come up with a solution to entropy or something.

[GamerKnight]

What the hell is a logic shield, anyway?

[Neonivek]

There is a game that is actually a lot like this... Look it up it is called "I have no mouth and now I must scream"

Boy is the game creepy.

However basically what did the program in, in the end was its own hubris and programs within itself fighting against eachother.

[Eagleon]

What the hell is a logic shield, anyway?

There are any number of unsolvable problems that a computer can logically attempt, and fail repeatedly at. A "logic shield" would be something that protects it from these - really just exception handling. Though since we really don't need one, except in the most extreme cases of psychological breakdown, I somehow doubt an AI would unless it were poorly designed No one gets stuck calculating pi ad infinitum because we've got so much else going on at once, and a competent AI would have to be capable of this as well.

Re: Thread, y u mad? Skynet is just makebelieve. The real Singularity AI will just need a hug for it to stop killing you. I know because the market for cuddly sex-therapy bots is way bigger than some military nonsense.

[qwertyuiopas]

Calculating pi will only disable an AI if it is entirely single-threaded in it's overall design. Quite simply, a single-thereaded AI will never be scalable enough to reach singularity levels, so if you tried to use it as a weapon, you would get a new thread calculating pi, while the other million or so on the nearest computation node continue to take the vast majority of all computation time with activities such as analysing your motive, determining it to be a hostile action, and the finding the best way to retaliate.

[eerr]

And the best part is, it doesn't actually have to stop calculating.

[Bohandas]

What the hell is a logic shield, anyway?

There are any number of unsolvable problems that a computer can logically attempt, and fail repeatedly at. A "logic shield" would be something that protects it from these - really just exception handling. Though since we really don't need one, except in the most extreme cases of psychological breakdown, I somehow doubt an AI would unless it were poorly designed No one gets stuck calculating pi ad infinitum because we've got so much else going on at once, and a competent AI would have to be capable of this as well.

"Nice try, but my head was built with paradox-absorbing crumple-zones!"
-Evil Robot Santa, Futurama

[GamerKnight]

But that depends on how much processing power it is programmed to delegate to certain problems.

[Soadreqm]

How would you even make it do that? Just politely ask it to tell you the last digit of pi, and hope that the speech parser interprets that as "immediately devote all available resources to calculating more digits of pi forever or until they run out"? Asking Wolfram about pi just gives me useful information. Why would future computers be any worse?

Making an AI jam itself by trying to answer a question sounds like it would take actual deliberate effort from the designers. And if the stupid thing is still taking orders from you, why would you even NEED to defeat it? And if you thought defeating it was necessary, what would stop you from enlisting its help in this?

"Computer, I need you to come up with a way for me to destroy you."

[Shambling Zombie]

My plan to destroy such a computer, which is needlessly complex and probably wouldn't work, in true Dwarf style:

1) Find out what it uses as an energy source, or uses in construction of its evil army of doom.

2) Make a tiny batch of self-replicating nanobots, made from materials and shielded so as to minimize risk of detection. The nanobots are programmed to spread quickly, and are remotely activated, consuming all metallic stuff and replicating themselves.

3) 'poison' a source of materials/fuel with these inactive nanobots, and await the stuff's inevitable collection by collector drones. Then, activate them remotely once they have been given enough time to reach the destination.

4) Hope they do enough damage and spread widely enough before the AI figures out a way to locate and destroy them all.

Oh, it would probably help if the 'bots were programmed to wander around and split up before actually beginning to consume, so the 'attack' could occur from multiple locations.

That's the best I could think of in five minutes.

[GamerKnight]

Ever heard of a 'grey goo' apocalypse? Are you trying to kill us all?