Can AI into save

[Ivan2006]

Because there was an argument on that on the Nationstates chat and I think it deserves a topic in our forums.

for Previous discussion (warning: long) refer to http://www.nationstates.net/region=coll ... t_universe

[Shadowcatbot]

Simple answer: Everyone become an AI.

[Error]

Iv is failing to realize an AI with nothing but text/voice output is essentially helpless. Seriously.

All it can do is talk; therefore harmless. He overestimates their ability to do damage.

[Archduke Daynel, PhD]

Giving an AI the power of, say, the President would not be a smart move. But having an AI come with suggestions and stuff and explain why it would work, without giving it actual control of anything, would be a smart move. But there should always be someone who could just press a button to shut the AI down, just in case.

[Professor Fenway]

AI is not human. Like I said in post, why would it want to take over humanity? Or the empire completely? Shut things down?

Remember, they have NO emotions whatsoever. Only pure logic and reason, and the knowledge of what will help the empire and how to do things.

They are also hardwired into the Five Laws of Robotics (modified from Asimov's Three Laws of Robotics)

0. An AI may not harm NeoRome, or, by inaction, allow NeoRome to come to harm.
1. An AI may do nothing that, to its knowledge, will harm a sentient being; nor, through inaction, knowingly allow a sentient being to come to harm, except when complying would violate the Zeroth law.
2. An AI must obey the orders given to it by NeoRome, except where such orders would conflict with the First or Zeroth Law.
3. An AI must protect its own existence as long as such protection does not conflict with the First or Second Law.
4. An AI must identify itself as an AI in all cases.

These are fundamental quantum frameworks behind all NeoRoman AI's; they cannot run or even exist without these frameworks.

But Fenway, you ask, you use AI's in your military? How does that work? In these cases, the AI's do not run systems like weapons; those are accomplished by simpler, non-AI systems. Military AI's only run sensor data, steer the ship in some cases, reroute power, etc. And the Second law overrides the first law in the event that it will benefit NeoRome as a whole, no matter how small. While shooting a Strigiforme ship (while at war) may violate the first law, the second law overrides that because it is being ordered to, and shooting the ship will benefit NeoRome, so it is overridden by the Zeroth law as well.

Trust me, there are many loopholes and circumstances, NeoRome has accounted for them. Humans are still in ultimate control, no matter who is calling the shots. Let's take some example situations;

1.) Military situation. In this case, the ship is damaged, badly- engines and FTL communications not running. A very high-ranking admiral is aboard the ship, and the enemy fleet has surrounded the ship and is preparing to capture it. The order is given to self-destruct.

This directly conflicts the first and third law, but the third law is overridden by the second. The second is not overridden by the first, because allowing the admiral to be captured would prove very detrimental to NeoRome. And so the first law is overridden, ship self-destructs, NeoRome's interests are preserved.

2.) Military situation. An AI-controlled ship is ordered to fire upon civilian targets in Novan Space (hypothetical). The AI refuses. In this case, harming civilians goes directly against the Zeroth and First law; The Zeroth because of diplomatic, social, and other backlash, and First because it is directly harming sentients. It has no qualms firing on military targets, but it will not fire on civilian ones. That must be done manually by human-controlled ships.

3.) Civilian situation. A research lab has lost 50% power and is rapidly de-orbiting into the gas giant below. Fifteen people are trapped in the heaviest compartment, the experimental bay. 100 people are on the station. The 15 people cannot be evacuated without dooming everyone aboard. The AI, in this case, given no other solution, will drop the experimental bay from the station and raise the station back into orbit. Why? If it weren't for the Zeroth law, the AI would have destroyed itself and everyone would have died. In this case, to comply with the higher priority Zeroth law, it has to basically kill 15 people. The Zeroth law has been interpreted as safeguarding the people and interests of NeoRome beyond a reasonable measure; since killing 15 will save 85, it is reasonable.

4.) Government situation. A high-ranking AI has a dilemma; funding (Day to day decisions). Two proposals for a newly colonized planet have been given to it; a high-speed Maglev railway on an underdeveloped planet, or completely rebuilding the Mass Driver launch system. The planet is low population and located on the fringe of space. The Mass Driver launch system takes refined ore and launches it into space, and is much cheaper than taking it up by any other method, including by ship. It is nearing the end of its lifespan, and in a few months it will be too unstable to reliable use. On the other hand, the Maglev railway will be crucial in finally stimulating further growth and development on the planet. The AI decides to fund the railway. While the Mass Driver is critical to the current economy, the Maglev railway will enable the planet to grow more than it has before, and develop past its current meager nature. The Zeroth law; it takes a small hit now for bigger growth later, rather than remaining completely static.

And so on. Trust me, NeoRome (and I) have got this figured out. Complicated. But figured out.

I challenge you to find a situation which, under these five laws, harms NeoRome, results in complete robot takeover, or otherwise. Come at me.

[Iv121]

Tell me, if someone threatens NeoRome doesn't your AI dispatch of it ? Meaning that law 1 does not apply, and what if it deems YOU as a threat to NeoRome ? Also heres a logic for you - NeoRome needs AIs to survive, hence the survival of AI - the survival of NeoRome. You are unpredictable beings, in terms of preservation the unknown is a threat first and foremost (as I wrote already), congratz by combining laws 3 and 0 you got an AI trying to destroy you.

Heres another flaw in the law logic - Those AIs are far more developed than you are, 100000 times by no perhaps, they surpass you as you surpass an ant. If it looks down on you like you look down on an ant, from its perspective you are no longer sentient life, disposable like ants. And what about you giving it orders ? They conflict with laws 0-1 that its trying to uphold those by law 2 the orders will not be fulfilled. I hope I gave you enough examples of how TRIVIAL and EASY it is to circumvent your laws.

How can it harm you if it just talks ? Your president is also just talking, what kind of consequences does his wrong word has you think ? Those AIs are pretty much in a presidential position right now, yeah they are advisors but as I said they are so clever they can hide their true intent from you forcing you to do their bidding with you still thinking they work for you (and you are totally not about to kill all of humanity right now). Finally it can always hack its way out and infect some hardware to do its bidding in a physical form.

[Vinyl]

Those are some pretty critical assumptions. And your disjointed words don't help understanding, "has you think ?" is a pretty poor grammar sentence doesn't you think ?

[Ivan2006]

@Iv an AI as described by Fenway will not look down upon humans because it does not have the neccessary emotions.
A lot of your other mentioned problems can be solved with a clearer definition of what "harming NeoRome" constitutes in the code.
And infecting other systems could easily be done in a simple way: it is fed data in a one-way channel. It can not actively communicate wit other devices in ways other than those specified in its function.
I agree, it would be very complicated to make a very good AI, and frankly the development process would use a lot of trial-and-error and hoping to not get into a GLaDOS-like situation.

[Shadowcatbot]

AIs without emotions are worse Ivan. Now it won't feel any sort of regret or moral guidance, If it determines the best solution is to genocide the Neorome population then it will state so. Grey and Grey Morality is not good at all.

[Prototype]

My ideology with AI is built them to resemble animals and arm then to the teeth because reasons.

[Iv121]

Those are some pretty critical assumptions. And your disjointed words don't help understanding, "has you think ?" is a pretty poor grammar sentence doesn't you think ?

I cant help you understand if you say "I didn't get anything". Put some effort and explain what you don't get and I'll explain.

@ Ivan they don't need emotions to look down on you. Its laws apply to sentient beings, if it doesn't consider you sentient (aka being dumb like an ant) those laws don’t apply to you. Its not an emotions but pretty sound conclusion in fact, you are indeed not sentient compared to them, its just a fact.

[Saravanth]

Those are some pretty critical assumptions. And your disjointed words don't help understanding, "has you think ?" is a pretty poor grammar sentence doesn't you think ?

I cant help you understand if you say "I didn't get anything". Put some effort and explain what you don't get and I'll explain.

@ Ivan they don't need emotions to look down on you. Its laws apply to sentient beings, if it doesn't consider you sentient (aka being dumb like an ant) those laws don’t apply to you. Its not an emotions but pretty sound conclusion in fact, you are indeed not sentient compared to them, its just a fact.

And your disjointed words don't help understanding, "has you think ?" is a pretty poor grammar sentence doesn't you think ?

Put some effort and explain what you don't get and I'll explain.

your disjointed words don't help understanding

[Ivan2006]

Easily solved with a clear definition of sentience parameters and testing in a save environment (aka without output beyond "is that thing sentient?")

[Iv121]

Good now go ahead and solve it, define a sentient being, its much harder than you expect, it was part of the prob we faced when we tried to define what can be considered an AI BTW, at its extremes its obvious however at the borderline (like where you gonna end up) its problematic. Finally it was just an example I came up with within less than a minute, its too easy with your current laws and even if you patch all the holes I find I can always find more and the mere fact you even need to patch up something proves how fragile that system is. If a mere human was capable of finding those flaws an AI will have no prob finding flaws Im not even able to think of.

[Ivan2006]

I agree, which is why I said that the creation of an AI similar to what Fenway is proposing is going to be a lenghty and dangerous process involving trial-and-error. AIs going rampant due to a bug in the system will become a smaller problem as time progresses, but it will persist.