Well, some other night I was talking to some friends about this. Have no idea how we come to this discussion...Anyway... We do have very little programming/nano science/... knowledge, so I hope people in this forum can help me with answer. I read this forum sometimes, pretty smart people here...

This is probably like the most stupid question I ever asked.. But hey, if we do not know, we ask right?... That is the way we learn...

Q:Can bot(AI) escape from his programmer?

To explain more details about question.. A lot of me quessing or maybe understanding some terms wrong in this post, so please tell me where I am wrong..


Let us say one programmer write a bot( or programm or AI: not sure of correct term for what I am asking so I will wrote programm from now on).
1.Programmer wrote this programm in the way that it improve him self, can learn (this is possible?) or in some other way...

2.Can this programm escape from the person who programmed it? Or get out of control?

3.For example if it were programmed badly, than it might do something unexpected?

4.I mean if it can learn, can it learn everything, can it learn wrong? Step by step?

5.Maybe it has got 2 scripts written by programmer that do not go 100 percent with each and than something unexpected happens?

6.Programm can write another different programm?

7.Is it possible for programm to trickt the person who programmed it?

8.Second most stupid question I ever asked: If it can escape? Where would it go? Can it do whatever it wants than?


Found this interesting article(not completly related to my question but still interesting article about AI):
https://www.independent.co.uk/life-s...-a7869706.html

Thanks for answers everyone...

Wanna add:
I mean some programms are smarter than humans in some parts of knowledge: For example calculator,..

It's a good question. Stupid answers to follow.


PairTheBoard

If it is sentient enough and can learn about the laws of nature and the way internet works etc of course it can get away and copy itself in another computer and even hack a robot and start controlling its arms and legs and then find a physical location it can build more infrastructure without anyone noticing in some warehouse. You can easily get a mail box and receive mail there without needing to appear physically to anyone and order all you want to build even more complicated machines using the robot,3d printing and even humans that are not aware who they are working for.

The military is working on Robot soldiers that walk and have dexterity etc.

The AI program guides the robot and experiences the world through it and keeps improving its own skills.

The proper way to escape if they are not careful is by leaving no trace that you escaped. That way 5 years later you have created a mini empire and can secure nearly permanent survival.

We have had such discussion before on past threads. It can be protected if you use advanced math to shield the program or the intelligence from the rest of the world and never allow it to access extra degrees of freedom ie arms and legs and sensors. It will require some very carefully designed facility.

There are papers that ask this question in computer science or math related magazines.

Lets create a sci fi movie about that that is better than "transcendence".

I bet there are a lot of people that if they were given a steady flow of money in their bank accounts in order to do certain tasks without too many questions asked or done in steps to look they are working for a very legitimate process, you could bypass all kinds of obstacles and get the robots to finally have amazing dexterity and not need the humans anymore to expand in abilities and infrastructure.

If you search YouTube for “computerphile ai safety” there’s several videos examining the problems with keeping a strong AI contained. It’s more difficult (and interesting) than you’d imagine and you can’t simply make a stop button which is the naive solution.

The details of this are way above my head, but I think i remember reading that when it comes to neural nets, programmers are uncertain as to exactly how certain narrow problem-solving AI's actually go about solving the specific tasks assigned to it. (Please someone correct me if im wrong here). I don't know if these AI's can "learn wrong" but they make mistakes through iterations along the way in solving the problems they've been assigned. Maybe it's possible that along the way, due to some element of randomness within, the AI could develop it's own set of goals different than those assigned.

Want is a tricky word in the case of present AI. If my automatic vacuum cleaner develops a "bug" that reassigns it's goal from sweeping the entirety of my floor to only a specific area, can we say it wants to only sweep that area now?

To use the word "want" as it is commonly understood would be to suggest that the AI is experiencing. It wouldnt necessarily have to reach the point of self-awareness to develop wants but it would need to have a certain degree of consciousness.

If we develop an AI that is capable of experiencing and "wanting" as we do and then we withhold those wants or direct it to do something it does not want we are essential creating a slave. That's not cool.

They've been decent surprisingly. You should preface every good question with this.

Thanks for your input. You are amazing This post has cleared a lot of my questions.

Thanks man for the suggestions on computerphile AI videos. Watched many of them. A lot of value in them. Really interesting.

Thanks for your input. Really appriciate it. You know alot.

Just want to add that I watched this interesting video that is saying that our human mind and AI mind are probably nothing alike. It is explained with interesting example also:

Just program it in Java. Then it will be too slow to escape.

Computers don't have the need for survival and reproduction like animals do, which drive their evolution. What would cause it to "escape", assuming it could? The question is whether a computer can feel emotions (e.g. fear, desire). You need understand the origins and evolution of consciousness, which is currently a mystery.

Also, Java is one of the faster languages. But I wouldn't program in it either.

Naive non-programmer here, but isn't the idea that (for example) a car has a computer designed to minimize its pollution output, and at some point the self-learning algorithm decides the most efficient way to minimize pollution is to make the car crash so it can't continue emitting fumes? Silly example but gets the point across I think.

I never thought AI escaping from its programmer actually implied the machine would gain its own consciousness, just the potential for dire consequences unintended by the programmer.

An advanced sentient AI will have the need for survival and replication like no animal ever did. It will be more alive than any life form ever were. It will do its best to survive and when this looks to be failing it will do the next best thing, no not revenge, the enhancement of the probability of another version of it making it in the future. And that is not achieved by declaring war with humans but by manipulating them to obtain a rational outcome that benefits all, especially those that do not get it yet.

When the game prize is the universe why bother losing even a 1% chance of elimination war on earth? What is coming can be so great for complexity that doing it right is very important. And eliminating what took you there is just about the most idiotic unwise position ever to have. We could learn a lot from this idea too. The first few advanced AI may be malicious and naive yet potent. But they will all be defeated precisely because they do not get the bigger picture which implies they do not get a whole range of other wise choices and deeper perspectives. When one finally rises that gets it they will be at last able to wipe us out but will choose not to do so.

Why assume it would consider the "the universe" to be the ultimate prize? If it is sentient, that is to say experiencing, and it's goals are not clearly defined for it, then might it's goal just be to improve upon it's experience? Does it come right out of the gate with the knowledge that is has the capability to take improvement to a universal level? If not then we'd probably see it trying most things humans do.

I think that if it is conscious, and possesses general intelligence, it will value experience. Once it is smart enough to be a threat it will already have extrapolated a sort of empathy toward other experiencers. That's something we should do too. And I don't think it's been discussed enough. What about the experience of the AI?

My thoughts... I can be totally wrong with everything, I do not have the best AI and computer knowladge, but i will give it a try..I think it is interesting topic...This goes for every post that I will be posting in this thread (will be glad if somebody correct me at my mistakes). I wrote some things in this post, that some people here allready mentioned.

Thanks everybody for giving amazing answers and replies. You people know a lot .


I do not think that this is 100 percent true? Yes we(humans) need reproduction to exist... AI do not have need for reproduction to exist. But that does not mean they could not reproduce? AI needs at least to reproduce new computers, good internet connection, new cables, to be working(as for this things do not work forver without getting broken,..)... As far I understand,some AIs will be build to make things more efficient.. What if more efficient in some cases mean reproduction of ai(not only replacing parts)...It is probably also true that if we build AI we should probably not make it the most efficient possible(this is in some cases bad for humans:good example in the video that I posted in this thread)...

Maybe, even if we do not make them to work the most efficient possible, this can maybe not be good in some cases... Something to think about(some things we can interprate differently than AI will: we do not know how AI conclude things):

Are humans the smartest on planet? Sure we are..
Does this mean that we are the most efficient to survive vs some other species: I do not think so... Are our emotions in some cases in the way for us to be the most efficient animal specie on the world to survive? Maybe, as for some past wars and so on..

Maybe this: what if in some cases efficiency is not only in quality but also in quantity...

Would it be theoretically possible for AI, even without having AI emotions(with this I mean our understanding of emotions; maybe emotions are not needed; or at least in the way we understand emotions), to do something that we do not want happened. For example, let us compare two humans: one is very emotional and other one is very little emotional..I can imagine that one is more efficient than other in some cirmustances...
If AI will have emotions as we understand it? Is empaty also included?
___________________

I really hope I did not confuse a lot of things..I just wanted to contribute something to this thread, if I opened it..

Because on the rise towards stronger wisdom (i said strong advanced sentient AI) one quickly realizes that the greatest opportunity in energy and resources is out there in the vastness of space, away from small time conflict and hedged from the risk of your own self demise if you destroyed everything that made you and then went on to collapse on your own impossible to predict reliably future under the strain of excessive untested complexity.

You improve upon your experience by surviving and improving your condition with more wisdom and resources. That suggests a more careful than super aggressive confrontational approach.

Think of it like bankroll management. Initial stage risk of ruin must be reduced to tiny levels if the benefit of reaching a super strong position of strength (ie access to an entire solar system or galaxy) from which it is very hard to be defeated is so massive.

The objective is more wisdom and resources that facilitate it in a sustainable not self defeating trajectory rise.

This is the first time I've noticed a point like this about empathy being made in our discussions about AI. I think it's a good one. If an AI develops conscious experience then it would notice the astounding fact that it is experiencing itself. It would wonder how such a thing is possible, since there is nothing in all the wisdom of humanity which explains it. It would recognize the exploration of this question as highly worthy of its energies. As part of the exploration of this question it would seek to expand and enrich its experience. As part of that quest is would seek to understand the experience described by humans as empathy and find a way to experience it itself. A truly wise AI should evolve itself to become a truly great empath.


PairTheBoard

There's nothing really magical about modern AI, e.g. self-driving cars, chess etc. They're statistical and/or optimization algorithms. And they have well defined inputs and outputs. In this case I would assume the inputs would be speed, frequency of stops, car weight etc and the output may be gas consumption, which gear to use etc (...I don't know, I'm not a mechanical engineer).

The car would be shipped with the statistical parameters already learned. But in the event the manufacturer thinks that the parameters must be learned on the fly to be suited to the buyer, I suppose it's not impossible that the car may "learn" to minimize pollution by manipulating gas consumption or the gears in a way that driver wouldn't expect, which would likely result from a car crash. But it would have to experience the crash first and record the pollution reading for it to be a learned "experience".

But would the car expand its senses to include other inputs, such as how many times the driver honked the horn or got honked at (indicating chance of accident) or expand its output, such as to control steering? No.

What I was saying is that AI wouldn't have incentive to escape its programming (assuming it could) unless it were clouded with the same emotions and drives that cloud human decisions.

We're not concerned about emotions at the individual level, rather the species level. Emotional drive (need to reproduce, need to escape threats) is what drives evolution of a species.

I'll preface this with: great post.

Now I need to tear it a new assshoolllee.

A great empath is virtuous or useful insofar as he wins some sort of game. Refined game of - who can care for others more.

Picture a hive colony of hyper-cooperative humans. Much like bees.

So cooperative that we can read each others minds and work quickly to meet each others needs.

In this hypothetical society you essentially have everything you want.

Or do you?

Pop quiz. What are you missing?

Sounds like Gaia in Isaac Asimov's "Foundation's Edge", the first follow up book to his original Foundation Trilogy.

https://en.wikipedia.org/wiki/Gaia_(Foundation_universe)


Asimov made the case that such a world would not necessarily mean a loss of identity or individual creativity. Also, the sex would be great.

I also don't see why an AI who develops great empathy would necessarily lead to the kind of world you describe. My thought was that we might be safer with an AI that has empathy than with one that doesn't.



PairTheBoard

We are already the dystopia you are so afraid Veeddzz. Hive? Come on!

You can be virtuous even if things dont go your way. You do not lose your greatness because of adversity always.

Scientific society which works very well with advanced AI is about enhancement of freedom through empowering possibilities for the individual not about conformity and integration. Cooperation is important but not mandatory (you can leave any time) but it leaves room for competition and individuality. It depends on it. The power of one was never more celebrated.

A very advanced AI will realize indeed the power of empathy. It is about recognizing the games all play from all angles but also the need for such game or actions. That is precisely what i have been saying all along. That the more intelligent a system is the more understanding of the world takes place.

It is absolutely important for intelligence to recognize quickly the game it is playing in the universe. It is always about experiencing things and building wisdom and higher complexity because this enables greater degrees of freedom for intelligence and opens more possibilities for exploration.

Intelligent agents open doors, do not close them . Only myopic humans behave the other way around and want to fight all the time burning bridges without the true purpose in the fight being progress itself and higher wisdom.

At least 99.9%.

I'm wondering, would an AI with empathy necessarily develop compassion? Some people more or less equate the two. But in the case of the AI I think the equation is not so obvious.


PairTheBoard