Objective Morality - God or Darwin?

Dee

@YeshuaBought

***** You have the right to your opinion, 

I know Wow ! Thanks for that 

***but you don't have the right to your own facts.

No one “owns facts” some like you deny them thought 

 ****You need to go read Philemon.

No I don’t 

***** I will never agree with you,

I know , yet facts are facts 

**** so don't bother.

I’m actually not , I was just correcting you ....again .......What? Oh you’re welcome 

YeshuaBought

Dee said:

@YeshuaBought

***** You have the right to your opinion, 

I know Wow ! Thanks for that 

***but you don't have the right to your own facts.

No one “owns facts” some like you deny them thought 

 ****You need to go read Philemon.

No I don’t 

***** I will never agree with you,

I know , yet facts are facts 

**** so don't bother.

I’m actually not , I was just correcting you ....again .......What? Oh you’re welcome 

Leave me alone.

Dee

@YeshuaBought

Leave me alone.

YeshuaBought

Dee said:

@YeshuaBought

Leave me alone.

Reported for harassment, leave me alone, troll.

GnosticChristian

@YeshuaBought ;

You replied with couth elsewhere so I thought I would engage here.

I still do not know what you view as an objective moral tenet.

Listen to this and you might change the way you define that term.

 

You used the first commandment as an example.

Concider that it is impossible to successfully order a person to love anything that they do not know.

It becomes even harder when the love is to go to what is described to be really vile by any moral standard.

Regards
DL

YeshuaBought

GnosticChristian said:

@YeshuaBought ;

You replied with couth elsewhere so I thought I would engage here.

I still do not know what you view as an objective moral tenet.

Listen to this and you might change the way you define that term.

 

You used the first commandment as an example.

Concider that it is impossible to successfully order a person to love anything that they do not know.

It becomes even harder when the love is to go to what is described to be really vile by any moral standard.

Regards
DL

I don't care about your feelings, and neither do the facts. if you were a true Christian, you would agree with the Bible. You are a and a false teacher, and i just don't care, if you don't like hearing that.

Dee

@YeshuaBought

I thought you were leaving the site never to come back ? At least you’re predictable as in you change your mind every 5 minutes . A couple of weeks ago you claimed to be Atheist , you also claimed your support for Bernie Sanders now you’re back wtth Trump .......The opposite of what you say always seems to be the truth do you see a pattern here?

GnosticChristian

YeshuaBought said:

I don't care about your feelings, and neither do the facts. if you were a true Christian, you would agree with the Bible. You are a and a false teacher, and i just don't care, if you don't like hearing that.

You have not agreed with the bible yourself and ignored the quotes I gave to refute your view on free will and elsewhere.

Take the log out of your eye.

Regards
DL

MayCaesar

Happy_Killbot said:

@MayCaesar I don't know about your views that an AI would do that, because it is just carrying out it's reward function, and if that function is +1 point for everyone converted to Islam, it's not going to start leading people away from it. It might do a lot of things that we wouldn't consider direct goals of Islam, such as manipulating non believers into wanting to become Muslim, maybe even lobotomizing people so that they could only except Islam, and colonizing other planets to make room for more people.

The problem is, even though the doctrine of any religion is utterly incomprehensible and illogical, the AI doesn't care because it isn't operating based on those principals, it is operating based on it's reward function program, which has to be completely logical or else the program will cease to function. That is why we have all these crazy thought experiments like paperclip maximizers and the stop button problem.

Now there are ways around this that are sort of cutting edge, and may cause things to go a completely different way, if for example it's reward function is also unknown and must be learned itself, because what the creators want isn't really what they think they want. This type of AI is inherently much easier to control, because since it doesn't know what it wants it is reliant on its users for feedback to know it is doing a good job. This is a severely oversimplified explanation of the solution to the stop button problem, and in theory it will stop paper clip maximizers and Jihad drones from taking over the planet. This is because it is trying to do things that will make it's user happy rather than just maliciously carry out the will of its creator.

Paperclip maximizer and other related experiments are interesting to ponder on, but I do not think they describe the real AIs in any way. An actual developed AI that can teach itself will not just be trying to maximise some mathematical function; there will be much more going on there. For example, I am fairly sure that an AI will need to have some sort of a randomizer mechanism, so its response to external stimuli is somewhat unpredictable - I could go in depth on why I think this is the case, and it is not a very trivial explanation, but the crux is its need to look at the data from multiple perspective in order to properly analyse it.

What is fairly likely, however, is that, regardless of the AI programming, it will have a very strong incentive to get as much power over humanity and other AIs as possible. No matter what sophisticated purpose ultimately drives the AI, the more power it has, the better it will be able to fulfil its purpose. This means that even a very benevolent AI will still strive to be a dictator and to subdue humanity.
Which is why I think that soon after we create an AI, we will lose control over our lives. Which might not be a bad thing: depending on the exact scenario, an AI controlling our society may be strongly beneficial to us. Nonetheless, it is a one-way road. In my opinion, we should focus more not on how to make an AI as harmless as possible, but on how to make an AI period.

Happy_Killbot

@MayCaesar What type of algorithm you are working with will decide if it is going to need random components or not. Some neural networks will create a bias at each node and then make a random decision from each node based on that bias, for example path A might have a 10% chance, B 60% and C 30%, and these types are not deterministic and will sometimes do random things that become less random over time.

However you can also do it by passing a value into the next node, where the value passed from one to the next is determined by the output of the last, in this case instead of picking a random path it adds a value for each sequential node based on whatever that learned value is, so A might get 100, B 600, C 300, and then the process repeats for the next set of nodes and values. This type is deterministic and will always output the same value for a given structure.

There are also methods that use a combination of the two.

I'm not convinced we will ever have to deal with an AI overlord, specifically because we are probably going to evolve ourselves at around the same time that an AI would feasibly be capable of doing that. Humans will have created the AI, and presumably will be able to control it to some extent should it start behaving in ways that we don't like. We have clawed and fought our way to the pinnacle of our capabilities, such that producing such AI's is possible.

What is more likely is that humans will use machines for power over other people, and I think that our current path to AI development is not to make fully fledged completely autonomous agents but rather to create dependent semi-autonomous AI's that rely on humans to know what they need to do.

MayCaesar

@Happy_Killbot

What I meant is randomness not just in the output of each neuron (which is an inherent feature of all neural network models I have worked with), but in the input as well. That is, the sensory input the AI receives must have artificial uncertainties in it, and also artificial uncertainties must be introduced into the functions it uses to evaluate the output. 

I cannot easily explain it, but if you read some papers on neural networks, you can get the gust of it. It is somewhat related to the overtraining and overfitting problems which plague all neural network models to this day (and are possibly unsolvable completely), but in a more general sense. In a way, neural network needs to experience more variation in inputs than its actual sensors allow it to process, hence there should be mechanism generating "fake data" and making the AI adjust to it.
Something similar needs to be done with neural networks learning on simulated data, but applied to real data: since real data has some inherent noise, artificial noise needs to be introduced into the simulated data, otherwise overfitting and, sometimes, overtraining problems occur. But an actual AI deals with a much more complex input than modern neural networks, hence it needs more noise introduced at more different points of its decision-making process, and noise is intrinsically random.

I am not sure if an AI will become our overlord, but it will have a strong incentive to try, at least. Of course, it is also possible that the AI will not be a completely separate creature, and, instead, humans will be somewhat integrated with the same technology as the one the AI is based on - in which case the AI really will be just an extension of us, and it will do our bidding as much as we will do its.