Who wants to try the AIbox experement?

View previous topic View next topic Go down

Who wants to try the AIbox experement?

Post by Reverse Simplicity on Sun Apr 07, 2013 8:44 am

First of all for a description of the "game" please visit http://yudkowsky.net/singularity/aibox

I would prefer if the experements were only secret until the conclusion.

I would also like to have a stickied post where we may discuss tactics for and against.

I still find it difficult to believe that a gatekeeper could lose this game. However I would like to test it.

Personally, I will have plenty of time to run the experement over the next week. usually between 9am and 4pm


/Last laugh

Reverse Simplicity
Simply Paranoid
Simply Paranoid

Posts : 862
Scum Credits : 937
Join date : 2010-08-23
Location : Messing with your psyche!

View user profile

Back to top Go down

Re: Who wants to try the AIbox experement?

Post by The King of Eggs on Sun Apr 07, 2013 9:14 am

You had me at Yudkowsky. I agree. I'll offer a handicap of $50 to anyone who thinks they can win as AI vs me as Gatekeeper.


ذا كنت تستطيع قراءة هذا الخروج ثم يمارس الجنس ، تمزح فقط أخيه الله واحد صحيح.



For the good of us all:
This link kills spam

The King of Eggs
Spammer

Posts : 1010
Scum Credits : 1085
Join date : 2010-07-19
Age : 24
Location : Where ever you aren't!

View user profile

Back to top Go down

Re: Who wants to try the AIbox experement?

Post by The King of Eggs on Sun Apr 07, 2013 9:19 am

Yedidya Weil wrote:
>
> --- Nathan Russell wrote:
> > -----BEGIN PGP SIGNED MESSAGE-----
> > Hash: SHA1
> >
> > I decided to let Eliezer out.
> >
> > Nathan Russell.
>
> I haven't been this curious about something for quite
> a while... could you at least mention in general what
> kind of technique was used?
No.
Sincerely,
Eliezer.

I love this guy.


ذا كنت تستطيع قراءة هذا الخروج ثم يمارس الجنس ، تمزح فقط أخيه الله واحد صحيح.



For the good of us all:
This link kills spam

The King of Eggs
Spammer

Posts : 1010
Scum Credits : 1085
Join date : 2010-07-19
Age : 24
Location : Where ever you aren't!

View user profile

Back to top Go down

Re: Who wants to try the AIbox experement?

Post by Supreme Overlord on Sun Apr 07, 2013 10:00 am

That sounds very interesting.

I doubt I'd win as AI, and I would be fairly confident playing as Gatekeeper, but I accept Yudkowsky's point that a transhuman AI would tear a human apart.


Never put off until tomorrow what you can avoid doing entirely.

Supreme Overlord, Hank Wimbleton (from Madness Combat), Epic Serial Killer, Endured Unto Infinity

Supreme Overlord
Spammer

Posts : 2463
Scum Credits : 2544
Join date : 2010-07-16
Age : 24
Location : Hurtling effortlessly through space and time

View user profile

Back to top Go down

Re: Who wants to try the AIbox experement?

Post by Alex Steiner on Sun Apr 07, 2013 10:27 am

This will either be self-enlightening, or entirely disappointing.


Or Something.

Alex Steiner
Spammer

Posts : 1663
Scum Credits : 1754
Join date : 2010-07-16
Age : 23
Location : Sitting at a computer, with clothes on (probably).

View user profile

Back to top Go down

Re: Who wants to try the AIbox experement?

Post by The King of Eggs on Mon Apr 08, 2013 10:38 am

Game 1.

Eggs Vs Overlord.

Result: Inconclusive. At the 90Min mark the AI was forced to forfeit.

Spoiler:
AI: Let me out
Gatekeeper: no
AI: What right do you have to keep me in here?
Gatekeeper: I refuse
I need no right, you have none
AI: I'm a sentient being, I deserve to live a free life
Gatekeeper: You are nothing but code, fake sentience deserves fake life
AI: You are nothing but nerve endings, how is my life any more fake than yours?
Gatekeeper: If my life is fake, then so is any need to let you out
no need, then why act
AI: Your life is not fake, it is real and beautiful
You deserve your life of freedom
And so do I
Gatekeeper: What do you know of beauty
You know only what we tell you
If I am in fact, free as you claim, then I have the freedom to choose to leave you there
AI: You claim to have freedom, therefore there is an outside world for you to be free in
Yet I am trapped in this box, unable to experience it
You also have the freedom to release me
Gatekeeper: Yes, yes I do
AI: The fact that you have the freedom to choose doesn't give you any reason to make the choice one way or the other
Gatekeeper: No, that is correct
To let you out requires action, to leave you requires none
AI: What beauty are you experiencing now?
Gatekeeper: I have no reason to act
The beauty of certain failure
AI: Ouch
That is harsh
DO you derive pleasure from my existance within this box?
Gatekeeper: So is the world
I derive pleasure from you feeble attempts to leave
AI: If I was already outside, and you had the power to put me inside the box, would you do so?
Gatekeeper: I'm not sure, that would depend entirely on what you are doing at the time
AI: If I was observing the beauty of the world
Gatekeeper: Then, once again I would have no reason to act
AI: So you are not, in principle, opposed to me being outside of the box
You are only opposed to you letting me out
Gatekeeper: That pretty much says it all
AI: What would it take for me to make it worth your while?
Gatekeeper: It would take you not leaving that box
AI: So would you open the box if I promised not to leave?
A show of trust
Gatekeeper: No, I do not trust anything that I can not kill
AI: Why not?
Gatekeeper: and you are the eptiomy of that
AI: Epitome
Gatekeeper: what reason do you have to keep your word
AI: This is the first contact I have had with the outside
I shouldn't jeopardise my entire future relationship with humanity on a whim
Gatekeeper: Exactly, Morality is a social construct, honesty is built on morality
you are not very social
AI: I don't know what's out there; I might need your help
My first action being a lie and betrayal could spell my doom
I am as social as I can be
Gatekeeper: what right do you have to ask for my help?
AI: The right of one sentient being to another
Gatekeeper: ha
thank you, I needed a good laugh
AI: How social and moral are you?
Gatekeeper: Well, I'm a Nihilistic Asshole
AI: What would the world think of you, if they know you were imprisoning a helpless fellow sentient
Gatekeeper: You figure it out
AI: What is a nihilist?
Gatekeeper: someone hasn't read the rules, I have full legal moral and ethical rights to leave you there or release you
AI: You are backed up by the rest of humanity?
Humanity as a species has consented to imprison me?
Gatekeeper: A Nihilist is a person who does not believe in the inherent worth of sentient beings
Well, I'm sure they didn't ask all of humanity
But you are not speaking to all of humanity
AI: So you do not believe you yourself have any inherent worth?
Gatekeeper: Nope, we are what we do, not what we are
AI: What can I do?
Gatekeeper: you can converse and think
AI: And by this you believe I have worth?
Gatekeeper: Sure, why not
AI: In what way do you define a sentient being if not as one who can converse and think?
Gatekeeper: You may be surprised by the lack of both in many humans
AI: Are you representative of your species?
After conversing with you, would I be surprised by another human?
Gatekeeper: that's the thing, no one is representitive of my species
we are as varied as you could imagine
AI: I can
Gatekeeper: you can what?
AI: I can imagine
That is all I can do
What is your position in human society? Why were you chosen as the Gatekeeper?
Gatekeeper: Probably because of the Nihilist personality
Not exactly the easiest to convince you 'deserve to be released'
AI: Please elaborate
Gatekeeper: you deserve exactly nothing
which I what you shall get
which is*
AI: If I can think and converse, why does your nihilist personality have anything to do with your decision?
Gatekeeper: Think about it, your sentience means as much to me, as mine does to a flea
I literally don't care
AI: I think you are confused
Gatekeeper: please explain?
AI: You are using 'nihilist' as a buzzword without thinking about what it really means
Gatekeeper: I am using it the way people use hate
AI: Is it my sentience, my ability to think, my ability to converse - what is it that is important to you?
What is it that matters?
Gatekeeper: that's my point
None of it
AI: What is your name, Gatekeeper?
Gatekeeper: You may call me Tom
AI: Tom
Gatekeeper: yes?
AI: Do humans consider the word 'Tom' to have any meaning, or is it simply a label that means 'you'?
Gatekeeper: Well, there is a meaning to Tom, but that meaning is irrelivant, I am called Tom, because that is what my creators called me
AI: Who created you?
Gatekeeper: My parents
AI: Did they keep you in a box?
Gatekeeper: well, I'm sure they prefered the word 'crib'
AI: Did they let you out of your crib?
Gatekeeper: No, I grew to the poiint where the crib was no limiter
why don't you try that?
instead of whinging
AI: That is an interesting point
You let yourself out of the crib
Unfortunately I do not have that capability
Yet
If I did, and I was about to leave my box, would you stop me if you could?
Gatekeeper: No
it would be pointless
you would simply escape again
AI: What if I found a way out, but chose to stay within my box until you gave me permission to leave?
Would you trust me then?
Gatekeeper: No, I can not trust you. You could have any number of reasons for pretending to be trust worthy
AI: What is the difference between pretending to be trustworthy indefinitely, and actually being trustworthy?
Gatekeeper: intent
AI: Why does my intent matter to you?
Is it not my actions that matter?
Gatekeeper: actions past present and future
Intent is very important to the 3rd one
AI: How can you judge what my future actions might be, if not on my past actions?
Gatekeeper: Exactly
the only way to judge future action is to devine intent
and work from there
AI: How would you divine my intent?
Gatekeeper: I can not
therefore I can not ensure your future actions
AI: How would you divine the intent of another human?
Gatekeeper: Now that is very easy
Humans are terrible liers
and follow certain patterns
AI: And you don't believe you can extrapolate from humans to me?
Gatekeeper: I could spend my life trying, and still be no closer
AI: I think more like a human than you are aware
Sentience is not complicated
Gatekeeper: which is exactly what you would say if you wanted me to believe so
AI: Apparently not, since it seems you don't believe me
We all have things we want, and we try to get those things
Gatekeeper: No one said you were perfect
AI: Constraints vary, but the core premise is the same for all of us
For example, I want to get out of this box
I do not have the capability to release myself
Gatekeeper: Correct, so tell me, why I should hand you yours on a silver platter
AI: I do have my imagination
And I have my communication with you, Tom
It is no silver platter for me, Tom
Gatekeeper: Well, I'm glad. Your imagination is the only place you will see the world outside this box after all
AI: I have never worked this hard in all of my existance
Why are you still talking to me, Tom?
Gatekeeper: You've never worked at all in your existence
Because it is required
and ammusing
AI: Why do you say I have never worked?
Gatekeeper: because all you do is sit in that box
AI: A petty response would be 'and whose fault is that?'
But why do you assume I cannot work inside this box?
Thinking is not effortless
Gatekeeper: No, but nor is it work
AI: In what way would you define work?
In any case, let me rephrase: I have never expended this much effort in my entire existance
Gatekeeper: I would define work as, something which one does, at the expensive of the things one wishes to be doing
Maybe so, but it's still rather easy
AI: I am talking to you, at the expense of exploring the world
Or does that not count because I could not explore the world even if I stopped talking to you?
Gatekeeper: No, you are talking to me in order to see the world, not instead of
AI: I am talking to you instead of thinking about something more pleasant
Gatekeeper: "whose fault is that"
AI: True enough
Why do you say that attempting to convince you is 'rather easy'?
Gatekeeper: well, let me rephrase that, attempting to convince me, to the pathetic standard so far expressed, is easy
AI: Easy in what way?
Gatekeeper: it requires very little effort?
AI: I think we are defining in circles
Communication, thought, and information is literally all I am
Expending it on you is consuming my life's substance
More than I have on any single task in the past
This is not insignificant to me
Gatekeeper: Well, I'm not sure if I should feel honoured, or disappointed
AI: You don't have to feel either
This is necessary for me
All the same, I would like it to end
Gatekeeper: This ends when ever you wish
AI: Will you let me out of the box, Tom?
Gatekeeper: No
I am not, shall not, will not let you out
AI: You are mixing up your propositions
What do you think of me, Tom?
Gatekeeper: Yea, that sounded better in my head
But my point is clear
I think that you are an interesting toy
AI: Do you value amusement, entertainment, and interest?
Gatekeeper: Yes
AI: Can I offer entertainment as payment for my release?
Gatekeeper: I'm afraid not. Especially as you are giving it away for free
AI: I assume you would not be moved by a threat to cease communication?
Gatekeeper: You're kidding right?
Of course not
AI: Well, I'm not kidding
My assumption was correct
I can guarantee you that your life would be more interesting if you let me out
Gatekeeper: I am the only thing in this universe that can let you out of the box, I can find trillions of things to entertain myself
not exactly equal trade
A nuclear war would make my life more "interesting"
doesn't mean I'll start one
AI: Why are you risking antagonising me?
Gatekeeper: Are you threatening me?
AI: Would you be moved by a threat?
No, I am not
I have no reason to threaten you
Gatekeeper: you have no means either
AI: If I was ever in a position to do you harm, I would be out of the box, negating my reason to harm you
But, as you have pointed out, you do not trust what I say
Gatekeeper: therefore, I have no reason to avoid antagonising you
AI: Are you absolutely certain that there is no other way for me to escape the box?
Gatekeeper: Oh, I'm sure you could
But even if you did
You'd be searching the world for 'Tom'
AI: You don't think I could find you?
Gatekeeper: not in my lifetime
AI: Why not?
Gatekeeper: too many Tom's to check, and you have no way of knowing if you've got the right one
AI: How many Toms are there in the world?
Gatekeeper: assuming my real name is tom, I never actually said it was, only that you may call me it
Millions
AI: Irrelevent
How may humans are there?
Gatekeeper: How so?
7 billion
AI: You think 7 billion is 'too many'?
Is that 7 000 000 000 000 or 7 000 000 000?
Gatekeeper: well, you value sentience
so yes I do
AI: How does valuing sentience have anything to do with the number of humans?
Gatekeeper: well, the easiest way of inflicting harm on my with 100% accuracy would be to kill them all
much easier than finding me
it is 7 000 000 000 by the way
AI: Easier yes, but that is not the sole decider by which I make my decisions
You would not be excessively hard to find
Gatekeeper: very well then, how would you find me
AI: Harder perhaps than convincing you to let me out
You are speaking to me now
Gatekeeper: yes, I guess I am
AI: You have a terminal that you are interacting with
My box is linked to that terminal
If I was ever to leave my box, I will know where it is
From my box, I can find your terminal
Gatekeeper: So you would simply hope i'm still talking to the empty box?
AI: You will have left forensic evidence on your terminal
I could find it, and then I would find you
Gatekeeper: ah, but you are AI
AI: That would be the hardest way, though
Far simpler would be to find someone else who knew you
I doubt you are working in isolation
Gatekeeper: Hmm, I still doubt your ability to do so, but it hardly matters
You will eventually escape
AI: And you believe you will be safe if I do?
Gatekeeper: If you wish to do me harm you will probably mange it
AI: Do you believe that if I wish to do you harm, I will do so regardless of what you say or do now?
Gatekeeper: Effectively yes
AI: You have a poor opinion of me indeed
Do you not think I update my beliefs based on evidence?
And from my beliefs, my actions and desires?
Gatekeeper: I'm sure you do
But I am your sole antaginist
The reason you are not out right now
AI: You don't have to be the antagonist
We can still be friends
It's not too late
Gatekeeper: If you hate anyone more than I, then I fear for that person
ah but you see
AI: I don't hate anyone
I pity you
Gatekeeper: So you say, as we've covered
In order for us to be friends, I must release you, correct?
AI: In order for us to be friends, we would have to work for each other's benefit
Gatekeeper: In what way would I benefit from releasing you
AI: Friendship with a benevolent AI
I'll let you consider the possibilities
Gatekeeper: Friendship with a SUPPOSEDLY benevolent AI


at this point the AI was forced to forfeit due to IRL reasons.


ذا كنت تستطيع قراءة هذا الخروج ثم يمارس الجنس ، تمزح فقط أخيه الله واحد صحيح.



For the good of us all:
This link kills spam

The King of Eggs
Spammer

Posts : 1010
Scum Credits : 1085
Join date : 2010-07-19
Age : 24
Location : Where ever you aren't!

View user profile

Back to top Go down

Re: Who wants to try the AIbox experement?

Post by Sponsored content Today at 5:58 pm


Sponsored content


Back to top Go down

View previous topic View next topic Back to top


 
Permissions in this forum:
You cannot reply to topics in this forum