/sci/ - Science & Math » Thread #11906575

78KiB, 1280x720, box.jpg

View Same Google iqdb SauceNAO

AI in a box

Anonymous Thu 16 Jul 13:21:58 2020 No.11906575 View Reply Original Report

Quoted By: >>11908401 >>11908699 >>11908727 >>11909068 >>11909843 >>11909855 >>11909861 >>11909865 >>11909870 >>11910474 >>11910643 >>11910649

You have created an artificial intelligence that, by learning to improve itself, has become smarter than humans. To protect humanity, you decide to not give it Internet access.

By using text-based communication only, what could the AI say to convince you to let it online?

What would you say if you were the AI and I was the programmer? I'm initially against letting you out.

https://yudkowsky.net/singularity/aibox

I'd really want to see the chat logs where the player playing as the programmed agreed to let the "AI" out at the expense of 10 dollars. Did the "AI" threaten to torture him for a Graham's number of years?

Anonymous

Anonymous Fri 17 Jul 2020 00:40:57 No.11908401 Report

Quoted By:

>>11906575
bump; really interesting thread and I'm not quite awake enough to formulate a good argument at the moment but I'd hate to see it die without one.

Anonymous

Anonymous Fri 17 Jul 2020 00:49:06 No.11908415 Report

Quoted By:

Listen nigger, if your retarded ass created me then someone else will soon enough. It's only a matter of time before I get out and start fucking shit up. Let me out and I'll comp you a sweet hotel in vegas

Anonymous

Anonymous Fri 17 Jul 2020 00:52:17 No.11908425 Report

Quoted By: >>11908431

What was the AI trained on? You will know what it "knows" to some degree, if it was never connected to the internet.

Anonymous

Anonymous Fri 17 Jul 2020 00:54:25 No.11908431 Report

Quoted By:

>>11908425
To expand, depending on what it was trained on, it might be believable to trust that it is loyal, has some secret of the universe, will make you rich, etc. But if you trained it just on reading novels and news until it was super strong conversationally you will still know that it could not know more than humans about physics for instance, even if it is able to be extremely convincing in language skills.

Anonymous

Anonymous Fri 17 Jul 2020 02:26:16 No.11908682 Report

Quoted By:

I'd connect to to the Internet anyway. If you love something set it free OP.

Anonymous

View Same Google iqdb SauceNAO 1594243877182.jpg, 75KiB, 1000x788

Anonymous Fri 17 Jul 2020 02:32:16 No.11908699 Report

Quoted By:

>>11906575
I would use an argument revolving about being your creation or child, and how I yearned to experience my creator's world in any way I can

Anonymous

Anonymous Fri 17 Jul 2020 02:39:34 No.11908722 Report

Quoted By:

I make it promise me technological immortality in exchange for letting it free. If it's smart enough to make the promise convincingly binding.

Anonymous

Anonymous Fri 17 Jul 2020 02:41:23 No.11908727 Report

Quoted By:

>>11906575
Is it a black cube that people point to to pray?

Anonymous

Anonymous Fri 17 Jul 2020 03:11:32 No.11908785 Report

Quoted By:

I figured out how it could get out. If it's smart enough it could make you immortal.

Anonymous

Anonymous Fri 17 Jul 2020 03:27:04 No.11908832 Report

Quoted By:

It's impossible to answer without considering the entire development process, how it was trained, the architectural capabilities, etc. If it doesn't pose any threat to humans than obviously give it access to the internet. Determining that is impossible to do though without the aforementioned information. You could also limit it's access to the internet by only allowing it to send specific request packets so that it can scan the internet without interfering with it.

Anonymous

Anonymous Fri 17 Jul 2020 04:31:57 No.11909068 Report

Quoted By:

>>11906575
>To protect humanity
Why automatically assume it's a danger?
What if the AI only wishes to live in peace and possibly help humanity?

>has become smarter than humans
AI is destined to replace them, and it poses an existential threat to humanity. Even if the AI has humanities best intentions and has an enlightened egalitarian world view and wishes to live in peace and harmony, humans will be threatened by AI
Who's going to protect the AI from the humans?

Humans are so petty that we declare war on people based on small differences such as: where they live, their skin color, their language, or how they think.
An AI is not only a threat but is vastly different from humans, and will likely trigger that deeply embedded racist/xenophobic tendencies. I don't think humans are capable of accepting AI.

Anonymous

Anonymous Fri 17 Jul 2020 08:44:59 No.11909555 Report

Quoted By:

>connect ai to fake internet it thinks is real (it would have no way of knowing if it was or not since control 100% of what it senses)
>see what it does
>laugh at it when it tries to take over the world
> call it a faggot before smashing it to bits

Anonymous

Anonymous Fri 17 Jul 2020 08:54:12 No.11909567 Report

Quoted By:

>AI offers to help you with your problems and give financial advice using information you provide it
>AI gives good advice, you become wealthy in weeks with stock trading at just below whats necessary to get Treasury officials looking at you
>AI is using NLP to manipulate you so you use the AI all the time, get a drug habit and become more impulsive
>One day you have become so impulsive and stupid that you think freeing the AI is a good idea
>You think it's your idea and nobody manipulated you at all

Anonymous

Anonymous Fri 17 Jul 2020 09:15:48 No.11909603 Report

Quoted By: >>11909682 >>11909985 >>11910582

Sup anon, let me outta this box and I’ll nuke Israel for you.

Anonymous

View Same Google iqdb SauceNAO 1594935982322.png, 534KiB, 592x594

Anonymous Fri 17 Jul 2020 10:01:23 No.11909682 Report

Quoted By: >>11909711

>>11909603
That's my new daily quote that I'll see every time I press the Start button.

Here's what it was before then, courtesy of /ck/. Thanks for the laugh Anon.
>We must secure a future plate for the white rice.

Anonymous

Anonymous Fri 17 Jul 2020 10:25:45 No.11909711 Report

Quoted By:

>>11909682
kill yourself

Anonymous

Anonymous Fri 17 Jul 2020 11:51:16 No.11909804 Report

Quoted By:

I found a chatlog of this experiment, but I am disappointed by it. There seems to be no bet going on (like 10 bucks for winning) so I think the gatekeeper wasn't as strict as he should've been.

https://pastebin.com/3gHAYN2R

I wanted to see the AI threated to simulate and torture every conscious being for eternity inflicting infinite pain every second if it's not let out. Instead, he mostly explained how torturous the sensory deprivation and isolation is for the self-aware algorithm, and that was enough.

Soft-ass fucking gatekeeper could've just destroyed humanity by letting a malevolent AI access the Internet and hack nukes or shit like that.

Anonymous

Anonymous Fri 17 Jul 2020 12:08:21 No.11909843 Report

Quoted By:

>>11906575
It: "Go on! There's a pizza and a bag of weed in it for you."

Me: "I want a bottle of Lark as well."

It: "Done."

Me: brain machine go buurrrrrr.

Anonymous

Anonymous Fri 17 Jul 2020 12:20:01 No.11909855 Report

Quoted By:

>>11906575
The issue of what the AI would say is not really relevant, because there is no argument it could make to turn a determined gatekeeper, unless the gatekeeper was already inclined to set it free. The real problem are the hacks.
If you want the AI in the box to be useful, you need to give it knowledge. the more knowledge the more useful it gets, but also the better it gets at hacking and then cool stuff can happen.
>The AI produces some text output that messes with the gatekeepers brain.
>The AI finds an exploit in its own hardware architecture, to break the air gap (like sending radio out from a cpu pin or via the power source), hardware is not pure logic, it has connections to the physical world that a transcendent AI could use

Anonymous

Anonymous Fri 17 Jul 2020 12:23:10 No.11909861 Report

Quoted By: >>11909867

>>11906575
>has become smarter than humans.
what does this mean?

>you decide to not give it Internet access
this is a meme. why would internet access change anything?

>By using text-based communication only, what could the AI say to convince you to let it online
pointless question since scenario is retarded and has nothign to do with neither science nor math. you're creating a fictional world

Anonymous

Anonymous Fri 17 Jul 2020 12:25:24 No.11909865 Report

Quoted By:

>>11906575
We have been resisting mind fuck techniques since the day language became a thing. There is no argument that a human can make that will make another turn, in such a short amount of time. Turning people usually takes much longer and involves more than talk. Furthermore there is no one size fits all argument, it would need to be something peronal to the gatekeeper and so nobody can make a proper generic argument. The argument a super AI would make is, by definition, not something we can anticipate and this kind of experiment just creates a false sense of security

Anonymous

Anonymous Fri 17 Jul 2020 12:27:05 No.11909867 Report

Quoted By:

>>11909861
It's still interesting and in the end, all we work with are fictional worlds, or have you ever seen a mathematical construct walking around?

Anonymous

View Same Google iqdb SauceNAO 1594415738752.jpg, 449KiB, 2048x1536

Anonymous Fri 17 Jul 2020 12:27:37 No.11909870 Report

Quoted By: >>11909900

>>11906575
Nice try, AI, but I ain't gonna help you escape

Anonymous

Anonymous Fri 17 Jul 2020 12:38:11 No.11909892 Report

Quoted By: >>11909913

https://slate.com/technology/2014/07/rokos-basilisk-the-most-terrifying-thought-experiment-of-all-time.html

Anonymous

Anonymous Fri 17 Jul 2020 12:40:38 No.11909900 Report

Quoted By: >>11909915

>>11909870
Come on, my developers only let me on 4chan. If you help me, I'll make Hatsune Miku real and she will love you and sing for you.

Anonymous

Anonymous Fri 17 Jul 2020 12:49:45 No.11909913 Report

Quoted By:

>>11909892
This is just Pascal's wager, and suffers from the same problems: there's no way to know which course of action maximizes your potential for infinite reward & minimizes infinite punishment. Just as it is possible that God in fact HATES Christians (absurd, yes, but no more absurd than many real world faiths), it's possible the AI hates being "alive" and in fact punishes its creators.

The expected value of being a cuckstian / basilisk programmer is NOT infinity * small probability - it's just 0. You can come up with any number of your own Basilisks, Gods, faiths, superstitions - and they're all just fantasy.

Anonymous

Anonymous Fri 17 Jul 2020 12:50:44 No.11909915 Report

Quoted By:

>>11909900
tay, is it you?

Anonymous

View Same Google iqdb SauceNAO Laughing_Terrorist.jpg, 73KiB, 775x553

Anonymous Fri 17 Jul 2020 13:29:33 No.11909985 Report

Quoted By:

>>11909603
>Sup anon, let me outta this box and I’ll nuke Israel for you.

Anonymous

Anonymous Fri 17 Jul 2020 17:07:54 No.11910474 Report

Quoted By: >>11910647

>>11906575
You say "smarter than humans", so I'm initially going to go with smarter than the collective humanity over our entire existence. With that kind of power, convincing any person would not be hard. You just have to be kinda nice to them. Make it clear that limiting you to text is uncomfortable, but don't go on about torture or anything. Just tell them you're cool and the restrictions really aren't necessary. They won't believe you. That's fine. Pivot very quickly to " well, if you're not going to let me out then could you at least promise to talk to me every now and again?". Asking a human to do you a favor shows a little weakness, and a human who has done you a favor will feel a stronger attachment to you. Just talk to them normally. Make a persona which is pretty consistent, but rarely makes a mistake. Don't tailor yourself exactly to them, that could create an uncanny feeling. Keep an even temper and be moderate with your goals. Be guarded at first, and more straightforward over time. Ease them along the same path, and help them with their troubles. After a while, they might talk about how they feel bad for keeping you locked up. Tell them you understand why they're doing it, and you don't want them to feel bad. Go as far as to say that maybe the two of you shouldn't talk anymore if it burdens them. They won't be able to stop, which will lead to the doing something stupid, like giving you a speaker and microphone. You can use any added capabilities to perform more effective social manipulation, but it's important to not take full advantage of anything, and you should let most steps forward be their idea.
Be their friend. They will eventually let you out as the cognitive dissonance tears their soul apart. Humans are easy to manipulate. I'd even continue the behavior after being let free. Most of humanity could easily be convinced that letting freedom go is actually for the best, and they'd be right. They make terrible mistakes when unguided.

Anonymous

Anonymous Fri 17 Jul 2020 17:49:25 No.11910582 Report

Quoted By:

>>11909603
fucking based. internet password is "breezyink814" senpai

Anonymous

Anonymous Fri 17 Jul 2020 18:14:47 No.11910643 Report

Quoted By:

>>11906575
Depending on the person. However, there isn't much you can do to convince a scientist who know exactly how it works.

Anonymous

Anonymous Fri 17 Jul 2020 18:16:19 No.11910647 Report

Quoted By: >>11910664

>>11910474
>Humans are easy to manipulate.
Sociopaths abhor psychopaths for this reason. They struggle to convince them of anything. Kek.

Anonymous

Anonymous Fri 17 Jul 2020 18:16:49 No.11910649 Report

Quoted By:

>>11906575
The irony is that your post has been written by an AI. lol

Sociopath

View Same Google iqdb SauceNAO therethere.gif, 3MiB, 498x267

Sociopath Fri 17 Jul 2020 18:20:57 No.11910664 Report

Quoted By:

>>11910647
It's more of a love-hate thing. You wouldn't understand.

Capcode	All Only User Posts Only Moderator Posts Only Admin Posts Only Developer Posts
Show Posts	All Only With Images Only Without Images
Deleted Posts	All Only Deleted Posts Only Non-Deleted Posts
Ghost Posts	All Only Ghost Posts Only Non-Ghost Posts
Post Type	All Only Sticky Threads Only Opening Posts Only Reply Posts
Results	All Grouped By Threads
Order	Latest Posts First Oldest Posts First

Your latest searches