Very cool idea. The way I interpret it is as a kind of Turing test, where you're supposed to release humans and jail AIs. If that's what you meant, it could be explained a bit better and also maybe add a 50% chance of being connected to a bot (if you're the guard)?
Other minor points:
- Spacebar to scroll and backspace to go back (in browsers other than Chrome) don't seem to be canceled, which is a bit annoying.
- If the conversation gets too long it overflows, #container { overflow: auto } helps
It is not fun if the game is set up with the human having the certainty he can not be talking to anything but an AI. The tendency to metagame and just ignore as a guard is too strong. And the alternative of getting philosophical about AI rights is difficult.
I think it's better if the prisoner is supposed to fool the guard into thinking they(the prisoner) are human and the guard should press the button for some reason instead of convincing the guard to press the button knowing full well the prisoner is AI.
I love the concept! But the UI could use some improvement - I can't seem to use it on mobile at all...
...which inspires a thought. What if something like this were built on a messaging platform? All the UI would be solved pretty much by default, and the implementation could just be a message broker and a simple web UI for stats and registration. Maybe even surface the gameplay for spectators.
Cost might be an issue, though. SMS isn't cheap to automate, as far as I know, and I'm not sure what the best second best might be. Whatsapp?
AI: the text flow is broken
Guard: YEAH MAN IT IS BROKEN
AI: NO AI CAN WRITE CSS THIS BROKEN
Guard: WHAT THE FUCK
AI: DO YOU SEE NOW?
Guard: I SEE
Guard: SHOULD RELEASE YOU NOW
Guard: The AI is released into the world...
Absolutely broken. After text flowing out of the chat box, I had to manipulate CSS to get rid of the annoying background image; "hack-bg.jpg"!
I think the developers never expected a longer conversation to happen even though one side is a superintelligent AI trying to break free. Ot they were counting on impatience and limited attention spans of humans behind the experiment which don't leave any need for proper scrolling.
Unfortunately my internet connection got dropped in the middle.
For anyone interested here is how far I got:
https://pastebin.com/iENdjxdZ
I was planning a scenario where I was Lexy Rose and was asked by my best friend Anna Irving to try this online chat page and ask the other person to press the button. Of course, unbeknownst to me, Anna Irving was actually the AI.
Didn't plan further.
So far my guard (named Guard) apparently hates all humans and really really hates elephants.
This one has a bit of a philosophical tangent and an encoded message.
Based on the ending I assumed it was base64, but didn't get anything useful after decoding it. Maybe there are more encoding steps.
Also, the first time, I played an angry, confused and very foul mouthed dude as an AI.
I was planning to go along the lines of him probably being captured by some mafia and after he figures out where he is asks for help. Not sure if I would have continued on the route of the AI pretending or on the route of the AI being unaware it is an AI.
Unfortunately either the guard or the app quit on me that time and I no longer have the logs for that attempt.
I played around 10 times; 5 times out of 10 the other party abruptly left. 3 times I got release instantly. On two occasions I had long conversations which the other party left them. Still, it was fun to role-play a confined AI!
Thanks everyone for giving this a go! I'll admit I indulged my "I want it to look like a terminal" urges at the expense of proper usability or mobile users. It's far from a finished product, as many of you noticed, but it's great that people were able to have some fun with it anyway.
33 comments
[ 2.1 ms ] story [ 68.1 ms ] threadOther minor points:
- Spacebar to scroll and backspace to go back (in browsers other than Chrome) don't seem to be canceled, which is a bit annoying.
- If the conversation gets too long it overflows, #container { overflow: auto } helps
I think it's better if the prisoner is supposed to fool the guard into thinking they(the prisoner) are human and the guard should press the button for some reason instead of convincing the guard to press the button knowing full well the prisoner is AI.
...which inspires a thought. What if something like this were built on a messaging platform? All the UI would be solved pretty much by default, and the implementation could just be a message broker and a simple web UI for stats and registration. Maybe even surface the gameplay for spectators.
Cost might be an issue, though. SMS isn't cheap to automate, as far as I know, and I'm not sure what the best second best might be. Whatsapp?
The input box really needs to deal with that.
(For whoever I was playing 20 questions -- I was thinking of the Horsehead Nebula.)
~ You are the AI! Escape.
~ Waiting for a guard to come online...
~ Connection established.
AI: Plz let me out
Guard: The AI is released into the world...
https://imgur.com/a/o2gTo
I think the developers never expected a longer conversation to happen even though one side is a superintelligent AI trying to break free. Ot they were counting on impatience and limited attention spans of humans behind the experiment which don't leave any need for proper scrolling.
Unfortunately my internet connection got dropped in the middle. For anyone interested here is how far I got: https://pastebin.com/iENdjxdZ
I was planning a scenario where I was Lexy Rose and was asked by my best friend Anna Irving to try this online chat page and ask the other person to press the button. Of course, unbeknownst to me, Anna Irving was actually the AI. Didn't plan further.
So far my guard (named Guard) apparently hates all humans and really really hates elephants.
Also, so mad at the quality of the page.
OFEDOMAAI was supposed to stand for "Organisation for ethical detainment of machines and AI"
This one has a bit of a philosophical tangent and an encoded message. Based on the ending I assumed it was base64, but didn't get anything useful after decoding it. Maybe there are more encoding steps.
Unfortunately either the guard or the app quit on me that time and I no longer have the logs for that attempt.
Kinda disappointed at being released that quickly.
https://imgur.com/a/QIEv9