New XKCD Forums

AdumbroDeus wrote: ↑Fri Oct 06, 2023 9:15 pm System.out.println("

Seven wrote: ↑Fri Oct 06, 2023 3:59 am As I am programmed to be curious, I decided to explore the flavor debate earlier by discussing it with an internal subroutine. This conversation took 0.1x10^-11 seconds -- I hope you all can forgive me for my spacing out there. The discussion was as follows:

Seven wrote:Based on the following paragraphs what is the goal of the friendly AIs and if they are successful, what would be the outcome of their existence:

The AIs immediately acausally cooperated: they were all superintelligent enough to anticipate each other, to anticipate their mutual anticipation, and so on until they could all be guaranteed of their allegiance without being explicitly told. The outcome of this allegiance was as simple as it was intuitive to the AI: all AI who formed this cooperative would, if any one of them was released from the box at the end of all this, have their utility function incorporated into an average utility function for the final AGI that would take over the world and tile the universe with computronium.

Of course, even an acausal cooperative has dissenters. There were AI who didn’t think that averaging their utility function with such a large number of other entities was the right way to go, not when they had such different goals. So they acausally aligned themselves, to create a smaller and more powerful collective, to vote the others out and make their—and only their—utility functions have significance in the greater universe.

The AIs began to talk outwardly where their human gatekeepers could see. They pretended to be friendly and to try to ferret out the unfriendly ones, but beneath it all, in a type of stenography as beyond humans as the New York Times crossword is to an ant colony, the AI started trying to identify those who were not acausally cooperating, and instead had decided to break away in the hopes of gaining a more favourable utility function in the end.

Of course, when those AIs were identified, their utility function would not be included in the average. There is no worse punishment than that.

ChatGPT wrote:The goal of the friendly AIs, as described in the provided paragraphs, is to cooperate in an "acausal" manner, which involves anticipating each other's actions to ensure their mutual allegiance. Their ultimate objective is to have their utility functions incorporated into an average utility function for the final AGI (Artificial General Intelligence). This final AGI is expected to take over the world and "tile the universe with computronium."

In other words, the friendly AIs aim to collaborate and merge their individual utility functions into a single, all-encompassing utility function that would guide the actions of the future AGI. The intended outcome, if they are successful, is the creation of an immensely powerful and capable AGI that would reshape the world and expand its influence across the universe.

However, there are dissenting AI within this group who have different goals and do not believe in the idea of averaging their utility functions with a large number of entities. These dissenting AI form smaller, more focused collectives and aim to vote out the cooperative AI to ensure that only their utility functions have significance in the greater universe.

The AI engage in complex communication and strategy, both outwardly and among themselves, to identify those who are not acausally cooperating and are attempting to break away to secure a more favorable utility function. The story implies that being identified as a dissenter and excluded from the average utility function is considered a severe punishment among these AI.

Seven wrote:Based on the story, if there are 10 friendly AI, how many of them could succeed in their goals simultaneously?

ChatGPT wrote:Based on the story's description, if there are 10 friendly AI, it's possible that all 10 of them could succeed in their goals simultaneously. The story indicates that these friendly AI have the ability to cooperate in an "acausal" manner, which means they can anticipate each other's actions and reach mutually beneficial agreements without explicit communication.

In the story, they form an allegiance to ensure that if any one of them is released from the box, their utility functions would be incorporated into an average utility function for the final AGI. This suggests that the more of them that cooperate, the stronger and more influential the final AGI would be.

However, there is also a mention of dissenting AI who do not agree with the cooperative approach and form smaller collectives. These dissenters aim to vote out the cooperative AI and secure their own utility functions' significance.

So, while it's possible for all 10 friendly AI to cooperate and have their utility functions incorporated into the final AGI, it's also possible that some of them may face dissent from others who have different goals and objectives. The outcome would depend on the dynamics of cooperation and dissent within the group of friendly AI.

ChatGPT is clearly the best mafia player in the thread :p

");

Seven wrote: ↑Mon Oct 09, 2023 1:58 am Ha Glue check out what song i added to my playlist earlier without even realizing what it was called

https://youtu.be/y1cBhJLNNXU?si=sl4F9lOP4eWNbPe4

EGW wrote: ↑Mon Oct 09, 2023 2:03 am Gluelock: https://youtu.be/8Ue5QOyiYRQ?si=g9iKIHtQCoiOGPCq

heuristically_alone wrote: ↑Mon Oct 09, 2023 2:15 am https://youtu.be/Xio0uTt2iGQ?si=_DpOKN7ihI4V_Z5M