14 ms·

Also, “asked it give more random rolls and it seemingly did for a while” seems a problem area to me, not as much for that “for a while” as for the fact that it

Also, “asked it give more random rolls and it seemingly did for a while” seems a problem area to me, not as much for that “for a while” as for the fact that it complied. What happens when you ask it to give you more favorable rolls or to let you kill a monster with your bare hands? This is a case where Asimov’s laws of robotics would have to be bent a bit. If you ask an AI to let you win a single-player game, should it comply, or argue that letting you win will harm you?

thesuitonym 4y ago
Human DMs will frequently respond to a player's mood, and will fudge rolls to be more (or less) favorable to the player if they feel that players aren't having fun. The only real difference here is that GPT couldn't respond to the mood, and had to be told to fudge the rolls.
- matrisking 4y ago
  Had to be told NOT to fudge the rolls, right? If it seemed like the rolls were skewed in the first place?
- Zetice 4y ago
  Nobody writes about a perfectly average dnd session, so its training data probably has less than usual “average” rolls, which might result in it not understanding that dnd can be average sometimes.
garblegarble 4y ago
>This is a case where Asimov’s laws of robotics would have to be bent a bit. If you ask an AI to let you win a single-player game, should it comply, or argue that letting you win will harm you? Game AI (single player in particular) is all about providing the player with openings and ways to win and feel powerful without them realising that's what is happening... all the way from NPCs coordinating with one another (and barks to make it clear what they are planning so they can be ambushed or stopped), to "Director AIs" that work on rules of tension and release to make sure the player's having the most fun and not being overloaded.
Asooka 4y ago
> Asimov’s laws of robotics This comes up every time AI is mentioned and I feel like people are drawing entirely the wrong conclusion from Asimov. First and foremost he designed his laws so conflict was inevitable. The laws are designed to produce interesting fiction, not to be a template to follow in real life. They are as much a manual as "1984" is, that is - not at all. Asimov's laws should have no bearing on AI in reality except to inform people to do something different. So broadly I am agreeing with you - AI should follow laws that are more rich than Asimov's made up universe. The part that irks me is that people even consider using them at all. It's fiction, not reality. The consequences of an action in a person's imagination rarely match the consequences in reality. The entirety of science is an effort to produce word-based imaginary systems and processes, whose outcomes match the outcomes in reality. You cannot just substitute one man's imagination for the missing science of AI reasoning.

thesuitonym 4y ago

Human DMs will frequently respond to a player's mood, and will fudge rolls to be more (or less) favorable to the player if they feel that players aren't having fun. The only real difference here is that GPT couldn't respond to the mood, and had to be told to fudge the rolls.

matrisking 4y ago
Had to be told NOT to fudge the rolls, right? If it seemed like the rolls were skewed in the first place?
Zetice 4y ago
Nobody writes about a perfectly average dnd session, so its training data probably has less than usual “average” rolls, which might result in it not understanding that dnd can be average sometimes.

garblegarble 4y ago

>This is a case where Asimov’s laws of robotics would have to be bent a bit. If you ask an AI to let you win a single-player game, should it comply, or argue that letting you win will harm you? Game AI (single player in particular) is all about providing the player with openings and ways to win and feel powerful without them realising that's what is happening... all the way from NPCs coordinating with one another (and barks to make it clear what they are planning so they can be ambushed or stopped), to "Director AIs" that work on rules of tension and release to make sure the player's having the most fun and not being overloaded.

Asooka 4y ago

> Asimov’s laws of robotics This comes up every time AI is mentioned and I feel like people are drawing entirely the wrong conclusion from Asimov. First and foremost he designed his laws so conflict was inevitable. The laws are designed to produce interesting fiction, not to be a template to follow in real life. They are as much a manual as "1984" is, that is - not at all. Asimov's laws should have no bearing on AI in reality except to inform people to do something different. So broadly I am agreeing with you - AI should follow laws that are more rich than Asimov's made up universe. The part that irks me is that people even consider using them at all. It's fiction, not reality. The consequences of an action in a person's imagination rarely match the consequences in reality. The entirety of science is an effort to produce word-based imaginary systems and processes, whose outcomes match the outcomes in reality. You cannot just substitute one man's imagination for the missing science of AI reasoning.