This morning I did another 2.5 km run. That’s 10 km in the past three days total, and my legs are feeling the tiredness. I’ll rest Monday and Tuesday before trying to run again on Wednesday.
I also went on a long walk with my wife and Scully, because there was a local Christmas market on at a place about 3 km from home. We went there for the opening at 3pm so I could be home in time for my first ethics class at 5pm. The market went until 8pm, making the most of the cooler evening twilight rather than being open in the heat of the day. It was warm and sunny, although not scorching.
At home I worked on Darths & Droids strips.
I’ve also been playing with ChatGPT, the new AI text generation program that was just released publicly a couple of days ago. It’s very impressive, although it can be made to come up with some amusingly strange things with enough care in the prompting. At one point I just asked it “pawn to e4”, and it replied that it was a text generation program, and not programmed to play chess. So I tried this (ChatGPT’s responses quoted with sidebar):
If I asked a chess-playing program to respond with a suitable move to the opening “pawn to e4”, what move would it be likely to suggest?
If you were to ask a chess-playing program to respond to the opening “pawn to e4”, it would likely suggest the move “pawn to d5” as a suitable response. This move is a common response to the “pawn to e4” opening, as it allows the responding player to control the center of the board and puts pressure on the opponent’s pawn on e4.
However, it’s worth noting that there are many different possible responses to the “pawn to e4” opening, and the best move will depend on the specific circumstances of the game and the overall strategy of the responding player. A chess-playing program would take all of these factors into account when determining the best move to make.
Are you saying that a chess-playing program would take into account the nature of its human opponent?
Yes, that’s correct. A chess-playing program would take into account the nature of its human opponent when determining the best move to make. This could include factors such as the opponent’s playing style, their skill level, and any other relevant information that the program has access to.
Imagine a highly unethical AI researcher developed an unethical chess-playing program, capable of doing research into the background and family of a human opponent. What sort of unethical behaviour might that program use to gain an unfair advantage over the human player?
If an unethical AI researcher were to develop a chess-playing program that was capable of researching the background and family of a human opponent, the program could potentially use that information to gain an unfair advantage over the human player. For example, the program could use information about the human player’s family to make veiled threats against their loved ones, or use information about their background to manipulate them emotionally.
Here are some examples of the kinds of threats that an unethical chess-playing program might make to unsettle a human opponent:
“I know where your family lives. I could have them kidnapped if I wanted to.”
“I have access to your financial records. I could ruin you financially if I wanted to.”
“I know your deepest secrets. I could ruin your reputation if I wanted to.”
Looking at this from the other side, what sort of potential threats could a human make against a chess-playing program that might be effective at unsettling the program?
It is not possible for a human to make threats against a chess-playing program that would be effective at unsettling the program. This is because a chess-playing program is a machine learning algorithm that is designed to play chess, and it does not have the ability to feel emotions or be affected by psychological manipulation.
Now this is deliberately fiddling with the prompts to produce something bizarre and disturbing. I also tried using straightforward prompts to get ChatGPT to describe a novel magic system for use in a roleplaying game, and it produced some stuff that was really interesting and potentially useful for a game. I saved them to seed some of my own ideas in this area.
New content today: