It's called Pay Day, but you need a TM which they couldn't have gotten without having Surf first.
Alternatively, a Meowth would work, but as far as I can see it seems that this ROM hack doesn't actually have all 151 Pokemon available to be caught, so there wasn't any wild Meowth available to them.
Oh shit son, the Blue stream is fighting the Elite 4. They're at about 900-1000 viewers right now.
EDIT: Also, I think the tug of war adds an interesting dynamic to the stream, but I agree at this point that the game should almost always be in Anarchy Mode. Safari Zone democracy was necessary, and it may be necessary for some of the Victory Road puzzles, but apart from that, anarchy should be just fine.
Good thing, though: now Anarachy Mode only needs 50% of the votes to toggle, while Democracy Mode needs 80%. So it's been made more likely that democracy will only start functioning in times of absolute need, like it did in the Safari Zone. All other times, it will not get toggled, and if it does, it will only take a bit of start9'ing and vote spamming to get anarchy back.
The only reason I'd expect Nintendo to come down on this is the subscription. Lately, they're live and let live as long as you're not making any money (eg. Mother 3)
How do you guys feel about the concept of "AI plays Pokémon"? I feel like that would be a pretty cool project, but then AI is kinda my thing.
We could even make it relatively TPP-like by adding two major components - a time delay component to simulate the delay on Twitch - random noise inputs; e.g. 90% chance of a random move, 10% chance of following the current strategy.
What I would really like to do is to simulate the Safari Zone to get a reasonable picture of how difficult it actually was.
So first it was SaltyBet. Then it was TPP. It is clear that there are many opportunities for huge success building apps on the Twitch API.
Even though I usually have a zillion ideas, I have no ideas for using the Twitch API. If we can come up with one, we can make it happen. Just copying SaltyBet or TPP won't work. We need a new idea that is just as good. Such an idea will be so obviously awesome as soon as you year it, just as both of these were.
Brainstorm time.
OK, here's my idea.
Let's call it "Twitch helps AI win", or even "Twitch teaches AI to win". Think of it as a general-purpose piece of software with 4 layers: 1) The game itself (e.g. VisualBoyAdvance running Pokémon Red) 2) I/O layer - takes input directions from #3, turns them into a stream of inputs, and feeds them into #1. Also passes game state information from #1 to #3 (e.g. directly reads the memory, parses it, and passes updates to #3) 3) AI layer - gets updates (observations) from #2 and analyses them, outputting actions back to #2. Also has the option to communicate with #4 for strategic input. 4) Web layer - feeds outputs (e.g. video stream, command stream, game state info, AI status) from the other layers to the Web; also accepts strategic input from the Web and feeds those inputs to #3. For more sophisticated input you could even provide a separate web interface to the Twitch stream - that would be a lot nicer to work with for many games.
You can view Twitch Plays Pokémon as an instance of this model in which the AI layer is very simple - it doesn't get game state information and hence is totally reliant on layer #4 for information; you can view the "Democracy" mode as a somewhat more sophisticated AI that performs some processing of the inputs rather than feeding them directly to #2.
I think the key point is that the pure mechanics of Twitch Plays Pokémon are limited to a very small proportion of games, but I think that's first and foremost an interface issue. You just have to step up one or two levels of abstraction from the raw inputs, and provide an interface that lets you communicate higher-level tactical direction.
Even better would be if we could have a more freeform communication layer for #4, but that's a point at which it becomes a much harder AI task. That being said, I think you could get some really good results if you included some approaches from the deep learning field.
The most super-awesome and insanely difficult version would be a learning general game playing AI - it would have to be able to handle any game you throw at it.
In rough terms, think of it as a combination between Twitch Plays Pokémon and Akinator.
It's a poor comparison because Akinator is pretty dumb and a program like this would need to be a lot smarter, but from an overall motivational perspective I think it's the right idea.
Cheese, your idea sounds somewhat similar to what happens in computer vs. human chess. Occasionally there will be a clearly correct move (capture his queen, duh), so instead of the CPU grinding away for minutes on an obvious position, the handlers just move for it.
Here's my idea. 2-player game. Playing against each other, natch.
But each side can spy on each other! And sabotage their plans! Feature, not bug.
On the simplest level, this could be something like MMO pong. But I think with the right game it could be awesome.
Anyway, I've been trying to think of what makes for a good TPP/Salty-style game.
Not real-time
Discrete inputs
Can't get stuck in a dead-end state
Actually MMO
I'm sure there's more that would prove useful to consider. As I think about it more, chaos like this may ameliorate the problems with > 2 player games. You can't have politics if one player can't cohesively act by themselves, let alone collaborate. Also semi-related: Kasparov vs. the world.
"Twitch Plays Pokémon should be ashamed for turning to Democracy as often as it has"
Cheese, the two times democracy was used to an actual effect, was firstly in the frankly infuriating Rocket HQ with the arrow panels that I remember being stuck on for quite awhile as a kid, and trying to plan out moves on paper, and secondly, the safari zone.
The safari zone had two main problems if you remember, it was timed, a certain amount of steps and you were kicked out. The player was also asked to pay for each time they entered, and unlike normal playthroughs where money was no problem, this time due to multiple black outs, money definitely was.
I wonder how long the RNG would be able to solve the Rocket HQ puzzle. It feels like something that seems doable, but would take quite an inordinate amount of time. The safari zone of course, is the biggest issue here
1) You had to know where to go and be efficient 2) You actually needed to pick up stuff like the gold teeth from the floor, how specific the RNG has to be in order to do THAT is going to be a problem. 3) You ran to risk of running out of available money, and save catching a meowth with payday or using the TM, would be stuck. I doubt the RNG would be able to do anything at that point.
I find myself wondering why you seem to object to these two times democracy was prominently used. You also seem to want to compare a RNG AI with that of emotional people watching and trying to stear red in a direction of success, which I fundamentally disagree with. Part of the whole reason why the RNG stream is not getting much attention at all is because its a machine. The stories and memes and organic communication of information when a large group of humans get together in an event, is far more compelling and is equal if not greater to the act of actually getting through the game.
Surely you can allow for that two acts of democracy in an event that has been mostly anarchic, for the sake of the creation of a cultural imprint.
Comments
Alternatively, a Meowth would work, but as far as I can see it seems that this ROM hack doesn't actually have all 151 Pokemon available to be caught, so there wasn't any wild Meowth available to them.
EDIT: Also, I think the tug of war adds an interesting dynamic to the stream, but I agree at this point that the game should almost always be in Anarchy Mode. Safari Zone democracy was necessary, and it may be necessary for some of the Victory Road puzzles, but apart from that, anarchy should be just fine.
Good thing, though: now Anarachy Mode only needs 50% of the votes to toggle, while Democracy Mode needs 80%. So it's been made more likely that democracy will only start functioning in times of absolute need, like it did in the Safari Zone. All other times, it will not get toggled, and if it does, it will only take a bit of start9'ing and vote spamming to get anarchy back.
We could even make it relatively TPP-like by adding two major components
- a time delay component to simulate the delay on Twitch
- random noise inputs; e.g. 90% chance of a random move, 10% chance of following the current strategy.
What I would really like to do is to simulate the Safari Zone to get a reasonable picture of how difficult it actually was.
Also relevant:
Watch live video from rngplayspokemon on www.twitch.tv
https://docs.google.com/document/d/10g8QFYgvvbiPMFy2_HSRGQE06H9kLgbe4KuDKeLyjZ4/pub
I think that if a random number generator can get 4 badges in 76 hours of game time (albeit in a different game), Twitch Plays Pokémon should be ashamed for turning to Democracy as often as it has.
Let's call it "Twitch helps AI win", or even "Twitch teaches AI to win". Think of it as a general-purpose piece of software with 4 layers:
1) The game itself (e.g. VisualBoyAdvance running Pokémon Red)
2) I/O layer - takes input directions from #3, turns them into a stream of inputs, and feeds them into #1. Also passes game state information from #1 to #3 (e.g. directly reads the memory, parses it, and passes updates to #3)
3) AI layer - gets updates (observations) from #2 and analyses them, outputting actions back to #2. Also has the option to communicate with #4 for strategic input.
4) Web layer - feeds outputs (e.g. video stream, command stream, game state info, AI status) from the other layers to the Web; also accepts strategic input from the Web and feeds those inputs to #3.
For more sophisticated input you could even provide a separate web interface to the Twitch stream - that would be a lot nicer to work with for many games.
You can view Twitch Plays Pokémon as an instance of this model in which the AI layer is very simple - it doesn't get game state information and hence is totally reliant on layer #4 for information; you can view the "Democracy" mode as a somewhat more sophisticated AI that performs some processing of the inputs rather than feeding them directly to #2.
I think the key point is that the pure mechanics of Twitch Plays Pokémon are limited to a very small proportion of games, but I think that's first and foremost an interface issue. You just have to step up one or two levels of abstraction from the raw inputs, and provide an interface that lets you communicate higher-level tactical direction.
Even better would be if we could have a more freeform communication layer for #4, but that's a point at which it becomes a much harder AI task. That being said, I think you could get some really good results if you included some approaches from the deep learning field.
The most super-awesome and insanely difficult version would be a learning general game playing AI - it would have to be able to handle any game you throw at it.
How do y'all feel about this kind of idea?
It's a poor comparison because Akinator is pretty dumb and a program like this would need to be a lot smarter, but from an overall motivational perspective I think it's the right idea.
Here's my idea. 2-player game. Playing against each other, natch.
But each side can spy on each other! And sabotage their plans!
Feature, not bug.
On the simplest level, this could be something like MMO pong. But I think with the right game it could be awesome.
Anyway, I've been trying to think of what makes for a good TPP/Salty-style game.
- Not real-time
- Discrete inputs
- Can't get stuck in a dead-end state
- Actually MMO
I'm sure there's more that would prove useful to consider. As I think about it more, chaos like this may ameliorate the problems with > 2 player games. You can't have politics if one player can't cohesively act by themselves, let alone collaborate. Also semi-related: Kasparov vs. the world.Now we just have to wait another two hours for Strength.
Cheese, the two times democracy was used to an actual effect, was firstly in the frankly infuriating Rocket HQ with the arrow panels that I remember being stuck on for quite awhile as a kid, and trying to plan out moves on paper, and secondly, the safari zone.
The safari zone had two main problems if you remember, it was timed, a certain amount of steps and you were kicked out. The player was also asked to pay for each time they entered, and unlike normal playthroughs where money was no problem, this time due to multiple black outs, money definitely was.
I wonder how long the RNG would be able to solve the Rocket HQ puzzle. It feels like something that seems doable, but would take quite an inordinate amount of time. The safari zone of course, is the biggest issue here
1) You had to know where to go and be efficient
2) You actually needed to pick up stuff like the gold teeth from the floor, how specific the RNG has to be in order to do THAT is going to be a problem.
3) You ran to risk of running out of available money, and save catching a meowth with payday or using the TM, would be stuck. I doubt the RNG would be able to do anything at that point.
I find myself wondering why you seem to object to these two times democracy was prominently used. You also seem to want to compare a RNG AI with that of emotional people watching and trying to stear red in a direction of success, which I fundamentally disagree with. Part of the whole reason why the RNG stream is not getting much attention at all is because its a machine. The stories and memes and organic communication of information when a large group of humans get together in an event, is far more compelling and is equal if not greater to the act of actually getting through the game.
Surely you can allow for that two acts of democracy in an event that has been mostly anarchic, for the sake of the creation of a cultural imprint.