I'm happy to see interest in this. I'm the creator of the platform that it is played on (SCHNAIL - www.schnail.com)
Happy to discuss any observations and questions about the bots and the whole scene. I recommend reading this first: https://schnail.com/#/faq
Feel free to try your luck against them, it is free to do so.
Usually the bot authors are not that active in tl.net, or any other social media. I'll forward the interesting questions and post the answers where applicable.
And I really hate this part but I have to say, I'm not interested in: - Shit talking the AI. These are labors of love from hobbyists, who put on considerably more effort than someone queueing on ladder. Treat them as such - Anything answered by the FAQ - Any kind of drama. I'm happy to participate in a civil discussion, everything else will be ignored.
On February 27 2024 05:42 Nirli wrote: Might've been answered somewhere, but still:
Do the bots have vision of the map, i.e. black sheep wall?
No, they pretty much play by the same rules as humans. The only thing that is slightly different is they know where invisible units are - they still can't target them. In practice, that hasn't been an advantage for them.
I always wondered - is there a way to create AI, which has similar APM to the human ? It would be more fair competition, let's say maximize it at 450 or smthing.
On February 27 2024 07:41 ZZZero.O wrote: I always wondered - is there a way to create AI, which has similar APM to the human ? It would be more fair competition, let's say maximize it at 450 or smthing.
ZZZero, I love you, but this is, in my opinion, a wrong way to view a bot's abilities.
Bot APM is not comparable to human APM. Also, human APM is also not a good metric of performance. It tends to correlate with higher skill levels, but spam-clicking is very much a thing, and we have seen pros with lower APMs.
I feel like the AI cannot win here - if they lose then "AI can never beat humans", if they win "oh because high APM". Essentially, I don't see that the high APM gives them a meaningful advantage at this point. If and when we reach that point, the it is worth talking about this. Even then, any amount of processing power can go into a decision that leads to exactly one click. It is not an effective metric in the first place, and it isn't that hard to game it.
On February 27 2024 05:42 Nirli wrote: Might've been answered somewhere, but still:
Do the bots have vision of the map, i.e. black sheep wall?
No, they pretty much play by the same rules as humans. The only thing that is slightly different is they know where invisible units are - they still can't target them. In practice, that hasn't been an advantage for them.
So that one time AIlien beat me on Benzene it was just me sucking, nice. Playing the bots has been really fun, way more fun than diving into the ladder after years of inactivity. I'd recommend them to anyone new or coming back to the game, obviously way better than the default Blizz computer. So thanks for the hard work!
On February 27 2024 07:41 ZZZero.O wrote: I always wondered - is there a way to create AI, which has similar APM to the human ? It would be more fair competition, let's say maximize it at 450 or smthing.
ZZZero, I love you, but this is, in my opinion, a wrong way to view a bot's abilities.
Bot APM is not comparable to human APM. Also, human APM is also not a good metric of performance. It tends to correlate with higher skill levels, but spam-clicking is very much a thing, and we have seen pros with lower APMs.
I feel like the AI cannot win here - if they lose then "AI can never beat humans", if they win "oh because high APM". Essentially, I don't see that the high APM gives them a meaningful advantage at this point. If and when we reach that point, the it is worth talking about this. Even then, any amount of processing power can go into a decision that leads to exactly one click. It is not an effective metric in the first place, and it isn't that hard to game it.
It's easy for people to be confused about what strengths an AI has, so you can't blame them. Big number = strong, right? But we know that isn't the case. Sure it can spend a thousand APM mineral boosting to get ahead economically, and do some really cool micro to get better trades, but humans have the benefit of a brain and all that entails in terms of decision making, foresight, contextualization, etc. which are hugely important tools in strategy games. Human players and AI players are just in completely different worlds as far as their strengths are concerned. Limiting a scripted AI's APM is only a huge detriment. It makes them a lot less interesting to play against. It's fun seeing what they can do, that you could never manage yourself. Monster chasing those zealots while move attacking literally perfectly was really cool! But I do wonder what called for it to give up that chase, since those were free kills right there.
This is not written with any malice, but can we consider these bots “AI” or just a bot executing a script?
I’ve seen the games and the bots aren’t good. How much learning/intelligence do they apply in game or are they simply following a script that says if X do Y?
On February 29 2024 12:07 RowdierBob wrote: This is not written with any malice, but can we consider these bots “AI” or just a bot executing a script?
I’ve seen the games and the bots aren’t good. How much learning/intelligence do they apply in game or are they simply following a script that says if X do Y?
Yes I wonder this too. If they play 1,000 games on the ladder will they become significantly more skilled by the end of it?
On February 29 2024 12:07 RowdierBob wrote: This is not written with any malice, but can we consider these bots “AI” or just a bot executing a script?
I’ve seen the games and the bots aren’t good. How much learning/intelligence do they apply in game or are they simply following a script that says if X do Y?
Yes I wonder this too. If they play 1,000 games on the ladder will they become significantly more skilled by the end of it?
I am 99% confident we are in the "if X do Y" field ... no matter how many matches are played.
Bot APM is not comparable to human APM. Also, human APM is also not a good metric of performance. It tends to correlate with higher skill levels, but spam-clicking is very much a thing, and we have seen pros with lower APMs.
I have entered into this "discussion" many times ... and NO, there have been no pro with low apm, unless you consider 250 to being low ... which honestly i dont.
Beside this i can safely say that from the replay era on there have been no pro with low apm (some approx 2005-2010 era EU good players cant be considered pro imho).
At the net of spam clicking...as a general rule apm is an extremely good index of HUMAN skill level and so mentioning exceptions to prove a statement is not a good way of demonstrating things.
Bot APM is not comparable to human APM. Also, human APM is also not a good metric of performance. It tends to correlate with higher skill levels, but spam-clicking is very much a thing, and we have seen pros with lower APMs.
I have entered into this "discussion" many times ... and NO, there have been no pro with low apm, unless you consider 250 to being low ... which honestly i dont.
Beside this i can safely say that from the replay era on there have been no pro with low apm (some approx 2005-2010 era EU good players cant be considered pro imho).
At the net of spam clicking...as a general rule apm is an extremely good index of HUMAN skill level and so mentioning exceptions to prove a statement is not a good way of demonstrating things.
Bot APM is not comparable to human APM. Also, human APM is also not a good metric of performance. It tends to correlate with higher skill levels, but spam-clicking is very much a thing, and we have seen pros with lower APMs.
I have entered into this "discussion" many times ... and NO, there have been no pro with low apm, unless you consider 250 to being low ... which honestly i dont.
Beside this i can safely say that from the replay era on there have been no pro with low apm (some approx 2005-2010 era EU good players cant be considered pro imho).
At the net of spam clicking...as a general rule apm is an extremely good index of HUMAN skill level and so mentioning exceptions to prove a statement is not a good way of demonstrating things.
I agree with what you say on AI apm though.
Didn't savior play like with 200 APM?
I have never seen a pro Korean below 250-300 (when playing relaxed) Savior is no exception and even if you can maybe mention one guy with "low" apm this is indeed an exception and not the rule. The rule is you need 300 apm to battle with the best or you will not win. APM is not everything but I basically agree there are no pro with low apm because you can not play at a certain level without.
On February 29 2024 12:07 RowdierBob wrote: This is not written with any malice, but can we consider these bots “AI” or just a bot executing a script?
I’ve seen the games and the bots aren’t good. How much learning/intelligence do they apply in game or are they simply following a script that says if X do Y?
They're almost exclusively scripted. AFAIK none use any type of deep learning, so these are not really comparable to the GM bot that played SC2, or alphazero (chess, go) or really any type of AI that either use self-play or apply machine learning techniques which require data for training.
On February 29 2024 12:07 RowdierBob wrote: This is not written with any malice, but can we consider these bots “AI” or just a bot executing a script?
I’ve seen the games and the bots aren’t good. How much learning/intelligence do they apply in game or are they simply following a script that says if X do Y?
In order to answer that, you would have to have a definition of AI. To this day, there is no real consensus at that.
Enemies in the vast majority of video games execute scripts, and we have no problem calling them AI.
As with the APM, I think it is meaningless to argue about this, people will call it whatever they want, I'll continue to call them AIs.
As to the learning aspect: Some of the bots do have learning, if you play them multiple times, they will adapt. Since SCHNAIL keeps learning files locally, this doesn't really shine - as humans learn too.
EPT
