Greetings, glorious testers!
Check out Alpha Two Announcements here to see the latest news on Alpha Two.
Check out general Announcements here to see the latest news on Ashes of Creation & Intrepid Studios.
To get the quickest updates regarding Alpha Two, connect your Discord and Intrepid accounts here.
Check out Alpha Two Announcements here to see the latest news on Alpha Two.
Check out general Announcements here to see the latest news on Ashes of Creation & Intrepid Studios.
To get the quickest updates regarding Alpha Two, connect your Discord and Intrepid accounts here.
how can intelligent AI boss learning mechanics go hand in hand with un instanced content
Chicago
Member, Alpha One, Alpha Two, Early Alpha Two
its been stated that raid bosses or world bosses have an AI learning mechanic that makes them more difficult the next time round with better/improved loot tables etc, this sounds extremley exciting however i dont understand how this can work when 80% of the bosses are un instanced, how can a boss adapt to a raid group when you have 3 or 4 raids competing over the same boss? also couldnt another raid just grief you by all dying to the same boss to reduce hes loot tables and the degree of difficulty he reaches?
0
Comments
I would assume you don't want a complete answer to this, since it would take at least five pages to explain even the base structures that one uses to design this.
So I'll focus on the part where you are concerned about 'weakening a boss by suiciding into it'.
That's not really what Ashes has claimed it will do. It would be a change in capacity during the fight based on performance. Or, perhaps, it will do what certain other games do and 'save' your personal progress against the boss, so if your group engages it, it raises a value that tells the boss how much to unlock of its powers, but is not connected at all to a different raid.
In most cases AI in games are decision trees and machine states
Relevant context from the Wiki on World Bosses: "Raid bosses are aware of the number of combatants within an area in proximity to them and that awareness is part of an indicator to which behaviors they're going to utilize as part of their behavior tree. So as it's assessing the types of combatants that are facing it, the number of those combatants, the position of those combatants, the abilities and totals of those combatants, it weighs certain actions in its behavior tree and then it acts on those actions; and in the scenario where you're bringing overwhelming odds to a particular fight, that might weigh heavier the AoE options that the boss has access to, where they're utilizing a lot more AoE abilities during an engagement due to that overage of of players. So in that sense it's a bit adaptive. It's a bit dynamic based on the encounter scenario" - Steven Sharif
I would think that based on the information currently available we won't have "AI" bosses. More likely we will see adjustable behavior trees that will provide a variety of combat patterns that can somewhat adapt to a changing situation on the enemy side.
With that being said: I could imagine that a future respawn of the boss will have slightly adjusted stats or maybe a new ability based on the raid that defeated them beforehand.
But 'true learning' is something that is out of question in my opinion, that is technology we have yet to master.
A dev company is doing it's join in using their brains, it is not hard and it can be quite enjoyable, more dev companies should learn from this and try it too.
I will give an example why the boss being aware matters:
One of the hardest boss kills in GW2 is killing at the same time three bosses at the Bloodtide Coast, called Triple Trouble. It needs 20-25 people to kill a boss, but people make 3 parties of 35-40 people, plus many randoms and other parties, so theres from 120 to 150 people at these bosses. You can imagine how this ends.
A handful of gold and a bag of boredom
While this is strictly speaking 'true', I don't think I understand what else you'd want.
Games themselves offer a limited possibility space to players (or they'd be impossible to think about 'balance' in), and therefore any AI will also have a similarly limited possibility space.
AI is about learning the conditions under which to utilize parts of that possibility space. That's nearly always just a small stored Neural Network map on top of the standard machine state anyway.
The 'AI' part has moreso to do with the fitness algorithm over time, which isn't really that hard to do since MMOs can just apply weighting to the 'experiences' of top level bosses. They don't even have consistent experiences.
The game also could add this later, by just 'logging' the entire battle from the boss' perspective and then feeding it back into the system with as many tweaks as necessary, as many times as the team sees fit. So my question would be, what's the difference, other than 'scope'?
Galactic Civilizations is a strategy game, it makes sense having a true AI with neural networks.
Each time you play, the game gets better. If you open the game's folder you can see the AI file growing
That game's AI is terrific!
One time I was in the corner of the map, being beat up and fullly surronded by aliens controlled by the AI.
When I was about to lose my last planet, this AI got invaded and attacked by the biggest civiliztion in that match, my enemy was being crushed.
Then I made many invasion ships full of soldiers and just slow boated towards his planets, when I was 1 turn away from invading many planets, my enemy open a conversation with me, he said he knew what I was about to do and he got offended.
Because of this, my enemy surrendered to the biggest civilization and the biggest civilization bacame disproportionally bigger than anybody else. The other races felt threatened and formed an alliance and fought a major war against him, which gave me some space so I could steal some planets from him and survive
This is all AI based, nothing is scripted
I'm going to assume that you decompiled whatever version of the game you are talking about such that you can say that with certainty, so I won't make any claims about that game.
All I can say is that I know other games that can do what you described that came out in 1999.
I'm not claiming anything against your experience, just noting that your example isn't very convincing/compelling even considering the existence of the AI file.
Anyways, since my question was about what the difference is, and I feel like the conversation will get too long and we'll pass each other like ships in the night consistently if I spend time trying to determine what your perception of AI is, I'll disengage. In this case I'll use the inspiration to work on mine rather than talk about the theoretics.
Yeap, it's, there's interviews and the file is there, it's that clunky mess file nobody understands
ai in games means the mobs, npc or things controlled by the computers. it doesn't mean there's a machine learning algorithm in place.
I'm sure the bosses loot tables and difficulties will change based on how long it took to kill the boss after the first hit was done onto it