Has The Time Come For Adaptive AI?

[Shaamaan]

Adaptive AI makes sense in games like Unreal Tournament. Each player death makes the AI easier. Each kill makes the AI more difficult.

How would it work in ED? Each NPC killed makes NPCs more and more difficult until the player gets overwhelmed and blows up? This sounds... a bit silly.

 

[kilowolf]

We do not need adaptive AI, just need the AI to fit with the rank of the mission. I should not be seeing Expert AI in a Harmless mission just because i'm Expert rank. AI needs to be locked to mission rank, and outside of mission it needs to be random as not everyone is weirdly better or worse then you and everyone being a god like pilot or a dumb     is just stupid. Combat zones and RES sites should be rated by how much traffic they get not what rank people are because everyone being the same few ranks just because is stupid.

Either that or just put a difficulty slider in and let casual players get into the game for the first time ever.

 

[Darth Ender]

It's the only way to do it, if the gameverse is supposed to make any sense.

AI should have the option not to attack, if they don't fancy their chances.
Powerful AI may ignore low value targets.

Other than that, the system should set the danger level.

The very fact that it makes sense is why i have so little confidence that FD will do it that way. Creating a somewhat intelligent equation that governs NPC creation based on system states and such just requires more effort than checking for zero on a modulus to a random() call or just straight guaranteeing npc pirates appear when certain criteria are met (you take a mission).

That time is much better spent getting avatar creation ready and piling more features on top of the game while the last ones you piled on top have already been forgotten.

 

[Kaocraft]

On a gut level I don't like adaptive AI (or level scaling), and it never works properly anyway. The challenge should be the challenge. The best solution is to have a varied range of rewarding interesting activities and regions with varying degrees and types of challenge to them, and to make this variation legible and predictable for the player. Some activities will be out of reach for some players FOREVER, and that's OK as log as they're not caught off-guard by it or led to believe that they MUST be able to participate in it. Likewise some activities are going to be so unchallenging as to be mind numbingly dull to certain players no matter what; and this is *also* OK so long as they are able to reliably predict ahead of time what they are getting into, and do not feel as though they are being *forced* to participate.

The Galaxy should be a varied, dynamic, chunky environment with jagged edges and irregular surfaces. Different areas should be different from each other not just in appearance or name but in opportunity and challenge.

The mentality behind adaptive AI, level scaling, or anything along the lines of "all gameplay styles should be equally valid," is to reduce the texture of the game world to a smooth bland grey paste.

 

[Susanna]

Since the new AI was introduced, many people have stated it was too hard, not hard enough, or just right and they were all correct. At least, they were right according to their own subjective opinions. There was talk of a difficulty slider bar, but that was (thankfully) shot down because it can be abused easily.

The issue is we are all unique individuals. Some players are teenagers, hepped up on Red Bull and sugar snacks; while some like myself are quite a bit older with our twitch combat days well behind us. Some gamers have physical and cognitive disabilities or various combinations of other challenges. How do you satisfy all those camps and maintain a seamless game experience?

For a solution, Karl Marx said it best, "From each according to his ability, to each according to his needs!"

I remembered a paper I'd read about ten years ago and it took me a while to find it (link below). Essentially, it is a treatise on how AI can dynamically adapt to individual game play. Those who find combat too easy will have their difficulty ramped up and those who find things too hard can have an easier time.

I think this is the way forward to even out the bumps in the road for those who have issues. The advantages should be obvious. The game continually adapts to your individual game play over time. The logic will track if you go into RES or CZ actually looking for a fight, someone who does missions, or see if you are a trader or explorer who tends to avoid combat all together. That can be taken into account to shape your overall experience as far as interdictions, AI ability and combat frequency go. Frontier tracks all sorts of data and has done since day one. They should have the basic building blocks in place already. All they need to do is come up with the algorithms to process that data and give players an experience they will enjoy. As abilities develop (or degenerate as the case may be) the game will adapt.

Some will claim this sort of system can be corrupted. You can purposely play below your talents and have the AI be subsequently easy to kill. And while that is perfectly true, such a player would be bored out of their minds after a week of play. Yet, even if that were the case, their experience would not affect yours, or any others. They play the way they want, you play the way you want. Under this system, combat should be within, or very close to, your ability range. Occasional powerful AI can appear and test your abilities. If you rise to the challenge and defeat them, the AI ramps up slightly the next time. If you run with 14% hull, the AI tones it down on the next round. This does not mean it gets progressively harder and harder. You'll have some easy fights and some hard ones, but hopefully none that are impossible.

Now, implementing this system will neither be easy, nor quick. This will take a considerable amount of effort and testing to get right. If it were any other game company I'd be worried, but Frontier has shown that it has a deep talent pool of devs to pull these sorts of rabbits out of hats.

I'm interested to know if you'd support such an initiative and any reasons you have that are for and/or against.

Improving Adaptive Game AI With Evolutionary Learning
Paper available here: http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.109.6055&rep=rep1&type=pdf

It's got issues but I think it's a good idea that FDEV should be looking into ��