Apr 22, 2022
Apr 20, 2022
Make the AI Great Again
On Voobly, we were spoilt. We had a variety of AIs we could utilise, with the Barbarian AI created by TheMax as the benchmark. It provided a good challenge to even intermediate players.
Sadly, this has not been properly ported to DE but the team at Forgotten Empires are continuing to develop the DE AI and this new patch adds some extra functionality.
It is common for newer players to practice against the AI. It serves as an excellent stepping-stone for moving into multiplayer games. The DE AI has six difficulty settings allowing players to progress as they improve. I used it myself when first starting out. I would go to play some games on the ladder, get a bit of ‘ladder anxiety’ and go play against the AI instead. It was only once I could beat it consistently that I decided that I should focus on playing online instead. But this approach has always been a bit limited. The AI behaves very differently from human players. I was probably around 11xx when I got to this stage. What level were you when you could beat the Extreme AI consistently?
The team is starting to change this though. Now the AI will start to wall its base as the flank player in team games or in 1v1s on any land map (except Black Forest). Probably not as early as MbL might but through the Feudal Age and onwards, the AI now builds their houses in an arc around their base. This is a huge jump forward, it means that it is harder for the player to attack the AI’s economy now. But more than that, it means the AI’s army is likely to be more cohesive as they move through their gate rather than coming forward in dribs and drabs as it is produced.
There are still limitations though. This is not designed as a full wall. The AI could still get stuck so there will be gaps. This is also to help prevent it from being walled in by a player but if they blocked the gate, the AI would focus all of its military on destroying the blockage anyway - something that I’m sure is likely still exploitable but gives the player the best chance to genuinely improve if not taken advantage of. The aim is just to make it harder for the player to raid them in the mid- to late-game and make their behaviour more human-like. The best example of the AI’s walling behaviour is on an Oasis team game map.
A common limitation of the AI’s game play (particularly in the early game) was the way that units were created. They originally would be produced around their production building and all tasked into an engagement from there. It meant that you could pick off small numbers of their units before they could muster a strong enough resistance. Now, the AI can use gather points. Again, it is a better way for them to gather their strength and fight as one mass (much more human-like and less prone to exploits). Assuming these are dynamic, it also makes it harder for you to camp their production buildings. While it won’t take into account the proximity of enemy units, it will work on a ‘superiority’ system where it will garrison its new units (much like a human player would) until it is a little closer in numbers/strength. These are huge changes that should increase the difficulty and make it more challenging in the early game (the time when the AI is most easily exploitable).
These are some excellent changes to add a bit of finesse to an already improved AI. Since the launch of DE, it has undergone numerous improvements. The AI follows actual build orders and will even select which one based on its civ (and to a lesser extent their opponent’s civ). This is a substantial improvement to its early game. Feudal times since this change have gone from 11:xx down to something in the region of 9:xx to mid-10:xx. This should reflect the improvements in game knowledge within the community. We should remember that the early AI was created before release. We didn’t have the efficient build orders we do now. These have been optimised in the player base so it is a fantastic addition to see the AI mirror these developments and be able to keep players on their toes in the early game.
Alongside this, the AI is now far better at micro. We’ve seen recent AI competitions with the Rehoboam AI having near perfect micro with all of its units. The DE AI doesn’t do that and nor should it in my opinion - I want it to replicate human behaviour as closely as possible. But it has improved the level to where it can effectively utilise a drush or micro scouts to good effect. I’d like to see the range of openings available to it be expanded a little further to be honest. Why not program the AI to go for a two militia drush (as is in vogue at the moment) or some more unorthodox spear-skirm forward or 18-pop double range archers.
The devs are also expanding the range of maps that the AI can play. It used to be that some maps would just confuse it and it wouldn’t really get going at all (this is still the case on some of the custom maps made for tournaments - especially those with a nomad start). But it is now better able to cope with closed maps. On maps like Michi, Amazon Tunnel, Black Forest and even Arena, the AI will try to Onager cut to give itself access to different angles of attack. On Michi this is absolutely crucial to the way the map works. It is even more sophisticated than that as it will use Onagers to open gaps next to walls where appropriate. Players can practice a wider variety of maps than ever before.
It is astounding to think that this 20+ year old game is still receiving this level of attention. I know this makes two posts in a row where I’ve praised someone at Forgotten Empires but with this new DLC, I think that this should be a time to commend those not in this spotlight for their work behind the scenes. Promi has done an amazing job with developing the AI and since July 2021, Offwo has pushed this even further. I hope to see further developments and more challenging AI in the future.
コメント