SMAX - The Will to Power - mod

[Alpha Centauri Bear]

Fuf. I think I am done with AI work. At least for now. Either way, I don't want to do any major changes in near future.
https://github.com/tnevolin/thinker-doer/releases/tag/296

The major achievement is that WTP AI constantly beats vanilla even from most inconvenient starting locations. I didn't plan to make it play like a gross master. Just to be challenging enough for human.


A word of advise. HIGHLY EXPERIMENTAL VERSION.
Due to deeply sophisticated algorithms it may fail. I tested it and squashed all bugs I could find but some could still lurk around. Please report.
Another thing is that I increased number of path finding overall. That may result in slower strategy computation phase. Sometimes I observed up to 10-15 seconds. Could increase with map size or number of bases/units. Let me know if you experience serious delay. That should be no too bad anyway as whole mid game turn may take 1-2 minutes just for pieces pushing time.

[bvanevery]

In WTP changelog for version 297 it is noted:

* Disabled tech stealing from faction with vendetta.

Does that mean if someone declares war on me, suddenly they are invulnerable to me stealing techs from them?  Awfully handy and convenient for them.  Would be a seriously bad design decision IMO.

Or does "tech steal" mean something else, like getting a free tech when you conquer a base?

[Alpha Centauri Bear]

Stealing tech with probes.
Yes. They are invulnerable from you stealing tech from them during vendetta. That is the idea.

[bvanevery]

So everyone's getting something approximately like the Hunter-Seeker Algorithm, although not with actual destruction of the probe teams, just for declaring war.  Bad idea.

How many games have I played where I waited until I was in a war with someone, before I started stealing from them?  Stealing while at peace risks war.  Unless there's no diplomatic relationship at all, which is unusual in practice.  Not unheard of though.  Happens with Alien factions fairly regularly, even after they're speaking to you.

[Alpha Centauri Bear]

What would be good idea to reduce amount of techs one can steal?

[bvanevery]

This sounds like a solution in search of a problem.  Who are you worried about doing the stealing?

Are you trying to balance things for multiplayer?  If you are focusing WTP on multiplayer concerns, that's a whole lot of specialized design that I don't have that much input on.  I don't have any experience in multiplayer really, having played a game against an actual human opponent maybe only 2 times in my life, on a LAN in my apartment back in the stone ages.  Even then it was only 1 vs. 1, so not very "multi".

If you are worried about AIs spreading techs too fast by stealing them from each other, do you have evidence that this is the actual major vector of how they acquire techs?  I have my doubts, particularly because in stock, probe teams tend to cross terrain very badly and get themselves killed.  I do have armored probe teams in my mod, so they perform a little better, but that's primarily about threatening the human player and endangering bases to mind control.

If you're worried about a human player stealing from AIs in a single player game, you just shouldn't.  It's not a problem.  Frankly it's standard drill to steal from AIs that are doing better than you are.  Especially on larger maps, it's not that easy to steal stuff, because it takes a long time to push those probe teams safely across the map.

[Alpha Centauri Bear]

I was trying to solve the issue you created.
https://github.com/tnevolin/thinker-doer/issues/92

If you think this is not part of the problem, I can revert it. No biggie.

[bvanevery]

Wrong approach.  The problem isn't AIs or anybody else stealing tech, the problem is them trading tech.  Said so in that Issue, 2nd paragraph:

Tech is ridiculously cheap. There is no reason to do any research at all, as you will simply buy all your tech for next to nothing, from factions that freely sell it. Or even give it away to allies. RESEARCH penalties that some factions carry, like the Believers and the Free Drones, don't mean a darned thing. To the extent they are meant to offset and "pay for" their faction bonuses, it's a complete deal / slam dunk. A disadvantage that means absolutely nothing to them.

Please revert the probe team no tech stealing stuff.  It's actually work to steal techs.  I played a 1+ month game of my own mod recently before getting tired of it.  The Spartans had me totally bottled up by land, and the Caretakers had me totally bottled up by sea, as far as me being able to steal tech.  Couldn't do it, for like forever.  Couldn't put armored troops out there to protect probe teams to get them to targets, because Spartan land onslaught was that fierce.  I eventually gave up the game because I got to a point where I thought I was making progress, only to have it continue as the WW I style infinite unit pushing bog down again.  I can't say that my mod beat me, but at least in that long game, it drove me to something pretty close to a stalemate.  With my patience at least.

Air cover might have snuck some probe teams around over land, if I had made more planes.  But, in my mod I do have cheap Laser Interceptors being made by the AI.  They do shoot down my planes.  Might have worked some but not great.

And I don't even have any of your fancy AI stuff from either your efforts or Thinker.  Just leveraging what the stock binary can do, by providing the right predefined units.

[Alpha Centauri Bear]

Sure. 298.

[Alpha Centauri Bear]

I am testing all WTP AI factions now and noticed that they populate most of the planet by 100 turn and completely all of it by 150 or something. This seems like a too rushy game comparing to vanilla. Not that I am completely against it. After all I specifically designed early game to be less stagnated. However, maybe it makes sense to slow down this rapid expansion to prolong the game in general. Raising nutrient cost to 20 (from 15) didn't help much, surprisingly. They still find enough capacity to expand quickly. They also not too much affected by bureaucracy just because they have enough capacity to absorb drones. After they cover all the planet the biggest cities are still size 5-6 which is not too much.

I am thinking on reverting my farm production back to vanilla's +1 (from +2 now). That will significantly reduce expansion to moist areas only until TF/HF and condensers appearance. Not on sea, though.
Other option would be to increase colony cost to 8 (from current 6).

Other thoughts?

[Alpha Centauri Bear]

Overall, the new AI effectiveness + all the speedups I gave from start result in game ended by AI itself by ~250 turns.
Need to slow down things a little.
 

[bvanevery]

Does that mean you played against the AI and it kicked your butt?  In which case I'd agree, you need to give a competent human player a chance to actually do something.

Or did you do an automated AI vs. AI test?  If so, that result doesn't prove anything.  Your game needs to be tested with human resistance.  For instance, a human might be better at grabbing all the Secret Projects, giving the AIs less advantages to win the game with.  The political calculus of the game might be a valid problem a human player can intervene in.

By way of comparison, AI for SMACX AI Growth mod is capable of winning by transcendence in 400 turns in automated tests.  Doesn't always happen, but it happens often enough that it can be expected.

[Alpha Centauri Bear]

Nope. I am not onto human vs. WTP AI testing yet. I mean I surely played against it for AI tuning purposes but didn't test it massively for the benchmark.
If have tested 1 WTP computer player against 6 "vanilla" computer players so far.

I agree that it does not show that AI is good enough against human. The question of this post is not the quality of AI in general. It is that AI becomes economically more savvy to the extent it finishes the game too quickly by itself. I don't remember vanilla even was able to do it in time. So I am thinking on generic game slowdown. Irrespective to AI vs. human competition.

[bvanevery]

Then don't make any drastic decisions about what needs to happen.  You need real player feedback.  Which can (and typically does / must) include you playing the game.

I mean I don't just sit around letting AIs get ahead forever.  I react.  Sure I've seen AIs transcend in AI only games.  They haven't transcended in my games, against me.

[valox]

to slow the game down, we could encumber players that buy and steal tech.  In terms of lore this could be explained as "additional energy is required to service non-proprietary technology as we apply it to industrial and scientific applications moving forward."  Perhaps this 'encumbrance' is much like facility maintenance costs already in the game.  Additionally, i think it would be ideal to scale the encumbrance according the number of stolen/bought techs already possessed, using some kind of more-than-linear but sub-exponential algorithm--to the effect that its perfectly fine to have 2 or 3 non-proprietary technologies, but you'd think twice going from buying/stealing a 9th tech.

These are the consequences I'd expect.
1) serves as a negative feedback loop that encourages varied play styles
2) maintains player agency by not telling them what they can or cannot do
3) increases the perceived 'meaningfulness' of how one acquires technology. Namely, developing your own tech is now much more exciting since it is now perceived as being more 'sustainable' and stealing/borrowing is more thoughtful, because players need to self-regulate.