Changes to the Social Engineering models

[bvanevery]

I think you'll see 4x AI improve a lot more over the next 20 years. Self-play neural networks are getting very strong at this sort of problem.

At the risk of derailing the thread, what's the evidence for this in the 4X TBS genre?  I've heard about such work aimed at Starcraft, but that's a RTS.  Humans are limited about what they can decide in a RTS and it's a rather scaled down game compared to 4X.  4X has piles of gewgaws.  I have no confidence in AIs solving it.  The piles of gewgaws arise from commercial production realities, the lack of game design and artist constraint.

Neural nets would be relevant to Collectible Card Games as well.  That is the genre where 'designers' heap rules upon rules upon rules, without any restraint.  They're selling more decks of cards.  Now there are card game genres that are a bit more restrained, like Dominion, where you buy big standard sets of cards as "expansion packs".  It's not about open ended collecting and deck building.  Don't know what a neural net is going to do with that either.  But at least there's some kind of horizon, for when the last game rule has come into play.

I will achieve stronger 4X AI in commercial reality, using only case based reasoning, long before any of these people with "well we don't know what to do, so we'll train a neural net to do everything" approaches.  The way I personally can do that, is having the discipline to integrate AI as part of the required design, production, and testing pipeline.  'Cuz I'm 1 guy, and I can be a fascist like that.

Steve Jobs taught us all how to vertically integrate design.  How not to be "corporate stupid" about what your business model is.

[Nexii]

It's only been very recent that neural networks can challenge and beat the top players in computer games more basic than 4x. I would say it will be a few years at least until 4x starts to make proper use of it. But the day will come. One big factor I'd say is that 4x is relatively niche market for gaming. It is a cost to develop such things, to be the first to implement it. Eventually though it will take less time/effort to go the AI learning route rather than the heuristic. Even the makers of SMAC knew that, they had AI learning all throughout their tech tree.

[bvanevery]

People tend to get conditioned to think of tech progress as a doubling of performance every X period of time, due to some things in the computer industry proceeding at that pace.  This is selective observation.  There are some things in computers that haven't gone fast at all, like VR progress over the past few decades.  And other things in science that do not proceed fast at all.  All years I've watched claims of "whatever, coming soon" over the years, has simply made me tired of hearing the hype.  The hype is cyclical and that's the real pattern to be observed over longer periods of time.  There was for instance an "AI Winter" in the 80s that presaged the VR hyperbole of the 1990s.

[vonbach]

Here's a silly question. If I wanted to change the blurb for knowledge to something else what would I have to change?

I''m going to try moving green to values. This is what I'm working on so far.

Police State,    Chaos,  ++POLICE, ++PROBE, +SUPPORT, -GROWTH
Democratic,      InfNet,  ++EFFIC, +ECONOMY, --POLICE
Fundamentalist,  Psych,   ++GROWTH, +EFFIC, +MORALE, -RESEARCH
Simple,          None,    None
Free Market,     IndEcon, ++ECONOMY, +EFFIC, +GROWTH, ---PLANET, --POLICE
Planned,         PlaNets, +ECONOMY, +SUPPORT, +POLICE, -GROWTH
Autarky,         DocFlex, +ECONOMY, ++GROWTH, +EFFIC, -SUPPORT
Survival,        None,
Power,           DocLoy, ++MORALE, +POLICE, +SUPPORT, -EFFIC
Green,           Integ,   ++PLANET, ++EFFIC, -SUPPORT
Wealth,          IndAuto, ++EFFIC,  +ECONOMY,  --MORALE, --POLICE
None,            None,
Cybernetic,      Algor,   ++RESEARCH,  ++EFFIC, +POLICE, ---PROBE
Eudaimonic,      EnvEcon, ++GROWTH, +ECONOMY, ++EFFIC, +PLANET, -PROBE, -MORALE
Thought Control, PlaEcon, ++POLICE, +SUPPORT, ++PROBE, +INDUSTRY, --RESEARCH

[Nexii]

There's lines in script.txt that you'd have to edit if you want other factions to say appropriate things in diplomacy. The dialogues start at #SOCIAL0CAT1GOOD.

Swapping 2 across categories would mean rewriting about 20 from scratch if you care about that detail. Because the dialogues are often written from the perspective of one SE thinking another is bad. For example the reason power dislikes green, green dislikes wealth, and vice versa. Then the same for Autarky against the other Economics.

Also theres the descriptions in the helpx.txt starting at #HELPSOC00. I've been modding too much to bother but once I settle I'll probably get around to it. Make sure to back these up before you change them

[vonbach]

Thank you.

[Geo]

Some wonders offer benefits or mitigate negative modifiers to certain SE choices. IIRC, these are linked to specific positions in the SE tabel. So you might find that for instance the Longevity Vaccine (less drones or +50% economy), Network Backbone (Cybernetic negative modifier), and Cloning Vats (Power/Thought Control negative modifiers) act not as exptected.

[vonbach]

Some wonders offer benefits or mitigate negative modifiers to certain SE choices. IIRC, these are linked to specific positions in the SE tabel. So you might find that for instance the Longevity Vaccine (less drones or +50% economy), Network Backbone (Cybernetic negative modifier), and Cloning Vats (Power/Thought Control negative modifiers) act not as exptected.

Is it tied to the name or the position in the tier? Do you know?

[Nexii]

SPs are tied to the position. This won't really affect swapping Green with Wealth much, aside from that Longevity Vaccine will get the benefit it had for Green

[vonbach]

SPs are tied to the position. This won't really affect swapping Green with Wealth much, aside from that Longevity Vaccine will get the benefit it had for Green

I was hoping that was the case thanks.

[Nexii]

Yet another SE set I'm playing around with.

Pros:
- AI likes all choices a lot except Police State which may need a buff somehow. I feel like it's super good with Planned though? Oddly going negative ECON isn't viewed as a huge deal by the AI even though it also tries to hit +2.
- 2 choices to get +2 ECON.

Cons:
- Two combinations aren't that viable. Free Market + Power, and Green + Wealth specifically. At least until Eudaimonic. But logically I feel like those combinations were always a bit strange.
- Negative effic and industry aren't represented. Though I've kind of moved away from having these. The former because the AI will tank its EFFIC way more than it should, and the latter because the AI needs a huge bonus to want to lower industry.

Notes -
Only Democratic can boom in this set, no easy booming without Golden Ages. It's a sort of middle ground to test out. It impacts faction design a lot since any Democratic averse or -GROWTH faction cannot boom until Eudaimonic. I didn't like the idea of spreading Growth because then it's just more aversions that prevent a faction from booming. As well it means that any faction with at least 0 ECON can get to +2 ECON no matter their aversions. I've concluded that +2 GROWTH is about as good as +2 ECON and thus Democratic has a fairly bad downside. I may need to buff Police State / Fundamentalist to have more staying power.

Politics, Economics, Values, Future Society
Frontier,        None,    None
Police State,    DocLoy,  ++POLICE, +SUPPORT, ++PROBE
Democratic,      EthCalc, ++GROWTH, ++EFFIC, ----POLICE
Fundamentalist,  Psych,   +MORALE,  ++SUPPORT, +EFFIC, --RESEARCH
Simple,          None,    None
Free Market,     IndEcon, ++ECONOMY, -MORALE, ---PLANET
Planned,         PlaNets, ++INDUSTRY, +POLICE, --PLANET 
Green,           CentEmp, ++PLANET, ++EFFIC, --ECONOMY
Survival,        None,
Power,           AdapDoc, ++SUPPORT, +MORALE, --ECONOMY
Knowledge,       Integ,   ++RESEARCH, ++EFFIC, -POLICE
Wealth,          IndAuto, ++ECONOMY, --SUPPORT, --PROBE
None,            None,
Cybernetic,      MindMac, ++INDUSTRY, ++SUPPORT, ++EFFIC, --GROWTH
Eudaimonic,      Eudaim,  ++ECONOMY, ++RESEARCH, ++GROWTH, ++PLANET
Thought Control, CentPsi, ++POLICE, ++PROBE, ++MORALE, ---RESEARCH

[vonbach]

I tried your suggestion of simply giving everyone ++support and it certainly has some interesting effects. The AI certainly seems more formidable afterward.

[Nexii]

I just redid my entire tech tree although I don't think it will have much impact on SEs for now. Just a few tech changes. May end up changing a few as necessary. I removed support from Police State, as the minimum cost on my units dropped from going to scient's patch. It should still be very powerful especially paired with Planned.

Politics, Economics, Values, Future Society
Frontier,        None,    None
Police State,    DocLoy,  ++POLICE, ++PROBE, ++SUPPORT, --GROWTH
Democratic,      EthCalc, ++GROWTH, ++EFFIC, --POLICE
Fundamentalist,  Psych,   ++GROWTH, ++SUPPORT, --RESEARCH
Simple,          None,    None
Free Market,     IndEcon, ++ECONOMY, --SUPPORT, ---PLANET
Planned,         PlaNets, ++INDUSTRY, +POLICE, --PLANET 
Green,           Ecology, ++PLANET, ++EFFIC, -POLICE
Survival,        None,
Power,           AdapDoc, ++SUPPORT, ++MORALE, --ECONOMY
Knowledge,       Brain,   ++RESEARCH, ++EFFIC, --POLICE
Wealth,          AdapEco, ++ECONOMY, --MORALE, --PROBE
None,            None,
Cybernetic,      MindMac, ++INDUSTRY, ++SUPPORT, ++EFFIC, --GROWTH
Eudaimonic,      Eudaim,  ++ECONOMY, ++RESEARCH, ++GROWTH, ++PLANET
Thought Control, WillPow, ++POLICE, ++PROBE, ++MORALE, ---RESEARCH

All politics are tier 1, except Democracy is tier 2
All economics are tier 2
All values are tier 3
Cybernetic is tier 7
Eudaimonic is tier 10
Thought Control is tier 9

Also the lower tiers are much easier to get to, tier 2 techs only have 1 prereq and tier 3 techs only have 3 prereqs. So it shouldn't take forever to get SEs.

[Nexii]

Some more tweaking
- Fundamentalist got +1 GROWTH instead of +1 MORALE. More of a family focus but not having the freedom and flourishing of Democracy.
- Free Market got -2 SUPPORT instead of -1 MORALE. Either works but I can see the increase price of military as realistic, more mercenary armies. And needed to vary penalties.
- Green got -1 POLICE instead of -2 ECONOMY to boost it. Green tends to be more freedom loving/anarchist, stock Gaia even got -1 POLICE.
- Power got +2 MORALE up from +1. -2 ECONOMY is quite painful.

Green and Knowledge have quite mild penalties. But I also consider their benefits much weaker than the alternatives. All pairings work a bit better with this setup, except maybe FM+Power. Even that nets you MORALE at the cost of PLANET. Thinker AI seems to value PLANET, EFFIC a lot more so Green actually gets picked by non-Green factions. I think even in the base AI this set should work decently. It doesn't hate -POLICE that much.

[vonbach]

Just as an experiment try adding +2 or +3 support and watch how the AI plays.