AI Beelines

[Adam_Smith]

Wow! This place has slowed down quite a bit in the last few months. I guess people aren't coming back after summer vacation like was thought by many. It might be just the opposite.

Anyway, I think this will be a very good thread if some people participate.

This thread is about particular beelines that each AI makes. It might be in particular interest for the Believer, Datajack (are there any?), or maybe the Drone players. The reason I say that is because those players, myself included get at least half of their techs on the grift. So you might want to know what techs you might steal ahead of time. Also, if you are playing against a probe-low tech team, and you know they plan to steal tech, then you might want to know what tech they will, or will not be stealing.

This is the catalyst for the idea of this thread. I was playing another one of my hotseat games (against myself). This time it was the Drones against the Believers. The Drones got a big continent with the Hive and the Spartans on it. The Spartans beat the Hive and Domai beat the Spartans. So he had a huge base to work on. Miriam started on a small size island. She actually was not doing too bad. She was kicking back (at war with no one since Santiago beat the #$%^ out of Yang) and stealing tech from Aki-Zeta.

Aki-Zeta beelines for Pre-Sentinent Algorithms and Fusion Power. She tends to neglect D: AP. AMA is included in her beeline though. Domai got Sythetic Fossil Fuels from an Artifact. Then he researched D: AP.

Even though Miriam had Fussion reactor and AAA tracking Domai was harrassing the #$%%^ out of her with his Aircraft, and there was nothing she could do with out air power. He was destroying her enhancements and killing her formers and supply crawlers. Then the game crashed. Boo hoo. It was a fun one.

Anyway, here are some more AI beelines that I know.

Yang, Marr- D: AP

Zak, Aki- Pre-Sentient Algorithms etc.

Santiago- Superstring, AMA (I think)

Dee- Centauri Empathy, Meditation

Morgan- Environmental Economics

Anymore you guys can think of?

[kt_butts]

This is a great idea! I think the best way to figure this out definitively is to make a beeline for the Empath Guild and then keep track of all the AI tech choices throughout a game. It would be interesting to see how they vary from game to game (does the AI research Doc: Flex earlier on a water-logged world?)or if the research is pre-determined. I wonder if it is diferent from level to level? I will take a look this evening.

[Simpson II]

As far as I have been able to tell, the AI always plays as if it were using blind research, doing essentially random techs in accordance with it's faction priorities. I'd be interested if kt_butts shows me to be wrong, though. The easier way to test would be with the scenario editor, just cheat EG or infiltrators and then switch back to a hidden map.

[kt_butts]

Hey, that would be easier! I'll see what I can do.

[Adam_Smith]

Blind research would give the factions different discoveries every game. I'm pretty certain that that is not the case. It is true though that they tend to stick to their faction agendas. For example, Morgan researches build techs first and Zak researches discover techs first.

[Neptunus]

Eh... Excuse me... hmm... eh... could someone tell me please... what's a beeline ... ?

[Adam_Smith]

Beeline is a term that builders have made popular. For example, beeline to Ind. Auto is researching everything in the path to Ind. Auto in order to research Ind. Auto as soon as possible. I don't really do it as much as some people do. What this thread really asks is "in what order do the AIs research technology.

[Simpson II]

Adam,

"I've never seen Yang research Centauri Meditation before Centauri Meditation."

Wow, really?

I think that mark13's point is that the AI doesn't really beeline.. that is, it doesn't pick techs because they are on the way to Doc:AP, but just because they suit it's preferences themselves.
Yang's priorities are build, conquer.. the techs on the way to D:AP mostly have high build and conquer values in alpha.txt. The list is

App. Physics
Biogenetics
Doc. Mobility
Ind. Base
Social Psych
Doc. Flex
Ethical Calculus
HEChem
Gene Splicing
Synth. Fossil fuels
Doc. Air Power

All of these have fairly obvious applications for build and conquer. You can check the alpha.txt values for yourself if you like. I think the lowest is Doc: Flex.
To get to Cent. Meditiation, he would need to get past things like Secrets of the Human Brain and Centauri Empathy, which are both very low for both build and conquer ratings. He ignores them for this reason, and just sort of ends up at D:AP (though there seems to be a random element, so perhaps he could get to Centauri Meditiation first once in a very long while.)
With regard to Morgan, it takes a 3 more techs to get to bio-engineering compared to environmental economics, many of which have low build values - neural grafting and MMI, for example. The beeline to Env. Eco is full of build techs.

In this case, kt_butts should find that the AI follows no set route through the tech tree, though it will tend to research many of the same techs. A record of how it progressed in a number of games should give clear evidence.

[mark13]

As far as I know, rather than using beelines, the AI researches the technology which (s)he is most interested in at present, rather than looking to the future. For instance, Santiago's interests are mainly conquer, so she will research the tech which has the greatest conquer value. If you look in the alpha.txt file, at the tech tree, each technology has four numbers next to it. These represent explore, discover, build and conquer (though not in that order) values for that particular technology. These are what the AI uses. For instance, Santiago might have a choice between High Energy Chem. and Nonilnear Maths.
HEC has a conquer value as 3, but NM's value is 4, so the Santiago AI will always go for NM.

If the AI has two "interests" (Hive has Build and Conquer) the two values are added together. So, faced with the same two choices, NM (Build value 0, conquer 4), and HEC (Build 1, conquer 3), Yang would choose entirely at random.

The only reason AIs generally go for the more powerful techs overall (D:AP for instance) is because overall, the numbers are higher than they would be for less powerful techs (Fusion Power reads 3,4,3,1).

[Adam_Smith]

It doesn't matter why they beeline to D: AP (per say). We are interested in which beeline they go for.

Also, they not only concentrate on one of the four research categories, but they concentrate on certain technologies within that category. For example, Morgan beelines to Enviromental Economics instead of Bio-Engineering (which is the same tech level and also a build technology). Also, Yang beelines to D: AP and not Centauri Meditation (which is the same tech level and also a explore technology). Honestly, I've never seen Morgan research Bio-Engineering before Enviromental Economics, and I've never seen Yang research Centauri Meditation before Air Power. In fact, Yang will always research Gene Splicing before Centauri Empathy just to get to D: AP faster, and Gene Splicing is the same tech level as Centauri Empathy, but it's a build tech.
[This message has been edited by Adam_Smith (edited September 13, 2000).]

[kt_butts]

So far, my findings have been inconclusive. I started with the SMAC 4.0 factions. As of now I've followed 4 games, 3 standard Random and 1 huge 90% ocean. I played all as Morgan (just for consistency). Although the paths were similiar, none of the factions followed the same paths from game to game. This would support the other theories posted. However, in the 90% huge game, Doc: Flex was researched much earlier by all of the factions (the one exception was miriam). This leads me to believe that there are certain map variables that affect AI beelines. At this point, I have limited data (a stream of visiting relatives and a girlfriend whom is becoming salty with my SMAC addiction are conspiring to drag me away from my laptop), but my next step is to check for research patterns when the map size changes. My guess is that the Momentum factions will not beeline to conquest techs as quickly when the map is larger. When I'm finished I will provide further update.
[This message has been edited by kt_butts (edited September 13, 2000).]

[Adam_Smith]

Oops! Sorry for that mess up. That sort of ruined my whole post. That should be, "I've never seen Yang research Centauri Meditation before D: AP."

As far as the techs that you have posted for beelines to D: AP, there are as many explore techs in there as there are anything else. There are even a lot of Discover techs in there.

That all misses the point though. Why are we discussing the path to D: AP? The fact is that Yang beelines to it. I don't care about the programming.

[mark13]

After several lengthy investigations of this phenomenon, it appears that in my original theory, I may have been slightly off the mark. The case that disproved my theory was the Hive getting D:AP very quickly in a number of games.

Although the four values have a large say in what the computer chooses to research, after several games I have concluded that it may be incorrect to state that the AI doesn't beeline after all.

On the way to D:AP, Yang would have to go through both Synthetic Fossil Fuels and Doctrine: Flexibility. These two, although they are explore techs, have a build/conquer value of 3 (not great), whereas D:AP has a build/conquer value of 6. The fact is that although better options were available for his build/conquer strategy, Yang decided to forego these alternatives to research Synthetic Fossil Fuels. There can only be three plausible reasons for him doing this:

1) It gives him better weaponry, and so may have a higher conquer value than it appears,

2) It leads directly to D:AP, a technology with massive build/conquer values.

3) It has the highest build/conquer value of the techs currently available (unlikely at only 3)

The former seems implausible considering the AI conquer value (surely it should be higher than 1?).
The latter may appear plausible at first, however:

Silksteel Alloys has a build/conquer value of 7, and its two prerequisites, Subatomic Theory and IndAuto also have very high build/conquer values (both have four).

In which, case, by process of elimination, it looks as though it does indeed beeline. However, this still doesn't explain why it doesn't beeline straight for Silksteel, because that actually has a higher build/conquer value than D:AP!!

This puzzling development leads me to only one logical conclusion: The AI chooses technologies, at least partly, on circumstance (a theory also proved by kt_butts), the main reason being:

- Yang is generally on the offensive in this part of the game, and may need D:AP to continue an effective assault.

Note that this is the case even if that means beelining to appropriate technologies. If that is the case, then I can only congratulate the programmers of this remarkable game for developing a consistent enough formula for this to work.
[This message has been edited by mark13 (edited September 14, 2000).]
[This message has been edited by mark13 (edited September 14, 2000).]

[Simpson II]

Adam,

The AI value of techs is more complex than build, conquer etc. See mark13's earlier post. The idea of knowing why the AI beelines is that if we do, then it becomes predictable, enhancing the empirical side.

mark13 & ky_butts,

Interesting! It appears that we underestimated Firaxis. I had rather assumed that Synth fossil fuels had a high conquer value (overestimated Firaxis. )
If you want to do more testing, it would be interesting to move the needlejet chassis to a completely different tech and see what happens. Same with the foil chassis, TFs, energy restricitons.. maybe also swap around the order of the factions in alpha.txt, to see whether this is caused by something in the faction files or by something about the faction's slot. That could be important to faction makers.
[This message has been edited by Simpson II (edited September 14, 2000).]

[mark13]

Make that a triple post. Someone delete this, pleeeeease!!!!
[This message has been edited by mark13 (edited September 14, 2000).]