Player's strategy, human behaviors

[zupa]

I'm looking over all AIs but only one I've found (trAIns) is "pretending" a human behavior as a main aim. Of course strategies are human player similar, but after a while it's easy to see things are synthetic.

I am making a research now on an AI which would pretend to be a human player. It does not have to be smart, but at the first sight it should look like a human player. Should not play "massive", create human manageable structures and fleet-sizes. I agree that generally AI aim is to pretend human, but I would like to focus on the human behavior aspects so while playing against you would have difficulties with a distinction.

I would like to define basic human-like behaviors in the game.

Maybe you could give me some ideas for strategies. I know how I'm playing what differs from available AIs, but I would be grateful for any other players’ experiences and suggestions how I could achieve the goal.
You can also describe how you play in an aspects which you think differs you and AIs.

[slickgoku]

I think you need to specify exactly what you mean by "acting human."

However you should realize that getting computers to "act human" is something that people are getting paid a large sum of money to attempt to achieve!

[planetmaker]

I'm looking over all AIs but only one I've found (trAIns) is "pretending" a human behavior as a main aim. Of course strategies are human player similar, but after a while it's easy to see things are synthetic.

I am making a research now on an AI which would pretend to be a human player. It does not have to be smart, but at the first sight it should look like a human player. Should not play "massive", create human manageable structures and fleet-sizes.

I wonder what is 'typically human' and what is 'synthetic'?

I actually think there are as many human ways of playing this game as there are players playing this game. And I'm pretty sure that it is not always easy to distinguish what is build by an AI and what by a player. 'Massive' definitely is not something exclusive to AI

[Transportman]

I wonder what is 'typically human' and what is 'synthetic'?

I actually think there are as many human ways of playing this game as there are players playing this game. And I'm pretty sure that it is not always easy to distinguish what is build by an AI and what by a player. 'Massive' definitely is not something exclusive to AI

I think variation it typically human. Most AI's build their lines the same way: from A to B and back, although I did the same when I started. Some make A and B RoRo stations, some make it Terminus, but they make all stations the same. Some AI's make junctions, but those are always build the same way with little to no difference between junction 1 and junction 2. I once had an AI that did make connections with every station, but those junctions were all the same with path signals and tight corners. Also the most AI's I know 'ignore' the acceleration model, and just build tight corners and a lot of bridges (or they build such short trains that it doesn't matter that there are tight corners), getting funny lines if there is a diagonal track that it keeps crossing with bridges.

And do you know an AI that build that kind of massive stations like on your screenshot?

For the topic starter: define manageable, because when I have a line running, it generally doesn't need my attention to operate correctly, and when it does need my attention, the game throws a nice warning at me that the vehicle made a loss. I place the trains in a group and on a timetable, and then it just keeps running without problems and thus without my attention.

[zupa]

I think you need to specify exactly what you mean by "acting human."

slickgoku, this is something I would like to achive with this discussion, because it's
quite hard to define in OpenTTD game.

And I'm pretty sure that it is not always easy to distinguish what is build by an AI and what by a player. 'Massive' definitely is not something exclusive to AI

planetmaker, the image you posted is very inspiring! AIs does not play so massively. They
create large number of vehicles (sometimes hitting the limit), but it's not done in a way you are showing.

I think it's a good point. Transportman pointed it out nicely

And do you know an AI that build that kind of massive stations like on your screenshot?

Sometimes I also play like that, but I've not thought it could distinguish human.

For the topic starter: define manageable, because when I have a line running, it generally doesn't need my attention to operate correctly

You are right Transportman. But I've noticed that for less experienced players they have sometimes problems
with "organization". Maybe it's a matter of getting used to available tools (multiple replacement, etc.).
Being lost it's a totaly "human behavior", but it's not the one I would like to simulate

So first (I hope not the last) conclusions came out of it:
- completely massive approach,
- variations of lines, junctions, stations
- take the acceleration model into account,

[Emperor Jake]

I think aesthetics also come into play. For instance, the AI doesn't care what its tracks and networks look like, just that they function (and not always that). I, as a player, pay quite a bit of attention to what my tracks and the things around them look like. Such as using a few different sorts of trains, adding decorative non-track tiles to stations, levelling small areas of land in order to get a smoothly flowing track, and so on. Another typically human behaviour would be to upgrade existing services as demand increases. Many AIs don't even upgrade their trains, and may still have steam locos puffing about in the 2100's. A human would have electrified that and lengthened the trains, then maybe improved a junction and station for capacity, and then upgrade it to high speed rail or maglev.

All of these things would be very, VERY hard to code for an AI to be able to do... if someone manages it, they are legendary

[zupa]

I think aesthetics also come into play.

Emperor Jake thanks for pointing it out, aesthetics play a role... but also aesthetics is very relative. What's aesthetic for me might not be ok for others.

I think I know what you mean with the networks look, when a human builds everything is just "adjusted" to the current map situation, sometimes a player adds something unexpected. AIs just build a generally planned "specialized network" and that's why it does not look nice.

These things you listed - I've written them on side for myselft to remember. Choosing different sort of trains and replacing is a very good suggestion should not be that hard to do and it will be very nice in results.

I'm surprised, each idea given inspires me more and more!

[Ailure]

Personality is probably the most defining factor for human gameplay. Some people level a mountain just to get the straight path possible and make huge networks, while others tries to disturb the terrain/cities little as possible. There is extremes and inbetweens.

I'm personally one of those inbetween players. I try to strive for efficiency, but without creating stations that are 50x bigger than the city/industry in question.

[Michiel]

I think the main problem for an AI trying to look human is how hard it is to get a view of the bigger picture. A human can look at the map and instantly see what might be viable connections, the lay of the land, etc. An AI is basically looking at the map through a tube, through which it can only see one tile at a time. The kind of visual pattern processing a human brain can do it amazingly sophisticated and will be very hard for an AI to emulate. Since AIs are limited in processing power (limited opcodes/tick), it may not even be possible. Just the algorithms required to handle the terraforming for a huge station like planetmaker posted are an incredible challenge. Still, I'm not saying we shouldn't try

[zupa]

I think the main problem for an AI trying to look human is how hard it is to get a view of the bigger picture.

Yes, but I think that's a partial true. Most of the AIs create a general plan and then go with realization. Human is able to modify the plan, even completely change the strategy (and I'm not talking about replanning procedure).

The kind of visual pattern processing a human brain can do it amazingly sophisticated and will be very hard for an AI to emulate.

That's a very good point! The reasoning of a human player is based on the visual impressions not only data processing. This will be very helpful.

Let's assume we can skip techniqual limitations of the AIs in the game at this point, humans don't have them, players does not think about limited opcodes/tick