[SPOILERS] JR4 tries his luck in PB 37

[shallow_thought]

My point is that it`s far easier to make it play chess than a game where a lot of the information is hidden.

Absolutely. It will be interesting to see how well AI's can be made (or have been made and I haven't been folllowing closely enough) to play superficially less complex games than Civ that have "hidden" information. Poker, bridge etc. spring to mind.

[RefSteel]

Good point about hidden information - but I'm not sure that humans can handle it any better than machines. In fact, a powerful computer could do "C&D" analysis like I was doing in the early game much more effectively and keep it up much longer, including retroactively "catching up" once it revealed the terrain on which the other players were playing, so less of the game's information would actually be hidden from it. I also suspect a well-designed machine would actually handle the fog of war better than a human, because it would be quicker to recognize and adapt to the impact of each newly-revealed tile (or other piece of information) on its gameplan.

(Further discussion spoiled for veering way off topic, and length.)

I tend to think in general that computer capabilities are way ahead of where the vast majority of people believe they are - and progressing more rapidly than nearly anyone expects.  (I also think that most of the time, the biggest obstacles are wildly different from the ones that most people - including me - anticipate.) On the other hand, it's very possible to over-estimate computer capabilities, especially when large companies have a vested interest in persuading people that they're already better than they really are. For instance, Google's self-driving cars don't get into many accidents on a per-mile basis, and as far as I know have never had a collision where they were supposedly "at fault," but they only move on low-traffic streets they know with speed limits under 35, and their accident rate is still something like 75% of the average human driver's, which suggests to me that they're doing a fine job of not hitting anything, but a lousy job of driving: The best driver I know - the guy who taught me about escape lanes and generally improved my situational awareness behind the wheel just by talking about what he was doing - once explained to me that driving safely is mostly about anticipating what everyone around you is going to do, preparing for it, and behaving predictably yourself, in order to make it easier for the drivers around you to safely share the road with you. I suspect that self-driving cars can't do either of these things yet; the computer doesn't know how to act like a human, which can result in human drivers misinterpreting what it's about to do, and (to take an extreme example) when it "sees" people driving erratically, the computer doesn't know to give them a wide berth - only that they are moving objects to be avoided on their present trajectories. Both of these problems would result in an increase in accidents where the other person is at fault, in situations most human drivers would have successfully avoided the problem. Similarly, the driver-assistance/collision-avoidance programs used by the likes of Tesla are great precisely because they can identify threats using a system independent of their human drivers, but no matter how they're advertised, they explicitly aren't meant to replace the driver - just to catch certain dangers that the driver might have missed. I want to believe we'll one day have computers that are better at driving than any human being, but we aren't there yet.

[RefSteel]

Absolutely. It will be interesting to see how well AI's can be made (or have been made and I haven't been folllowing closely enough) to play superficially less complex games than Civ that have "hidden" information. Poker, bridge etc. spring to mind.

Poker's been done, apparently.

[EDIT: The article I linked is over-selling or misinterpreting the results though. It's no surprise that a computer can "bluff" - it's not like they have any "tells." And in spite of the bet size being unlimited, the range of possible behaviors for it to analyze and predict is highly constrained - but it is definitely handling hidden information very nicely.

I find the Bridge question interesting because it requires coordination (but forbids communication) with a teammate. Computers can apparently play it very well, but as far as I can tell with a brief search, they aren't world-beating yet. I'm curious about how well a computer could be trained to work with human teammates. Ultimately though, (most) games are practically designed to be ideal computer environments, because they have straightforward, hard-and-fast, essentially mathematical rules that a computer can learn to exploit optimally. The poker computer doesn't understand the other players' psychology; it's just really good at guessing what their cards are likely to be, based on their bets in the context of the game's history, while effectively preventing them from guessing its cards on the same basis. In a real-world environment, computers can't just be "taught the rules" because not all the rules are clear-cut or even known, and there are no win conditions to target. Except ... at the highest economic levels, modern western financial systems are highly exploitable by computers - and as I understand it, that is already being done actively.]

[JR4]

I`m a bit surprised that the computer hasn`t learned how to beat the best bridge players yet. Well, it`s probably inevitable in a few years. The fact that it`s a team game (without allowing direct communication) probably doesn`t help the AI.

An advantage for the humans playing civ (in contrast to blitz, rapid or classical chess) is the lack of time pressure. With a timer of, say, 24 hours you`d easily be able to play even the most complicated turns without feeling that pressure. It`s more resembling of correspondence chess than anything else. Also, two brains tend to work better than one, so I still think that putting a couple of the top players in charge of a civ would beat anything the AI could come up with. At least in the foreseeable future.

EDIT: I agree that the AI could be taught to do the C&D and the city management almost perfectly, but still think it`s going to struggle with defending against intercontinental invasions.

[RefSteel]

I`m a bit surprised that the computer hasn`t learned how to beat the best bridge players yet.

Likewise, and it seems to support to your point about hidden information. Poker does feature hidden information, but the nature of that information is relatively trivial: A given hand is better or worse than another hand. With Bridge, high-level play (as I understand it at least) requires inferring relatively detailed (probable) information about other players' hands on the basis of the cards they play and their expectations about other players' hands formed on the same basis, while helping a teammate to better make the same inferences. That's a far more complicated interaction ... and in some ways resembles the (even more complex) player interactions in a multi-player game of civ.

[mackoti]

very interisting discution here about how good you can make an AI to play civ.What i want to say to make it to play good you need a huge data base with best game and  because of that i didnt report mine because i was afraid of that   .And i dont know how you can make an AI to atack proper and dont get trashed in fight.I can keep lower power bait it to atack and having 30 units finishing in my que and distroy all theyr stack in my culture.And the number of games which actualy AI can learn is low and 99% of them are played under Seven skill for example.So dont see how actualy would win without giving some 'unfair' adavantages.

[JR4]

Heh, you might be right that creating a very big database for the AI to use could be a problem because of the lack of well documented games by top players. I definitely agree about the AI struggling with attacking and defending (where it needs to take decisions by itself).

In other news, it seems that PB 37 is about to come back online soon! If we can get that reload and dtay will finish his turn, then maybe I`ll get to play turn 87 sometime today.

[JR4]

TURN 87

A quiet turn for us. We`re making +24 gold per turn at full tax, I wish it`d have been a bit more than that. We need to save a lot more gold before being able to finish Currency.

Our scouting chariots continued their work. The western one jumped back on the galley. I think we`ll be moving it back toward our main lands now. The other one discovered a few new tiles E of Kokejin. The galley is probably going back to Muqa now to add more flexibility for troop movements across the water.

The missionaries continued toward Oghul Qaimish and Borte. Would it be better to let the one destined for Borte jump on the galley that`s finishing in Muqa this turn? We could also whip out the galley for overflow into e.g. a spear this turn. Oh, and Muqa is growing into unhappiness now (it has 4 turns of whip anger left). We could turn on avoid growth or just let it grow and then whip away excess population.

Joey`s phalanx still covered his worker cutting down the forest W of Nambui so I moved our own worker back toward Toregene.

EDIT: I started a Barracks in Borte. The plan is to switch to a settler right before it grows into unhappiness.

[JR4]

Graphs and demos: It looks like our neighbors turned science on a couple of turns ago. Nothing too frightening, really, as we`re getting close to the 100 beaker mark (97). Power is still looking good. Dtay building yet another city is however, a bit scary. Still, he`s not built in our border lands so as long as he expands elsewhere it isn`t too bad.

We`re still competitive in food and mfg. It looks like we`re going to struggle to match my T90 prediction of having the highest crop yield, though!

[JR4]

TURN 88

Today the turns are flying by! We have something to make up for, as the game was paused for four days prior to that. And dtay already played before us.

I decided to offer Couerva open borders. He.. accepted. That`s very good news for both of us. We got +3 commerce, 1 each in Borte and our island cities. According to the demo screen we`re still at 0 in export/import so it`s a fair, balanced deal. And once we`ll get Currency it`ll be even better. We`re now at +30 at max tax, it`s looking better. And Currency is a 5 turn tech now (we`ve enough gold for 3 of those turns).

The worker on the desert hill N of Muqa hopped on the galley. It`s going down to Radnashiri to help improve the gold asap. We`re getting it online on turn 96. Muqa started a spear that`ll be ferried over to Radnashiri. We can`t afford to skimp on the military there imo. The galley carrying the E chariot is going back to Muqa.

Two workers completed a road just SE of Toregene. They`re available for other duties next turn. The plains pig is being improved as well. Both workers from the corn are there now (we`ve got 1 turn invested already). The spear in Oghul Qaimish was moved S to the sheep, it`ll go to the forest next turn. A worker is lined up to chop it. The missionaries are spreading the faith to Oghul Qaimish and Borte next turn. We could whip Oghul Qaimish now to finish the lighthouse and regrow this turn. What do you think?