Beta testers wanted for Ogre video game

[Steve Jackson]

One of these days we will have self-modifying game AIs. Hmm. Unless we do already.

Difficulty level "just crawl under your desk" ?

And shortly after that, Skynet.

[GranitePenguin]

Originally Posted by Steve Jackson

Anybody managed a complete victory in that scenario?

Playing the Mark III as the Ogre? Pretty much every game.

As the defense, I was even able to beat the Ogre when it took two turns in a row because of the INF ramming bug.

The AI in the beta needed a lot of work.

[GranitePenguin]

Originally Posted by piningforthefjords

I played thru the Mark III attacking twice and easily scored complete victories every time. Not sure why but my Mark III started at M4 and that made a big difference.

Started at M4, or was it just the bug that it didn't count the entry hex as part of your initial movement?

I deliberately moved only three hexes total to at least give it a chance. The defensive AI did n't do a very good job making me choose between multiple poor options. It usually just ran everything up the middle and it felt more like bowling than Ogre. ramming is really effective when it gives you two extra attacks.

They also went too far the other way when I suggested the CP can be placed somewhere other than the last row. When they changed that, the AI consistently placed the CP 2 or 3 hexes forward. *shrug*

[Steve Jackson]

Yes, do not underestimate the meat processor that you were born with. It retrains rapidly and adapts to a lot of situations. Equaling even an average human opponent is a high bar for an AI, which is why some games tweak difficulty just by giving one side more or different units, or even (and I do NOT recommend this) giving a stupid AI better die rolls than the player. Note that I list that last bad idea only to vent about it in general. Auroch is NOT doing that - they are playing fair with die rolls and showing us what we rolled!

[GranitePenguin]

Originally Posted by Steve Jackson

Note that I list that last bad idea only to vent about it in general. Auroch is NOT doing that - they are playing fair with die rolls and showing us what we rolled!

Absolutely. I come by my bad rolls honestly. :-)

[dwalend]

Originally Posted by Steve Jackson

One of these days we will have self-modifying game AIs. Hmm. Unless we do already.

Difficulty level "just crawl under your desk" ?

And shortly after that, Skynet.

(Un)fortunately it's really hard to imitate people.

The current big successes in data mining cited in the press are high-value problems that combine expert analysis with mountains of data and man-centuries of specialized R&D. It's gone from sci-fi to unbelievably expensive. Worse, the machine learning techniques often make it really hard to figure out what the computers have decided to do. See Google's current dilemma in the EU around "Why did you choose to show this ad to that citizen." (Maybe they could build an AI to give satisfactory answers to the EU courts.)

The usual approach for the rest of us to be able to imitate some very smart people in their domain, with some parameters that a very limited feedback controls. I've used playbooks with great success in the past. Also, there was some hope of understanding what the machine was doing.

I know we asked how Auroch's AI worked, but they've been busy delivering the system. Depending on how they did it it could be either really easy or nearly impossible to take your help.

[sir_pudding]

Weird, I must really suck at this game and not know it. I had a lot of trouble beating the AI. I dominated in the one multiplayer match that actually loaded during the first sever test, so apparently there are people who suck worse.

IMO you probably want an AI setting that is dumber than this if you want new players to stay with the game.

[Steve Jackson]

A valid point. A game that you cannot win is no fun ever. Even if you eventually "lose" at a high level, you feel like you have won if you extend play time by long enough. But a game that you cannot lose eventually becomes boring. (Though it can be a lot of fun for a WHILE. Cough, cough, Farmville.)

You know, all I just said there is "People are different and games are different."

[dwalend]

I think there's more hope.

The AI needs to be able to keep the game challenging. I play defense vs a new player; I should give up three tanks or so or. GP suggested it was more fun for everyone to put all the tanks on the table and get the balance by intentionally putting two or three of them too far forward. (And he's right.)

If the AI is able to predict that it has a near-certain complete victory then it could choose to make a rookie mistake to keep it interesting. (And Auroch maybe could add a switch to turn that off and have the AI bring its best game to give us bragging rights.)

[GranitePenguin]

Originally Posted by dwalend

If the AI is able to predict that it has a near-certain complete victory then it could choose to make a rookie mistake to keep it interesting. (And Auroch maybe could add a switch to turn that off and have the AI bring its best game to give us bragging rights.)

I saw one such rookie mistake during a live stream I participated in tonight. The defense had managed to get two GEVs behind the ogre, had it down to M2 with the MB gone, and did not retreat them far enough away so the Ogre was able to backtrack and take both of them out. Had that not happened, it would have been a lot more interesting.

To be fair, it's really easy to screw up GEV placement. :-)