Simultaneous Games

[tbontob]

Originally posted by couslee:

quote:

---

Originally posted by tbontob:
(1)One of the threads discussed how the AI was able to adjust by reasearching and then modifying its ships to attack human empires without first observing the changes in the human ships first.

(2)And yes, it is probably the total score. But that is still a cheat in games where we humans cannot see the others score.

1. not sure about that. bet it was a good thread.
2. setting are NOT CHEATS you can make the "view all player scores" in the game set up.
some "harder" settings give the AI bounses.
some "harder" settings give the player handicaps.
Not selecting "view all players scores" is one of the second variety.

---

1) I search the old threads on a regular basis. Must do 4 or 5 searches a day and read at the very least 100 old Posts, some going back as far as 2001 and earlier.

I came across this particular thread in one of my searches and think it was about a year old although it could be earlier by as much as a year or two. I can't remember as I do so many searches. It was not something I was looking for but found it interesting and didn't take any notes.

Anyways a bunch of "well known guys" were discussing it and had devised some tests which to their minds "proved" the AI cheated.

About two months ago, I asserted the AI did not cheat. Nobody took me up on it. But then I started wondering how AI's could as a group suddenly declare war on me...some of which I had the most minimal of contact. We couldn't view each others scores and yet they knew somehow that my empire was the most strongest of all.

In 1981, I bought my first computer, an Apple II+. At that time, games were few and far between.

So, I thought I would program my own game for the time when a bunch of us guys got together on Saturday for beer and pretzels.

It was called "Prices" and could be said to be an earlier Version of "Capitalism".

Essentially, each player had a company which produced "widgets", with a current retail price of $10.00. Each turn, the CEO had to decide on the price of his widget, research budget, production, the number of plants to build (if any) and a few other variables.

I didn't play as I knew the algorithym to success.

Essentially, the two most important factors were price and research. These had long term effects not immediately obvious.

The other factors also affected the profitability but generally only for that month. So overproducing meant higher inventory costs and a lower profit for the month. Underproducing meant lost sales, and a lower profit.

Price had a major impact on sales. What they didn't know was that a stable price was important to the consumers confidence in their product. So, changing the price up and down from month to month to get rid of excess inventory or to generate sales was not the means to success. IIRC a .1 % increase per month gave the best results. No increase or .2% would still give very good results. Reducing the price was not a good thing as its adverse effects would be felt over many turns.

Reseach had a major effect on sales as well, but its effects were not as obvious as a price change and were spread out over many periods. It also had a multiplier effect. Consistancy in research within a certain range from period to period also provided the best results.

Each month, my friends would be supplied with a monthly profit and loss statement, a monthly balance sheet and a 'decision sheet' on which they would have to submit their decisions for the next turn. They would have no access to their opponents P & L statement and Balance Sheet.

Each year end, they would be supplied with a government statement on such things as total sales, investment,etc in the economy.

It was a lot of fun. And the ones who felt they were losing would often "try" use their decisions to make the program give wierd results to the players. I had anticipated it, but even so the economy would sometimes show signs of a depression or inflation as a result of these particular decisions.

The important thing is that I had not programed an AI to play the game.

If I did, how would I have programed it? If I programed it to price its product with a 1% percent price increase each turn, is that not cheating?

And if it isn't, who would want to play against an opponent who consistently wins?

If I let it just flounder around, it would not be a very good opponent.

Or I could let it cheat and have it take a peak at it's opponents inputs or their profit and loss statements, massage the data and use it to affect the AI's input.

On a basic level, my feeling is the computer must "cheat" in some way because it is not able to learn from its mistakes and successes.