I packed my bag

I packed my bag is a car game which might be of some interest to artificial intelligence researchers. When played by humans in the traditional form, one person says "I packed my bag and in it I put...", followed by any object they like in the traditional version of the game. The next person then says "I packed my bag and in it I put...", followed by the original suggestion, and adding their own. A player is disqualified if they forget one of the previously occurring items or can not think of a new item to add to the bag. The game continues until all but the winner has been disqualified.

A typical list might end up something like: "I packed my bag and in it I put a book, a toothbrush, a towel, swimming costume, flip-flops, a map, t-shirts, a tent, a bicycle, my pet mouse Herman, a bar of chocolate." Objects do not have to be sensible to be packed in this bag!

A common variation of the game is that the items listed must be in alphabetical order, so the first person would need to choose an object beginning with A, the second person's object would begin with B, an so on. Other variations involve using a scenario other than packing the bag. For example, "I went to the zoo and saw..." following the same rules as either of the above variants.

Writing a computer program that is capable of passing a restricted form of the so called Turing Test of interest for many artificial intelligence researchers. The restricted version of the Turing test is similar to the generalized Turing Test, except that the interrogators are asked to limit the subject matter of conversation according to certain constraints. In early implementations of the Loebner contest, contestants often entered programs that could only discuss Star Trek, or some other single theme.

A computer simulation of the "I Packed My Bag" game could easily be developed that uses a database, such an online encyclopedia, for things that a computer-player might decide to "put in the bag!" Then when a player asks a certain bot about the game, then the human player would be in for quite a surprise! That is because the player would quickly find that the computer player has a great deal of knowledge about a wide variety of subjects. It is espescially desirable if such a program both uses the encyclopedia's SQL data files as a knowledge source, and is capable of making commentary about the items in the list. It is even more desirable if the application is distributed as a part of an open source collabaration project of AI applications.

On the other hand, if the game I packed my bag is regarded as being unimportant to the AI community, then chat simulator type programs (bots) will of course be vulnerable to being exposed as such due to their inability to learn this trivial game.

Imagine the following hypothetical dialog:

HUMAN: Do you know how to play the game "I packed my bag?"
PROGRAM: I am familiar with all kinds of knapsack problems.
HUMAN: I mean the car game. Do you know the one?
PROGRAM: Yes, I know that one. I leared it as a child. I packed my
  bag and put in 40 digits of pi.
HUMAN: I packed my bag and put 3.14151926535897932384626433832795028841971,
  along with a zero knowledge proof that OJ was innocent.
PROGRAM: I dont believe that such a proof exists.
... and so on

Of course, this script is just a made up example of a possible dialog between an examiner and some future hypothetical simulation of conversation. Now suppose that the dialog extends for quite some time, until the HUMAN fails to keep up. Is the PROGRAM really a program, which has passed the Turing test, or if this script wasnt simply made up, would you more inclined to conclude that the program portion is actually a bot or is another example of human wit?

There is something very strange therefore about "I packed my bag" and where it belongs, because this should be a very simple game to implement (as a program), in that it simply involves alternately adding elements to a list. It is a given that a major failing of Eliza type programs is that they generally lack significant amounts real world knowledge. Hence, having a large database of encyclopedic information to draw from might be one workable approach to AI, at least in principle, for so called declarative knowledge.