I spent a few minutes looking at the CalTech Turing Tournament website http://turing.ssel.caltech.edu/index.html I came away rather puzzled. This seems to be a number guessing game. Sure, it includes both emulator and detector algorithms, but such a specialized domain seems less interesting than algorithms that play chess, bridge, go, or whatever.
Does anyone here have any idea what the value of this Tournament is? Other than having fun and spending taxpayer dollars, that is. >From the website: "The Human Behavior to be Emulated. "An even number (16) of human subjects are matched in pairs: subject 1 is matched with 2, subject 3 is matched with 4, etc. For each pair of subjects, the odd player is the row player, and the even player is the column player, and they then play a repeated normal form game, whose stage game looks as follows: [matrix table ommitted] "The game is played for 50 rounds. In each round, the row player chooses the row, say i, and the column player chooses a column, say j, resulting in a cell (i, j) that is chosen in that round. The first entry aij in each cell represents the payoff, in cents, to the row player if that cell is chosen, and the second entry bij in the cell represents the payoff to the column player if the cell is chosen. Before the experiment begins, the subjects in each pair are shown the payoff matrix, they know that they will play exactly 50 rounds of the game with the same partner, and they observe, after each round, the row and column choices that were made by themselves and their partner, and the payoff that each of them received in that round. At the end of the experiment, the subjects are paid the total amount that they have earned over the course of the 50 rounds. The normal form game used as the stage game in the experiment is recorded in a file in the format of a stage game. The output from the experiment consists of a 16�50 matrix of integers, whose (i, t)th entry is the strategy selected by player i in round t of the match. This is written to a file in the form of a dataset file." Kevin Copple -----Original Message----- From: [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED]]On Behalf Of Ed Heflin Sent: Thursday, January 23, 2003 2:37 AM To: [EMAIL PROTECTED] Subject: Re: [agi] Turing Tournament Damien, This is great stuff...and I wouldn't expect less from CalTech, although I might have expected MIT to be the first to formalize this ;-). Just kidding...I became aware of the sometimes intense rivalry between the two institutions during my post-doc work in experimental gravitational-wave detector development some years ago...imagine trying to observe a fluctuation the size of a hydrogen atom over the earth-sun distance...a proposition in 'quantum fluctuation measurement insanity' that both institutions took on with real scientific fervor and seemed to make some progress in! And, as you point out, the new Touring Tournament at CalTech comes from the unlikely source of the Division of the Humanities and Social Sciences. Now that has got to be a first! I do think that they add something entirely new and necessary, that isn't captured by the "The First Turing Test" for the The Loebner Prize. I think the new element of automated testing is a real plus for the competition and will tend to objectify the evaluation process and enhance the standards of the competition. Furthermore, it is an enhancement that probably even Touring himself might not have expected...the use of both, an algorithmic detector with an emulator. But more importantly, I think this sets the stage for ultimately asking what the real test for mimetic behaviors should be, especially given the larger goals of AGI. When I stop and ask myself...exactly what is being tested in a 'Touring test for mimetic behavior'?, I know my answers different for an AI system, say a chatbot, v. an AGI system, say a cognitive virtual player. Since I expect that the behaviors of an AGI system are of greater 1. breadth and 2. depth than an AI system, I would seek a different forum to test greater mimetic behaviors appropriately. Over the summer, I came to the conclusion that a more complete Touring test of an AGI system would involve a richer I/O environment, say a virtual world, and the ability to endow virtual characters in this I/O environment full cognitive abilities. In this sense, the Touring test of an AGI system becomes more of a test grounded in game play whereby everything from speech interaction to strategic interaction is tested in a human player v. virtual player (computer). The game play than becomes the basis of the judgment on the virtual player (computer)'s ability to mimic human players. Just my $0.02 worth. EGHeflin PS: Please excuse the slight out of sequence submission...but, I'm just getting caught up with things after a bit of R&R in the sunshine state! ----- Original Message ----- From: "Damien Sullivan" <[EMAIL PROTECTED]> To: <[EMAIL PROTECTED]> Sent: Tuesday, January 14, 2003 10:20 PM Subject: [agi] Turing Tournament > Hey, look what my alma mater is up to. The Humanities and Social Sciences > department, no less. Although it was common for undergrads to be in economics > experiments, and this 'test' looks pretty similar. No hard language stuff. > > http://turing.ssel.caltech.edu/ > > -xx- Damien X-) > > ------- > To unsubscribe, change your address, or temporarily deactivate your subscription, > please go to http://v2.listbox.com/member/?[EMAIL PROTECTED] > ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED] ------- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to http://v2.listbox.com/member/?[EMAIL PROTECTED]
