|New Mexico Supercomputing Challenge|
Challenge Team Interim Report
The goal of this project is to create a sophisticated computerized artificial intelligence system that enables the computer to think logically and formulate the best strategy for accomplishing a task. The computer will both make "educated guesses" in-simulation based on its likelihood for a successful outcome, and analyze previous attempts post-simulation to determine what strategies could be improved upon in the next simulation run. Eventually the analyzation of previous attempts at success will be conducted in-simulation, so that the computer can successfully formulate the most effective strategy possible in a real time environment. This will empower the computer to think in the same general manner a human does in order to, in most efficient way possible, determine the best method for completing its function. However, the major advantage a computer would have over a human making the same decision would be, of course, that the computer could run through a very detailed thought process and determine an effective strategy much faster than could a human. In order to determine if the finalized "intelligent" AI, is, in fact, making the best decisions, a "control" AI will be run through the simulation as well (one which tries every possible combination of simulation variables and deduces the true best result thusforth; however, this AI would do this much slower and less efficiently than the "intelligent" AI). After the "intelligent" AI is found to work as planned, two "intelligent" AI's will be pitted against each other in the simulation to test their interactions with each other.
The simulation used to test the AI's will be a fairly realistic strategic battlefield environment. A flat map (hopefully to integrate some form of obstacles in the future) will be set up, on which two players (AI's) will be given a base structure (Barracks) in which to produce units to attack the other player. There will be a small variety of units with specialized functions (none clearly better or worse than others), forcing a computer to build some of each type for an assault. In a simulation, two AI's will construct units, move them to strategic defensive positions, and send groups of units along strategic routs to scout for the location of enemy forces or to attack such forces. The simulation would be complete when one player's barracks is destroyed by the other player, who would be the winner of the scenario. The two players that would be pitted against each other in the simulation would be user-selectable from a "preset" player (one for which its strategy was hard-coded and unchanging...this is the standard player for testing the "intelligent" and "control" players against), the "control" player (one which tries every possible combination of strategies sequentially and determines which is the best after all have been tried), and the "intelligent" player. The "intelligent" player will be programmed to act as would a human (but more efficiently and quickly), taking the knowledge of enemy troop positions, time into the simulation, units specialties, previous failures/successes, and other factors, and, using logical paths, decides both simple, immediate decisions and an overall long-term strategy for winning the simulation.
After programming, testing, and refining of the truly "intelligent" artificial intelligence system being created in this project, this final system could blaze the trail for future, more advanced systems that could develop the "best" solutions for any given problem much more quickly and efficiently than could a human (these decisions would often be better than those a human would make, as well, because the computer can base its choices upon many more variables than a human could comprehend). Using such a technology, many tasks could be accomplished by computers alone. This technology could be implemented in unmanned spacecraft, mechanical probes entering areas too dangerous for humans to traverse, and even in machines that would perform household tasks. Such implementations would take input from the real world and logically decide what method of action would be the best to take, just as the "intelligent" player in the simulation described above takes input from its simulated surroundings and determines the best route to success, and just as humans take input from their surroundings and formulate the best route to accomplish their goals.
For questions about the Supercomputing Challenge, a 501(c)3 organization, contact us at: consult1516 @ supercomputingchallenge.org
New Mexico Supercomputing Challenge, Inc.
80 Cascabel Street
Los Alamos, New Mexico 87544