The Algorithm
In order to find the gold, our agent creates an internal representation of the board using a list of squares. Each square contains information about its position and safety (e.g. whether or not there is a pit or Wumpus).
​
On each move, the agent travels to the safest adjacent square that has not been visited more recently than the others. This makes it so the agent is able to explore unique paths without stumbling onto an unsafe square.
​
Once the agent moves to another square, it reads the state of that square to determine whether it has a breeze, stench, glitter, or gold. We use recursion to effectively reorganize our agent's current information. After the current square's state has been updated, all the adjacent squares' states are updated as well. Then, their adjacent squares are updated too.
Is It Intelligent?
Our agent is capable of making logical decisions and using rational to overcome obstacles and retrieve the gold, so it definitely has the characteristics of a typical artificial intelligence. However, to classify it as "intelligent" might be a bit of a stretch.
​
Looking towards other artificial agents that are built to overcome challenges, we find that there are several that have far exceeded humans in terms of computational speed. For example, the brilliant chess engine AlphaZero has constantly come out on top even against the best players in the world. Our agent can be likened to AlphaZero, as both attempt to carry out their tasks in the most efficient way they know.
​
However, they are also similar in another fashion: they are unable to change, grow, or develop. Both AlphaZero and our agent are limited to the confines of their respective games, and they would be useless in any fundamentally different setting. The lack of adaptation shown leads to the conclusion that they are not truly intelligent, only adept at calculations.
The Results
Overall, the agent was very successful and its logic was sound in deciding the safest paths to take. We were able to successfully clear three maps, and most likely could have completed more if we were aware of one bug.
​
Our main flaw was that our agent did not understand that there would only be one pot of gold on the map; thus, on the fourth map, it did not stop its search after finding the first pot of gold, resulting in an endless loop.
​
Although our agent did very well, it definitely could have been improved. For example, other groups were able to solve maps much quicker than we could. This was most likely due to the fact that our agent was designed to take paths that have been least explored.
​
If we were to redo this project, we might consider some form of advanced memory system, so as to be able to track the agent's path and location. In doing so we would be able to return to the start immediately after finding the gold, instead of further exploring the map.
​
​
​
​
​