1.2 Heuristics

1.2 Heuristics

Heuristics, the trail and error of the AI algorithmic world. What I mean is this technique used in gaming AI, will be how the AI learns from its previous experience; and it does this by a simple trial and error rule. An example of this would be if you were playing the first person shooter and if a AI character ran towards you and you were to shoot at the AI, the AI would then trigger that it was under attack and take for cover. Heuristics is where the AI learns from its mistakes or from the situation it is in, it is a trail an error process.

"Heuristics can be said to be estimates of how far the goal state is. Heuristics basically predict how far the goal state maybe or how much it will cost to get to the goal state from a particular node."(2) I can see that Agrawal mentions that the heuristic method is that of a goal state and that its a prediction, I don't feel that in this method of artificial intelligence that it is the heuristics that predict how a goal is met, I feel that the AI using the heuristics algorithm will know exactly what its goal is and it wont be a prediction. An example would be a AI soldier in a game like Battlefield, if the AI is in an attack state then the AI will charge at the playing character but once the AI was under attack from the player character, then the AI would look for cover, somewhere to hide. If the heuristic method is that of a prediction, then we or even the AI would not know what to do when it was under attack. 

What Argrawal does say that I agree with in his paper is how much it will cost to get to a particular goal state. Now what he means is what will it cost the AI to move from one place to another and this is a algorithm know as pathfinding and the preferred method used is the A* technique. I cover the A* algorithm is section 3.1 of my blog (3.1 Pathfinding Concepts - A* Algorithm). To close up this function, there is not much to the heuristic algorithm. It's main priority is that of sending the AI information, is the other algorithm that feed of which helps the AI on its way.

I like to say that there are four features that make up heuristics which make up its advantages and disadvantages... Optimality - this will link to a problem in real time in a game in which a solution will be given to the AI, sometimes this might not always be the best outcome as the heuristic algorithm will find out the best solution, the best solution might not always be the right solution. Completeness - when all the solutions are found with a problem, can we rely on heuristics to find all the solutions and in the end heuristics are only used to find the best outcome. Accuracy - It's important to feel like the heuristic measures each outcome using the best method, is the heuristic giving you the most reliable data? Can you rely on the best outcome given? Execution - Now we know that heuristics are know for finding the best possible solution, can it be executed quickly to an AI? Some heuristics are faster than others and some can take longer, its an algorithm we need to depend on to get the best outcome for our AI.