 Today we are going to discuss on the topic AI problem characteristics. At the end of this session student will be able to demonstrate the key dimensions of problems to use heuristic search. Heuristics means a technique that is going to help the search to be made more efficient. A heuristic function defines a number which leads us closer to our particular goal from the start state. We encompass a variety of specific techniques particularly effective in a small class of problems. We analyze the problem to select most appropriate methods and finally we use key dimensions that a problem exists to define the control strategy. Find out now what makes a problem a candidate for artificial intelligence. The problem key dimensions which we are talking about which will be using heuristics at different levels are is the problem decomposable into easier sub-problems. Can solution steps be ignored or at least undone if unwise? Is the problem universal predictable? Is a good solution obvious? Is a desired solution a state or a path to a state? Is the large amount of knowledge required absolutely or only a constraint for the particular search? Can the problem give a problem solution or interaction of human is required? When we consider if the problem is decomposable the characteristics to be tested are break the large problem into smaller problems and try to solve each of them with few specific rules so that you get a solution for every sub-problem which finally relates to getting a solution to the whole problem. An example of this might be an integration problem of calculus where the integration involves n number of terms and each term can be integrated separately after we test this and we find out the term can be integrated we then see or breaking the problem into sub-problems which are solved separately and then we integrate all this to get our particular result. The second example is solving the problem in a blocks world where we have a start state and a final state. We look at the start state and predict how the different blocks have to be placed and then finally step by step after solving each stage we go to solve the full problem. The second key aspect is can a solution step be ignored or undone? We usually see that there are two types of problems ignorable in which certain solution steps can be ignored because they have been inbuilt in higher solutions. For example we will use this in theorem proving recoverables problems are those in which the steps can be undone. For example in eight puzzle if we take a particular step and we are wrong or we detect that we are not reaching a solution we may backtrack and go till we come to a particular stage where we take another path irrevocable is in which the solution cannot be undone for example if we take a move in chess we cannot go back we have to go forward to get a particular solution which might never lead us to a particular goal. Therefore these problems have certain characteristics ignorable problems are simple structure and never backtrack. Recoverable problems are backtracking to recover mistakes irrevocable problems need planning since more effort is required. The next key aspect is is the universe predictable do we know exactly what can be the consequence of a happening possible to a plan an entire sequence of moves that start from the start to the goal. We embedded it into our control structure that allows backtracking wherever it will be necessary for certain outcomes for example in the eight puzzle or certain outcomes that are in the bridge game. Perfect predictions detected in the open loop approach where certain outcome problems are computationally expensive because they require repetition of computations. The next key aspect is a good solution absolute or relative. For classes we see that any path can be getting to us a goal state or the best path which we select to avoid time elapsed as well as getting best solutions. The characteristics that we have to see here are any problem gives us easier solutions any path problem are solved in reasonable amount of time but good path problems are more suggested because they have better heuristics associated with them and the best path problem would have a possible miss of the best solution and we have to do some exhaustive search to get to the particular goal. The next key aspect is a solution or a state of the path. If a solution is existing it might be a state for example if we take a problem of natural language processing there we see that there exist different states like a banker is eating some noodles in a restaurant that banker might refer to the particular bank or it might refer to the bank of the particular river which he is owning and therefore we have two states associated with it and hence this problem is better to be represented by a solution that is a state but in a two jug problem we go for the consequences of operations. Therefore problems are not expressed as states alone but the path that leads us to the goal state. Problem definitions take care of this problems state corresponds to the situation in the world and not sequences of operations. It may become necessary to record the path of the problem process as it proceeds. The next key aspect is is a large amount of knowledge required absolute absolutely or only to constrain the search. Some problems as we know require very less knowledge for example in case of search we just require the rules to determine legal moves and simply control mechanisms to implement the appropriate search. Some problems do require huge amount of knowledge for example a newspaper story that has to be understood and implemented and interpretation is of importance. This requires a variety of facts and complexity constructs. The next key is can computer give a problem solution or interaction of the human is required. Some problems which we have discussed have the problem is in and the solution is out. When computers are to have immediate interaction with the user where information from the user is required is the other domain of problems where we use resolution as a solution aspect. The solutions might be of two types solitary where the computer gives the problem description and produces answer with no immediate communication and demand of reasoning and therefore the human is not involved. There is additional input to the program where additional reassurance to the process may be given by a human if required and conversational there is an intermediate conversation between the person and the computer for assisting the computer to give additional information. As our differences we have used these. Thank you.