University of the People CS 4408 Artificial Intelligence Term 5, 2018-2019 Assignment 6 Please Note that I referenced a lot of my thoughts from exercises 6 AIspace.org: Complete exercise 8.A. As part of your assignment, provide a response IN YOUR OWN WORDS to the following questions. (Assignment part A) 1. The STRIPS representation for an action consists of what? o They consist of preconditions and effects. Preconditions consist of a set of variable value assignments that must be true for an action to occur. Effects are a set of variables whose values change as a result of the action. 2. What is the STRIPS assumption? o This assumption states that the values of the variables not mentioned in the action description remain unchanged after the action is carried out. 3. What is the frame problem in planning? How does it relate to the STRIPS assumption? o The Frame problem is the difficult of properly representing all things that remain unchanged in STRIPS. The frame problem seems to be very important to the STRIPS assumption. Since the STRIPS assumption states that the values of the variables not mentioned in the action description remain unchanged after the action is carried out and the frame problem is having difficulties representing unchanged elements then these two concepts go hand in hand. 4. What are some key limitations of STRIPS? o STRIPS does not support quality or disjunctions. State cases can only come in the form of positive literal conjunctions and goals are straight conjunctions. Complete exercise 8.B. As part of your assignment, provide a response IN YOUR OWN WORDS to the following questions. (Assignment part B) 1. What is meant by the horizon in a planning problem? o The horizon refers to the number of time steps required for the problem to be rolled out. 2. What are initial state constraints in a CSP problem? o The state variable seems to be constrained before any action has been implemented. So the constrain occurs at time 0. 3. What are goal constraints? o The goal seems to be constrained at an unknown time (x). Since the goal will be constrained at the end of the problem, then the goal is constrained at the horizon which is (x). 4. What are precondition constraints? o These are constraints found between the state variables at certain time (t1) and actions that occur at another time (t2). These constraints are needed and must be true for the action(s) to occur. 5. What are effect constraints? o This is when an action affects a state variable and as a result, the state variable’s value changes. Complete exercise 8.C. As part of your assignment, you must provide a reflective statement that documents your findings about the various search approaches to include your observations, which forms of search were more efficient and which less efficient for the Vancouver neighborhood graph and why? You statement must be at least 200 words in length. Statements under this length will receive less points. (Assignment part C) So the exercise in 8C focuses on STRIPS Representation and CSPs. Loading up the test program was fairly simple and executing the test program was also easy. This is what I have learned about STRIPS Representation so far. The STRIPS representation is used to determine the truth values of primitive predicates based on the previous state and the action. This is derived from the idea that most predicates are usually unaffected by a single action. Everything in STRIPS is time based. This begins with the states and proceeds all the way to the goal. The goal in actually serves two functions. The first one is to see whether and action can be carried out in the state and secondly, to see what is true in the resulting state. In order to better facilitate this, each state is given a value and these value are affected by events called actions. An action is a specific event, which requires constraints to be true for it to occur. An action will declare which state variables would be affected when it occurs. The state variables that are not declared in the action description remain unchanged. When an action occurs and state variables change, this is called an effect. I don’t really think more detail need to be added on that since it is selfexplanatory. Complete exercise 9.A. As part of your assignment, provide a response IN YOUR OWN WORDS to the following questions. (Assignment part D) 1. What is meant by a one-off decision? How can this be applied in the delivery robot example? o With one-off decisions, the agent in question uses its knowledge of available actions and preferred goal outcomes to make its choice before any action is made. To apply this to the delivery robot example, the robot can make a decisions to conserve power or take the longest or shortest route to before it starts moving. 2. Define utility in a decision problem. o This is basically the agent’s preference to its world environment in terms of desirability out of all the possible words available to the agent. 3. How do we calculate the expected utility of a decision? o The expected utility can be calculated by summing over the possible worlds that meet a certain decision requirement (Pw) for each available world (Uw). This results in (Uw * Pw). 4. How do we compute an optimal one-off decision? o First we calculate the expected utility for each decision per last question and we then choose the best overall decision that will maximize our expected utility. 5. What are the three types of nodes in a single-stage decision network? o The three types of nodes are: i. Decision Nodes ii. Random Variable Nodes (Chance Nodes) iii. Utility Nodes 6. What is a policy for a single-stage decision network? What is an optimal policy? o A policy for a single-stage decision network is each decision variable must be assigned a value. The optimal policy is one where the expected utility is best. 7. Describe the variable elimination steps for finding an optimal policy for a single-stage decision network. o The steps for finding the optimal policy are: i. Prune all non-ancestor nodes of the utility node. ii. Sum all chance nodes. iii. Once that is done, a single factor representing the expected utility for each combination of decision variable will be left. Let’s call that (H) iv. Find the maximum value is (H), let’s call that s and return s. Complete exercise 9.B. As part of your assignment, provide a response IN YOUR OWN WORDS to the following questions. (Assignment part E) 1. How is a sequential decision problem different from a one-off decision problem? o With a one-off decision, whether it be a single decision or multiple decision in a single macro decision, the decision is still made before an action is carried out. With sequential decisions, the agent basically uses cause and effect. The agent will observe then carry out and action based on its observations that it will observe the outcome of the action and repeat until it gets to the goal. Basically many sequential decision are made. 2. What types of variables are contained in a decision network? o The types of variables are: i. Decision Nodes ii. Random Variable Nodes (Chance Nodes) iii. Utility Nodes 3. What can arcs represent in a decision network? Relate this to the types of variables in the previous question. o What can arcs represent: i. In Decision Nodes, arcs show the information that will be available when the decision is made. ii. In Random Variable Nodes (Chance Nodes), arcs show probabilistic dependence. iii. In Utility Nodes, arcs show what the utility depends on. 4. What is a no-forgetting decision network? o This means that information that was used for an earlier decision is available for later decisions. 5. Define decision function and policy. o A decision function specifics a value for the decision variable e for each assignment of values to its parents. A policy is each decision variable must have a decision function. 6. A possible world specifies a value for every random variable and decision variable. Given a policy and a possible world, how do we know if the possible world satisfies the policy? o The possible world can only meet the policy requirements if the value for each decision variable in that possible world is the same as the value in the decision function for the corresponding decision variable in the policy. 7. To find an optimal policy, do we need to enumerate all of the policies? Why or why not? o No we do not need to enumerate the policies because we can use variable elimination instead. It’s easier. Reference(s) Poole, D., & Mackworth, A. (2017, January 1). Artificial Intelligence. Retrieved August 5, 2019, from https://artint.info/html/ArtInt_36.html