• Unit 5: Problem-Solving Methods in AI

    Problem-solving using AI includes both ML and other paradigms. This unit provides you with a foundation for problem-solving using ML methods, which always requires the availability of data from which to generalize. AI also provides us with other concepts, such as a general problem-solver (GPS), to solve problems more broadly. AI problems come in many flavors, and you can leverage the right algorithm to solve them based on their characteristics.

    Completing this unit should take you approximately 5 hours.

    • Upon successful completion of this unit, you will be able to:

      • discuss how to assess a problem description to extract key machine learning requirements;
      • apply the machine learning models as dictated by problem characteristics to deliver the optimal business outcomes;
      • describe the critical components of an AI-based general problem-solver (GPS);
      • explain how the problem-solving agent relates to different environments;
      • analyze a new problem to determine its environment using the various characteristics and methods for analysis; and
      • explain the computational complexity of general problem-solving agents and how to improve it.

    • 5.1: Integrating ML Skills

      Now, we will look at how to solve new problems based on what you've learned about ML. We will analyze the nature of the available data and how it needs to be processed and conditioned to preserve any underlying mathematical relevance. We will also look at data cleaning, normalization, and other helpful techniques. The section will demonstrate how to identify key features of any ML problem, determine which ML models would be the most relevant, and provide some benchmarks for effectively training and validating models over their lifetimes.

    • 5.2: General AI Problem-Solver Architecture

      General problem-solvers (GPSes) have always captured the imagination of computer scientists. They are useful to understand as one of the models of how to build an intelligent agent for almost any problem. GPS uses the PAGE model: percepts, actions, goals, and environment. In particular, we visualize the problem-solver as starting from its initial state in the environment and, through its actions, eventually achieving its goal. GPS is a simple algorithm that will find a sequence of actions that achieves the goal state, starting from the agent's start state. We will give examples to illustrate how to formulate problems in this style so that a GPS can be executed to solve them.

    • 5.3: Designing a General Problem-Solving Agent

      Different types of problems can be modeled so that they can be solved by applying a simple GPS. The problem with GPS is computational complexity. The algorithm has exponential complexity, making it useless outside of the smallest cases – in larger examples, the algorithm does not perform adequately.