IT4535 introduces the fundamental problems, theories, and algorithms that define artificial intelligence as a discipline. The course moves beyond popular conceptions of AI to examine the computational foundations: how do you build a system that searches for solutions in large problem spaces, represents knowledge in a form machines can reason about, learns patterns from data, and makes decisions under uncertainty? These questions have driven AI research since the Dartmouth Conference of 1956, and they remain central to every modern application from autonomous vehicles to medical diagnosis.
AI search strategies: comparing algorithmic approaches
| Algorithm | Type | Completeness | Optimality | Best Use Case |
|---|---|---|---|---|
| Breadth-First Search (BFS) | Uninformed | Yes (finite graphs) | Yes (uniform cost) | Finding shortest path in unweighted graphs; small to medium state spaces |
| Depth-First Search (DFS) | Uninformed | No (may loop in infinite spaces) | No | Memory-constrained problems; deep solution paths; backtracking puzzles |
| A* Search | Informed (heuristic) | Yes (admissible heuristic) | Yes (admissible + consistent) | Pathfinding, route planning, puzzle solving with domain knowledge |
| Minimax with Alpha-Beta Pruning | Adversarial | Yes (finite game tree) | Yes (perfect play) | Two-player zero-sum games (chess, tic-tac-toe, checkers) |
| Greedy Best-First Search | Informed (heuristic) | No | No | Quick approximate solutions when optimality is not required |
Knowledge representation and automated reasoning
A central challenge in AI is representing knowledge in a form that allows a computer to reason, infer new facts, and solve problems. IT4535 covers several foundational representation schemes, each suited to different types of knowledge. Propositional logic represents simple facts and their logical relationships (AND, OR, NOT, IF-THEN) but cannot express generalities about categories of objects. First-order predicate logic extends propositional logic with variables, quantifiers (for all, there exists), and predicates, allowing statements like "All mammals are warm-blooded" to be expressed and reasoned about formally. Knowledge graphs and semantic networks represent knowledge as nodes (concepts) connected by labeled edges (relationships), providing an intuitive visual structure that supports inheritance reasoning: if "dog" is connected to "mammal" by an "is-a" link, and "mammal" has the property "warm-blooded," the system can infer that dogs are warm-blooded without that fact being explicitly stated (Russell and Norvig, 2021). Expert systems apply these representation techniques in practice, encoding domain expertise as production rules (IF conditions THEN actions) and using inference engines to chain those rules forward or backward to reach conclusions. The course examines both the power and the limitations of rule-based systems, including their brittleness when encountering situations outside the encoded knowledge and their inability to learn from new data without manual rule updates.
Machine learning represents the modern frontier of AI, addressing the fundamental limitation of hand-coded knowledge systems by enabling computers to improve their performance through experience. IT4535 introduces the three major learning paradigms: supervised learning (learning a mapping from labeled input-output pairs, used for classification and regression), unsupervised learning (discovering patterns in unlabeled data, used for clustering and dimensionality reduction), and reinforcement learning (learning through trial and error by maximizing cumulative reward, used for game playing and robotic control). The course covers neural network foundations, explaining how artificial neurons compute weighted sums of inputs, apply activation functions, and connect in layers to approximate complex functions. Students learn the backpropagation algorithm for training multi-layer networks and explore how deep learning architectures (convolutional networks for image recognition, recurrent networks for sequential data) have driven recent breakthroughs in computer vision, natural language processing, and game playing (Goodfellow et al., 2016). Throughout these technical topics, IT4535 integrates discussions of ethical AI, examining issues of algorithmic bias, transparency, accountability, and the societal impact of deploying AI systems that make consequential decisions about hiring, lending, sentencing, and medical treatment.
Writing a search algorithm analysis, expert system design, or AI ethics paper?
Our AI and computer science writers apply formal reasoning and current research with the depth Capella's IT4535 rubric requires.
Key topics in IT4535
- Search algorithms: breadth-first search, depth-first search, iterative deepening, uniform-cost search, A* search with admissible heuristics, and their time/space complexity tradeoffs
- Game trees and adversarial search: minimax algorithm, alpha-beta pruning, evaluation functions for imperfect real-time decisions, Monte Carlo tree search
- Knowledge representation: propositional logic, first-order predicate logic, semantic networks, knowledge graphs, ontologies, and frames
- Expert systems: production rule systems, forward and backward chaining, inference engines, certainty factors, and the limitations of rule-based approaches
- Machine learning foundations: supervised learning (classification, regression), unsupervised learning (clustering, dimensionality reduction), reinforcement learning (reward signals, policy optimization)
- Neural networks: perceptrons, multi-layer architectures, activation functions (sigmoid, ReLU), backpropagation, gradient descent, overfitting and regularization
- Natural language processing basics: tokenization, parsing, sentiment analysis, bag-of-words models, word embeddings, and the challenges of ambiguity in human language
- Ethical AI: algorithmic bias and fairness, transparency and explainability, accountability frameworks, privacy concerns, and societal impact of autonomous decision-making
AI milestones students should know for IT4535 assessments
- Turing Test (1950): Alan Turing proposed the "imitation game" as a practical test for machine intelligence, asking whether a human judge could distinguish machine responses from human ones in a text conversation. Still debated as a measure of intelligence today
- Expert systems era (1970s-1980s): systems like MYCIN (medical diagnosis) and DENDRAL (chemical analysis) demonstrated practical AI applications using hand-coded rules, but their brittleness and high maintenance costs led to the "AI winter" when expectations outpaced capabilities
- Deep Blue defeats Kasparov (1997): IBM's chess-playing system used minimax search with alpha-beta pruning and custom evaluation functions, processing 200 million positions per second. Demonstrated brute-force search combined with domain knowledge
- Deep learning revolution (2012-present): AlexNet's victory in ImageNet (2012) demonstrated that deep convolutional neural networks trained on GPUs dramatically outperform traditional computer vision methods, triggering massive investment in deep learning across industry and research
- Large language models (2017-present): the transformer architecture (Vaswani et al., 2017) enabled training on massive text corpora, producing models capable of generating coherent text, answering questions, and performing tasks specified through natural language instructions
Get Help With IT4535
Search algorithm analyses, knowledge representation projects, neural network reports, AI ethics papers. Artificial intelligence coursework grounded in theory and application.
Place Your OrderView All ServicesRelated courses
Frequently asked questions
A* search finds the optimal (lowest-cost) path from a start node to a goal node by evaluating nodes using the function f(n) = g(n) + h(n), where g(n) is the actual cost from the start to the current node and h(n) is a heuristic estimate of the cost from the current node to the goal. The heuristic must be admissible, meaning it never overestimates the true remaining cost. When the heuristic is also consistent (the estimated cost from any node to the goal is no greater than the cost of reaching a neighbor plus that neighbor's heuristic estimate), A* is guaranteed to find the optimal solution while expanding the fewest possible nodes among all optimal algorithms using the same heuristic. In practice, choosing a good heuristic is the key design decision: Manhattan distance for grid-based pathfinding, straight-line distance for geographic routing, and domain-specific estimates for specialized problems. IT4535 assignments typically require students to prove heuristic admissibility and compare the performance of different heuristics on the same problem.
Supervised learning trains models on labeled data, where each training example includes both the input features and the correct output (target). The model learns to map inputs to outputs and can then predict outputs for new, unseen inputs. Classification (predicting a category, such as spam or not-spam) and regression (predicting a continuous value, such as house price) are the two main supervised learning tasks. Unsupervised learning works with unlabeled data, where the model must discover structure or patterns on its own. Clustering (grouping similar data points, such as customer segmentation) and dimensionality reduction (compressing high-dimensional data while preserving important structure, such as PCA) are the primary unsupervised tasks. The practical distinction matters for IT4535 because each paradigm suits different problem types: supervised learning requires labeled datasets (expensive to create), while unsupervised learning can work with raw data but provides less precise, harder-to-evaluate results. Many real-world systems combine both, using unsupervised methods for feature extraction and supervised methods for the final prediction.
Backpropagation is the algorithm that computes how much each weight in a neural network contributed to the overall prediction error, allowing the network to adjust its weights to reduce that error. The process works in two phases. In the forward pass, input data flows through the network layer by layer: each neuron computes a weighted sum of its inputs, applies an activation function (such as ReLU or sigmoid), and passes the result to the next layer, ultimately producing a prediction. The loss function then measures how far the prediction is from the correct answer. In the backward pass, the algorithm uses the chain rule of calculus to compute the gradient of the loss with respect to every weight in the network, starting from the output layer and propagating backward through each hidden layer. Gradient descent then updates each weight by a small step in the direction that reduces the loss (scaled by the learning rate). This process repeats across many training examples (epochs) until the network converges on weights that produce accurate predictions. IT4535 expects students to trace through backpropagation calculations on small networks and explain concepts like vanishing gradients and learning rate selection.
AI systems increasingly make decisions that affect people's lives: who gets approved for a loan, which job applicants advance to interviews, what content appears in news feeds, and how autonomous vehicles prioritize safety in unavoidable accidents. These decisions carry ethical weight that cannot be separated from the technical choices that produce them. Training data that reflects historical bias produces models that perpetuate or amplify that bias. Opaque "black box" models make it difficult to explain why a decision was made, undermining accountability and due process. IT4535 addresses these issues because Capella's program outcomes require students to evaluate the societal impact of technology, not just build technically functional systems. Assignments may ask students to analyze a real AI deployment for potential bias, propose fairness metrics, design transparency mechanisms, or evaluate the tradeoffs between model accuracy and interpretability. Understanding frameworks like the EU's AI Act, IEEE's Ethically Aligned Design, and the ACM Code of Ethics gives students a vocabulary for navigating these tensions in professional practice.