Course Description
EEE 4872 Artificial Intelligence is a broad survey of AI techniques taught from an engineering standpoint. Where a machine learning course concentrates on learning from data, this one covers the wider field: search and problem solving, knowledge representation, learning, natural language processing, robotics and computer vision — and, notably, the limits of each.
The Statewide Course Numbering System describes it as covering “basic artificial intelligence (AI) techniques of search, machine learning, natural language processing, robotics, and image processing. Potential/current limitations are analyzed; human interaction in a decision-making environment.”
⚠ The last two clauses of that description are unusual and worth taking seriously. Most course descriptions list capabilities; this one commits to analysing limitations and to examining human interaction in a decision-making environment. Those are not decorative additions — they describe a course that treats AI as an engineering technology with failure modes and human consequences, rather than as a catalogue of methods. In a period of rapid and often overstated claims about AI, that framing is a genuine strength.
Two Florida institutions carry it, both at 3 credits and both under the identical title Artificial Intelligence: Florida A&M University and Florida State University, sharing the joint FAMU–FSU College of Engineering. FSU describes it as instructing “students in basic artificial intelligence techniques of search, machine learning, natural language processing, robotics, and image processing,” with prerequisites of COP 4530 and EEL 4021. The two institutions and the statewide record agree exactly — an unusually clean case with no divergence to warn about.
Learning Outcomes
Required Outcomes
- Formulate a problem as a state space with states, actions, transitions and a goal test, and explain why the formulation determines what search can achieve.
- Implement and analyse uninformed search strategies — breadth-first, depth-first, uniform-cost, iterative deepening — and compare them on completeness, optimality, time and space complexity.
- Implement informed search using heuristics, apply A* search, and explain the conditions of admissibility and consistency under which A* is optimal.
- Design heuristics for a given problem and evaluate their informedness and computational cost.
- Apply local search methods — hill climbing, simulated annealing, genetic algorithms — and explain when they are preferable to systematic search.
- Apply adversarial search: minimax, alpha–beta pruning, and evaluation functions for game playing.
- Formulate and solve constraint satisfaction problems using backtracking, constraint propagation and heuristic ordering.
- Represent knowledge in propositional and first-order logic and perform inference by resolution.
- Reason under uncertainty using probability, Bayes’ rule and Bayesian networks.
- Apply supervised and unsupervised machine learning methods and evaluate them with an appropriate validation methodology.
- Describe the natural language processing pipeline — tokenisation, parsing, semantics — and the statistical and neural approaches to it.
- Describe robotics fundamentals relevant to AI: sensing, localisation, mapping, path planning and control architectures.
- Apply image processing and computer vision fundamentals: filtering, feature detection, segmentation and recognition.
- Analyse the limitations of AI techniques — computational, representational, data-related and ethical — and identify where a method will fail.
- Analyse the role of the human in an AI-supported decision environment, including automation bias, the handover problem and appropriate levels of autonomy.
Optional Outcomes
- Apply automated planning methods including STRIPS-style representations.
- Apply reinforcement learning: Markov decision processes, value iteration, Q-learning.
- Apply deep learning to vision and language tasks.
- Analyse multi-agent systems and distributed decision making.
- Apply expert systems and rule-based reasoning.
- Analyse explainability and interpretability methods for opaque models.
- Analyse AI safety, alignment and adversarial robustness.
- Complete a project implementing an AI system for a defined problem.
Major Topics
Required Topics
- Foundations — what constitutes an intelligent agent, agent architectures (reflex, model-based, goal-based, utility-based), environment properties (observability, determinism, dynamism), rationality, and a brief history including the reasons for previous AI winters.
- Problem solving by search — state space formulation, the search tree and graph, and the trade-offs among breadth-first, depth-first, uniform-cost, depth-limited and iterative-deepening strategies.
- Informed search — greedy best-first search, A* and its optimality proof, admissible and consistent heuristics, heuristic construction by relaxation, and memory-bounded variants.
- Local search and optimisation — hill climbing and its failure modes, simulated annealing, beam search, genetic algorithms, and continuous-space optimisation.
- Adversarial search — game trees, minimax, alpha–beta pruning and its ordering sensitivity, evaluation functions, cutoff and quiescence, and stochastic games.
- Constraint satisfaction — variables, domains and constraints; backtracking search; forward checking and arc consistency; variable and value ordering heuristics.
- Knowledge and reasoning — propositional logic, inference and satisfiability; first-order logic, unification and resolution; forward and backward chaining; knowledge representation and ontologies.
- Uncertainty — probability as a representation of belief, Bayes’ rule, conditional independence, Bayesian networks and inference in them, and utility and decision theory.
- Machine learning — supervised learning (decision trees, linear and logistic models, nearest neighbours, support vector machines, neural networks), unsupervised learning and clustering, overfitting and validation, and evaluation metrics.
- Natural language processing — tokenisation and morphology, n-gram and neural language models, syntactic parsing, semantics, word embeddings, and an overview of transformer-based models and what they do and do not represent.
- Robotics — sensors and perception, configuration space, localisation and mapping (including SLAM at an introductory level), path and motion planning, and reactive versus deliberative architectures.
- Image processing and vision — image representation, filtering and convolution, edge and corner detection, segmentation, feature descriptors, object recognition, and convolutional networks.
- Limitations — computational complexity and intractability, the brittleness of representations, data dependence and distribution shift, adversarial examples, bias in training data, opacity, and the gap between benchmark performance and deployed performance.
- Human interaction in decision environments — human-in-the-loop and human-on-the-loop designs, levels of autonomy, automation bias and complacency, the handover problem, trust calibration, and accountability for automated decisions.
Optional Topics
- Automated planning and scheduling; STRIPS and PDDL representations.
- Reinforcement learning: MDPs, value and policy iteration, Q-learning, and deep reinforcement learning.
- Deep learning architectures for vision and sequence tasks.
- Multi-agent systems, coordination, auctions and game-theoretic reasoning.
- Expert and rule-based systems; fuzzy inference.
- Explainability and interpretability: LIME, SHAP, saliency methods and their own limitations.
- AI safety, robustness to adversarial input, and alignment.
- A substantial implementation project.
Resources & Tools
- Artificial Intelligence: A Modern Approach (Russell and Norvig) is the field’s standard text by a very wide margin, and its structure matches this course’s topic list almost exactly — search, logic, uncertainty, learning, language, vision, robotics, and a substantial treatment of limitations and ethics. Assume this is the text unless told otherwise.
- Artificial Intelligence: Foundations of Computational Agents (Poole and Mackworth) is free online and is a good alternative; Artificial Intelligence: A Systems Approach is a more compact engineering-oriented option.
- For the component areas: Speech and Language Processing (Jurafsky and Martin, free online draft) for NLP; Probabilistic Robotics (Thrun, Burgard and Fox) for robotics; Computer Vision: Algorithms and Applications (Szeliski, free online) for vision.
- Python is the working language: scikit-learn, NumPy and pandas for learning; NLTK or spaCy for language; OpenCV for vision; PyTorch or TensorFlow for neural networks; NetworkX for graph search. The AIMA project publishes free Python implementations of the textbook’s algorithms, which are genuinely useful for checking your own.
- Robotics — ROS (Robot Operating System) and the Gazebo simulator where the course includes a robotics component.
- Environments — Jupyter notebooks and Google Colab (free GPU access, which matters for the neural network material).
- Reference material — AAAI, NeurIPS, ICML and IJCAI proceedings, all freely available; the Stanford AI Index Report for evidence-based data on AI capability and deployment, which is a useful antidote to both hype and dismissal.
Career Pathways
- Machine learning engineer and AI engineer (SOC 15-2051, Data Scientists; SOC 15-1252, Software Developers).
- Robotics and autonomy engineer — perception, planning and control; a substantial employment area in Florida’s defence and space sectors.
- Computer vision engineer; natural language processing engineer.
- Decision support and human–automation systems engineer — the area this course’s human-interaction content points at directly, and one that is genuinely under-served: many organisations can build a model and few can design the system in which a person acts on it.
- Simulation and training engineer — Orlando is a national centre for modelling, simulation and training, and intelligent agent behaviour is core to that work.
- Systems engineer for autonomous platforms; test and evaluation engineer for AI-enabled systems, an emerging specialisation as procurement begins to demand evidence of AI system performance.
- Florida employers include Lockheed Martin (Orlando — autonomy, simulation and training, and one of the largest employers of AI-capable engineers in the state), L3Harris (Melbourne and Palm Bay — autonomy, target recognition and signal classification), Northrop Grumman (Melbourne), Leidos, SAIC and Booz Allen, the Central Florida Research Park simulation cluster, the US Army Combat Capabilities Development Command Soldier Center – Orlando and Naval Air Warfare Center Training Systems Division (both in Orlando), NSWC Panama City for unmanned maritime systems, and NextEra Energy for grid optimisation.
- ⚠ A realistic note on competition. AI is a crowded field at entry level, with many applicants holding computer science or statistics graduate degrees. An electrical engineering graduate competes best where AI meets physical systems — robotics, sensing, signal classification, control — because the domain knowledge is what generic candidates lack. This course plus a signals or controls background is a stronger position than this course alone.
Special Information
⚠ The prerequisites tell you what kind of course this is
FSU requires COP 4530 (data structures and algorithms) and EEL 4021; the statewide record lists the same pair.
The data structures requirement is the substantive one, and it is a direct signal about the first half of the course. Search — the opening and largest block of material — is graph algorithms: frontiers are priority queues, explored sets are hash tables, and the analysis is in terms of branching factor, depth, time and space complexity. A student weak in data structures and algorithmic analysis will find the search material hard for reasons that have nothing to do with artificial intelligence.
⚠ Programming fluency is assumed rather than taught. You will implement search algorithms, constraint solvers and learning methods. Students whose programming experience is limited to short numerical scripts should build fluency — particularly with recursive algorithms and with the standard library’s data structures — before the term.
Comfort with probability is also assumed for the uncertainty and learning material, and with linear algebra for the learning and vision material, even where neither is named in the prerequisite list.
⚠ Course-code variation across Florida
Artificial intelligence is taught widely in Florida, and almost entirely outside the EEE prefix:
- EEE 4872 — FAMU and FSU, in the electrical and computer engineering department.
- CAP 4630 and related CAP numbers — the standard computer science AI course at most Florida institutions.
- CAI prefix courses — a newer prefix specifically for artificial intelligence, increasingly used as institutions build AI programmes.
- EEE 4773 Fundamentals of Machine Learning — the narrower, learning-focused companion at FAMU, FSU and UF.
- EEE 6772 Foundations of Intelligent Systems — a graduate course at UWF, in a Department of Intelligent Systems and Robotics.
⚠ This is one of the clearest prefix-divergence cases in the EEE prefix. A student who took AI under a CAP or CAI number has covered substantially the same textbook — Russell and Norvig is near-universal — but a receiving programme naming EEE 4872 will not match it automatically, and vice versa. Credit will transfer; the requirement match may not. Read target-programme prerequisites literally, carry the syllabus, and raise substitutions early.
⚠ How this course differs from EEE 4773 Machine Learning
Both are offered at FAMU and FSU and students sometimes take one expecting the other:
| EEE 4872 Artificial Intelligence | EEE 4773 Machine Learning |
|---|
| Broad survey: search, logic, uncertainty, learning, NLP, robotics, vision | Deep in one area: learning from data |
| Prerequisite: data structures | Prerequisite: signals and systems, linear algebra |
| Algorithmic and representational emphasis | Statistical and mathematical emphasis |
| Includes limitations and human factors explicitly | Includes model evaluation and validation in depth |
They complement each other well, and taking both is a coherent plan. If you can take only one: choose EEE 4773 if you want to work with data and models, and EEE 4872 if you want to build systems that reason, plan and act — robotics and autonomy in particular.
Position in the curriculum
EEE 4872 is a senior-level elective following data structures. It pairs naturally with controls and robotics coursework, and with EEE 4773. It is well suited to supporting a senior design project involving autonomy, perception or decision support.
Difficulty and time commitment
The characteristic difficulty of this course is breadth. It covers in one term what several specialised courses cover separately, and each area has its own vocabulary and its own way of thinking — logic is not probability, search is not learning, and vision is not language. Students who look for a single unifying framework do not find one, and that is not a failure of the course; AI is genuinely a collection of methods rather than a single theory.
The implementation load is substantial: expect to write real programs, not to work problems on paper. Plan on nine to eleven hours a week, more where a term project is set.
Practical advice: implement the algorithms yourself rather than importing them. A hand-written A* search on a small problem teaches more than a library call on a large one, and the search block early in the term is where the habits for the rest of the course are set.
Articulation and transfer
SCNS records EEE 4872 as guaranteed to transfer to an institution offering the same course. Two Florida institutions carry it, both at 3 credits. The course is upper-division and carries no general-education or Gordon Rule designation.
FE exam relevance
The NCEES Fundamentals of Engineering (Electrical and Computer) exam covers algorithms, data structures and probability, which this course applies, but does not test artificial intelligence. This is an elective taken for its own value and for employability.
AI Integration
This is a course about artificial intelligence, so the reflexive question — may I use an AI tool? — has an unusually pointed answer: you are studying the thing you would be using, and that is the best reason to use it carefully.
The course’s own description makes this a first-class topic. The statewide entry commits to analysing “potential/current limitations” and “human interaction in a decision-making environment.” Large language models are a live case study in exactly those two things, and a student in this course is better placed than most to evaluate them.
Where a general-purpose assistant genuinely helps. Explaining why A* is optimal under an admissible heuristic; walking through a resolution proof; generating boilerplate for search and constraint frameworks; debugging recursive implementations; explaining an unfamiliar method from a paper; and summarising the state of an area before you read into it.
⚠ Where it fails, and why the failure is precisely this course’s subject. Three failures, and each is one the course explicitly teaches you to see.
First: a language model will produce a confident answer where its representation does not support one. Asked to run a search, evaluate a game position or perform multi-step logical inference, it produces text that resembles a solution rather than the result of a computation — frequently correct on small examples and wrong on larger ones, with no change in confidence. The distinction between a system that computes an answer and one that generates a plausible-looking answer is a central topic of this course, and the tool in your hand is the cleanest available demonstration of it.
Second: models overstate the state of the field. They will describe benchmark results as though they were deployed capabilities, blur the gap between demonstration and product, and reproduce claims from press material rather than from evaluation. The course’s explicit “limitations” content exists to counter exactly this, and the Stanford AI Index and the primary literature are the correctives.
Third, and most relevant to the human-interaction material: automation bias. The tendency of a person to accept an automated recommendation over their own judgement, particularly under time pressure, is a documented and well-studied phenomenon — and it is on this course’s syllabus. A student who uncritically accepts generated answers while studying automation bias is demonstrating the phenomenon rather than learning it. That is worth noticing, not as a rebuke but because self-observation is genuinely useful here: the pull toward accepting a fluent answer is exactly what the literature predicts, and feeling it is instructive.
The engineer’s responsibility. AI systems increasingly make or shape decisions that affect people — in hiring, credit, medicine, policing and weapons. The engineer who builds such a system is accountable for its failure modes, its bias, and the conditions under which a human can meaningfully override it. This course’s inclusion of limitations and human-interaction content is not a soft addendum to the technical material; it is the part that determines whether the technical material is deployed responsibly. The habit worth forming is to ask of any AI system, including the one helping you study: what does it not know, how would I detect that it is wrong, and who is accountable if it is?
Academic integrity. FAMU and FSU both maintain academic honour policies covering AI-generated work. Algorithm implementations, analysis and project work are normally expected to be your own even where explanation assistance is permitted — and in a course where the assessed skill is understanding how these systems work, outsourcing the work to one is straightforwardly self-defeating. Ask your instructor what is permitted, and disclose its use where the syllabus requires it.