What belongs in a low-math undergraduate AI elective besides ML?

Question

The primarily undergraduate institution at which I teach does not have an Artificial Intelligence course. We do have classes on Data Analysis and Machine Learning:

DATA 150: Introduction to Data Analysis (4 Credits)

Data analysis is the extraction of knowledge and insights from complex data. This course introduces the concepts, issues, and techniques of data analysis. Topics include data cleaning and preparation, feature selection, association rules, classification, clustering, evaluation and validation. Tools implemented in R and Python will be used to explore data sets using these techniques.

CS 141: Machine Learning (4 Credits)

This course provides a broad introduction to machine learning and statistical pattern recognition including both supervised and unsupervised learning from a computational perspective. Topics include generative/discriminative learning, parametric/non-parametric learning, neural networks, support vector machines, clustering, dimensionality reduction, and kernel methods. Additional topics as time allows.

It's been a long time since I was in school, and even then my knowledge of AI was pretty eclectic. What belongs in an undergraduate AI course that is not covered by the above two courses?

A constraint is that we do not require Calculus in our program, and most students have only had discrete mathematics. (While the above classes nominally have Linear Algebra and/or Calculus-based Probability and Statistics as prerequisites, we sometimes waive those requirements.)

Answer 1

I took both the AI and (basic) ML undergrad courses at Princeton and currently teach an AI elective at the HS level for some very bright students. I've seen some good online material from Berkeley (CS188) and MIT (6.034), and of course Stanford's ML lecture series is amazing. Russell & Norvig's textbook is basically the bible for these sorts of courses.

From what I've seen, AI undergraduate courses have changed a lot in the past decade to focus on ML, since much of the recent breakthroughs and publicity are in those areas. AI is a very broad and historically rich field beyond just Data Analysis or ML. I would say a comprehensive AI course would start from Alan Turing's classic (and often misused) Turing Test and ground itself in an ongoing definition of rationality as effective problem-solving.

From there, there are lot of "classical" AI fields that are very approachable without any calculus; most of their discrete algorithms are intuitive, and you can wave your hands and say the continuous (calculus-involving) counterparts are basically the same idea. Here's my shortlist of essential first topics:

• Goal-based state-space searching algorithms (e.g. DFS, BFS, Greedy, A*). These are a great start since the notion of searching is very universal and lays the groundwork for the next two:
• Adversarial Game-playing AI (minimax and alpha-beta pruning)
• Constraint-Satisfaction Problems (ARC-3, backtracking search) and/or Boolean SAT (DPLL, WalkSAT).

These three areas form my HS curriculum, and provide really rich applications and philosophical fodder ("is this how rational humans think? Is it intrinsically better, or worse, or completely unrelated?")

Some courses might go from these topics into Bayes Nets and/or Hidden Markov Models, though I think this leans math heavy since its all about probability.

Usually by this point AI courses have become more or less completely focused on ML, often starting with k-means clustering, basic classification/regression using perceptrons or decision trees, ensemble learning/boosting, reinforcement learning (usually in the framework of Markov Decision Processes), and maybe deeper neural networks.

EDIT: A follow-up; my course has now changed its third unit topic from CSPs/SAT to Markov Decision Processes (Value Iteration and Policy Iteration) and Reinforcement Learning. Its more exciting, leads into ML, and flows out the Adversarial Games nicely because of the shared concept of finding a "policy", especially if Expectimax is covered.