CSCI 151 “Artificial Intelligence” is an advanced elective course in computer science. Its goals are twofold: to introduce students to fundamental techniques of AI work, and to problematize popular notions of AI in order to cultivate a critical technical practice. The main takeaway is that AI is not a well-defined term because it is not at all clear what “intelligence” means. Instead, this course focuses on automated decision-making and explores how computers can come to make several different kinds of decisions. Even this gloss is somewhat awkward (in the end, if humans act on the determinations of an automated decision-making system, the humans themselves are the ones making the decision), but it is an improvement over “AI.”
Python is somewhat notorious for having an awkward time of managing different OS, language, and package versions.
In this course, I will encourage the use of conda to manage the Python installation and associated packages; I cannot provide technical support for other installation methods.
We are trying something new with the assignments in this course: they are all motivated by, and many take place within, Minecraft. You do not need to own a copy of Minecraft to execute the programming assignments; a special version is made available via Microsoft Research as Project Malmo.
This is a project-based course in AI with a strong analytical component. Homework projects comprise the majority of the grade in the course (60%). Because analytic reasoning depends on a strong grasp of the fundamentals, we will frequently practice remembering material encountered earlier; every class will start with a quiz, and these quizzes will sum up to 15% of the grade. The final project makes up the remainder of the grade for the course, and this project has both analytic and programming components.
- Homework (60%, 10% per assignment)
- Quizzes and reading reactions (15%)
- Final Project (25%)
Homework projects are the cornerstone of the course. The homeworks consist of both analytic work (typed reflections) and implementation work (Python code)…
To sum up, weighting for the homeworks is as follows:
- 25% Reasonable attempt at completing the assignment by the due date (judged by my visual inspection)
- 25% Effortful and meaningful reflection by the due date
- 25% Automatic grader points before end of semester
- 25% Substantive peer review of another student’s program (within four days after due date)
…
The final project will consist of a significant implementation effort coupled with a strong analytic or critical justification for the project and a class presentation. Final projects must be executed in teams of up to five people.
I expect to see final projects in three main categories but I am open to others:
- A tutorial introduction, in the Minecraft domain, for an AI technique which was not covered by previous homework projects; this could serve as homework for a future iteration of the course.
- A project leveraging an AI technique for a different domain, where either the given application is not usually tackled with that technique or the scope and execution are significant or novel.
- A significant extension of a homework project, either using new techniques or into new domains.
…
All projects are due five minutes before midnight on the due date. The only exceptions are the project post-mortems which are due on 12-19 at 14:00.
(Instructors: Please feel free to rearrange these topics as you like. Some of the “reflecting on” topics can be merged into the preceding day’s lecture.)
| Topic | Subject | Notes | |
|---|---|---|---|
| 1 | Introduction and Search | Students can start on assignment 1 | |
| 2 | Informed Search | At this point all the topics of assignment 1 are covered in lecture | |
| 3 | Reflecting on Path Planning | It may be useful to go over (some of) the reflections in lecture before assignment 1’s due date | |
| 4 | Classical Planning | Students can start on assignment 2 | |
| 5 | Iterative Widening | At this point the topics of assignment 2 are covered in lecture | |
| 6 | Constraint Satisfaction | ||
| 7 | Reflecting on Planning | Here you can go over assignment 2 a bit before its due date | |
| 8 | Linear Programming | Students can start on assignment 3 | |
| 9 | Modeling with Linear Programming | Students have all they need for assignment 3 | |
| 10 | Integer Programming | Students can start on assignment 4 | |
| 11 | Reflecting on Linear Programming | You might go over assignment 3 just before it’s due | |
| 12 | Planning via Constraint Satisfaction | Students have what the need for assignment 4 | |
| 13 | Machine Learning | ||
| 14 | Reflecting on Integer Programming | You might go over assignment 4 just before it’s due | |
| 15 | Perceptrons | ||
| 16 | Backpropagation | ||
| 17 | Search by Sampling | ||
| 18 | Reinforcement Learning | Refer to Sutton & Barto chapter 6 | |
| 19 | Deep RL | For the next class, read Reinforcement Learning’s Foundational Flaw and How to Fix Reinforcement Learning | |
| 20 | Logic and Satisfiability | ||
| 21 | Project Proposals | Decide projects and teams by end of day, email me. | |
| 22 | Logical Resolution | ||
| 23 | Logic Programming | ||
| 24 | Convolutional Neural Networks | ||
| 25 | Logic Revisited | ||
| 26 | Recurrent Convolutional Neural Networks | ||
| 27 | Problems and Opportunities in AI | ||
| 28 | RCNNs Revisited |