Human-level Artificial Intelligence

EECS 395-22, Spring 2010

Course description

The combination of increasing computational power, off-the-shelf resources, and steady scientific progress in AI and Cognitive Science has lead to a revolution in the kinds of AI systems that can be built. This seminar will explore the state of the art in AI systems that capture larger constellations of human cognitive abilities and the problems that lie ahead in creating human-level AIs. Extensive reading, writing, and discussion will be required, as well as a term project.

Prerequisites: EECS 348 or equivalent AI course, plus at least one other AI course.

Practicalities

Instructor: Ken Forbus
Location: Tech MG28
When: Tuesday-Thursday 3:30pm to 4:50pm

Course Requirements & Mechanics

This course requires significant amounts of reading, thinking, discussing, writing, and presenting. You should typically expect between 24 and 70 pages of reading per week. In addition to the reading, you are expected to do the following:

At least four class presentations plus briefing papers during the quarter. These will summarize and critique the assigned articles.
- Briefing paper: A brief summary of the article (3-4 pages) which also raises discussion points. It should include key figures, algorithms, and/or examples. It must be posted on the class BBoogle site at least 24 hours before the start of class, so that everyone in the class can either download it or print it, as desired.
- Presentation: Presentations should be at least 15, and no more than 20, minutes. The 20 minute limit will be ruthlessly enforced. You must use PowerPoint or some other form of talk presentation software. Like the briefing paper, the presentation must be posted for download 24 hours before the start of class, in either PowerPoint or PDF format. The presentation must include
  - A concise summary of the contents of the paper. The audience should be able to understand from this the basic ideas, methods, and results of the paper. This part of the presentation is factual, not evaluative.
  - Your evaluation of the paper, including both positives and negatives. Rhetorical problems, experimental problems, theoretical problems, are all fair game.
  - A set of discussion questions for the class. These will be used to organize the discussion for the next 20 minutes.
Reaction notes. Before the start of each class, you must turn in via email a brief (equivalent of one page) written reaction to the readings for that class. The idea is to explore the implications of the papers, by comparing and contrasting them, and thinking about what questions they raise, implications, etc.
Term project. Term projects can either delve further into the literature in some area discussed in class, in ways relevant to our focus, or can be computational experiments to explore some of the ideas discussed in class. In both cases, you will be expected to explore additional sources beyond the reading in the class, and synthesize the relevant literature. Term project writeups will be between 15 and 20 pages, excluding source code, if any.
Pop quizzes. Generally unnecessary, since a vibrant discussion where everyone is participating provides ample evidence of understanding. But if it seems that students are not doing the reading, pop quizzes may occur.

The 24 hour deadline for posting materials is crucial: People need to be able to download materials and have them in class. How you choose to bring materials to class is your choice: some like soft-copy, others like paper. However, to minimize our environmental footprint, all materials turned in must be soft-copy -- no dead trees allowed. Similarly, the one page Reaction Notes must be turned in (via email only) before class, so that they represent your own thinking about the materials. Late postings will be penalized.

Syllabus

The syllabus is still under construction. Local copies of the papers are available through Blackboard.

Last edited 3/26/10 by KDF