 |
|
AAAI 2008
|
|
|
AI Education Colloquium
|
Welcome to the discussion-summary page for the AAAI 2008 AI
Education Colloquium.
Back to the original AI Education colloquium page
Thank you!
for sharing these ideas, both specific and broad-reaching, at the workshop.
Thanks especially to Kiri Wagstaff for the bulk of these notes!
Feel free to email Zach Dodds (dodds@cs.hmc.edu) if you'd like to add anything.
breakout survey form excel survey
results
Quote of the Day (BF Skinner) The real problem is not whether machines think but whether people
do.
- Broader context:
- How can AI education serve more than just the AI community?
- How can we incorporate AI into other, non-AI classes?
- Maybe teaching AI should really be teaching about thinking.
- Effective teaching:
- What are student attitudes about AI, prior to taking an AI class... especially negative ones?
"Studying AI won't help me get a job; AI will take jobs away from people" and so on
- Students want something tangible and concrete -- to experience AI, not just learn about it.
- How can we effectively share our advances in teaching methods/tools/assignments?
And if we do share, how can we effectively avoid students cheating?
- Resources:
- Motivation:
- What feedback is there from industry about AI-related skills they want students to have,
and how can this inform
course design as well as show students that AI skills can translate into job offers?
- AI disillusionment is a real factor: AI initially seems exciting, but then when you start
studying it, it comes
down to non-exciting mechanics and/or only works on toy problems.
- Many students come to AI motivated by other fields, like game development;
could capitalize on this to keep
energy and excitement going.
- Challenges:
- articulating why AI is important - answering "why do students need AI?"
- articulating the subset of AI that is fundamental "what AI should all students know?"
- balancing breadth vs. depth and finding an appropriate level of abstraction at which student can engage
- connecting AI course content with current research work and topics
- creating an AI-specific "nifty assignments" resource similar to SIGCSE's / Nick Parlante's
http://nifty.stanford.edu/
- Specific tips. suggestions, or insights:
- Get students involved in assignment design, especially incremental projects with multiple stages.
- Play-act important algorithms, e.g., backpropagation; act out an original (gender-based) turing test.
- using on-line resources, e.g., for 20 questions http://www.20q.net/ or
for chatbots
http://www.chatbots.org/, for Eliza (on-line or via emacs)
- have a blindfolded student try to determine the "categories" of objects pulled from a bag;
the class
provides the reinforcement: using color as one of the categories illustrates
the importance of feature selection!
- competitions can be excellent motivators
- real-world examples worth considering
- spam filtering
- information rettrieval
- credit-card scanning
- iris recognition
- A* used for character navigation in video games
- John Conway's game of life via the excellent GOLLY simulator: http://golly.sourceforge.net/
- "AI in the real world" class discussions using NYTimes articles.
These ccan illustrate cutting-edge work or misperceptions about AI
- Use peer-reviewing techniques as a deliverable in order to engage students in assigned papers or articles
- Many small assignments with one term-long course capstone works well
- The positive initial reaction to physical embodiments of AI ideas (for example, robots or vision systems)
can quickly
turn quite negative if the embodiment is harder to work with
or hides the AI topic it means to reinforce.
- Provide data structures and, in general, a coding framework so that the AI concepts stay at the fore
- For machine learning, Weka http://en.wikipedia.org/wiki/Weka_(machine_learning) is an
excellent example of such a framework