Big Debates, Successes, and Failures in Artificial Intelligence

This quarter's 590a seminar is going to be a series of discussions and talks about big debates, successes, and failures in the history of AI. Our primary goal is to get students excited about their work by giving them a historical perspective on current research in AI.

We'll be discussing topics like these:

• What are the big debates in AI (e.g. statistical vs. logical)?
• What is an ambitious project but failed project from the past that we are better suited to tackle now?
• What is behind the hype of things like big data, deep learning, and nonparametric Bayes?

Format

Each week, a disucssion leader will pick a topic and find some good reading material. They'll be responsible for summarizing the topic (10-15 minutes) and moderating a discussion. To keep things informal, the discussion leader should come equipped with questions and talking points, but no slides.

Schedule and Readings

Date	Discussion Leader	Topic
January 9	Tony and Abe	The Chomsky-Norvig Debate: What is the role of statistics in AI? Peter Norvig's On Chomsky and the Two Cultures of Statistical Learning Atlantic interview Noam Chomsky on Where Artificial Intelligence Went Wrong If you want, watch this video, starting at 2:04 A look at the original roots of AI Some other useful links: Leo Breiman's paper Statistical Modeling: The Two Cultures. Norvig talks about Gold's Theorem, which is a theoretical result about learning languages from examples. Here is the paper. Language Identification in the Limit by E Mark Gold, 1967. Chomsky keeps mentioning Charles R. Gallistel and his research agenda. The abstract of this paper summarizes it: Machinery of Cognition by Charles R. Gallistel. Responses on the Language Log, here and here. Some discussion on Reddit here.
January 16	Rob Gens	A nice overview: Deep Learning Architectures for AI by Yoshua Bengio. The "cat face" paper from Jeff Dean and Andrew Ng. Building High-level Features Using Large Scale Unsupervised Learning Some recent press: Optimistic: Scientists See Promise in Deep-Learning Programs Pessimistic: Deep learning takes us, at best, only a small step toward the creation of truly intelligent machines. A blog post talking about why deep belief networks make less sense for language than vision.
January 23	Morgan Dixon	AI vs. HCI punchout Direct manipulation vs. interface agents by Ben Shneiderman and Pattie Maes. Some other readings from Morgan: How Might People Interact with Agents? (Norman, 1994). Compelling Intelligent User Interfaces: How Much AI is Enough? (Horvitz, 1997). Steps To Take Before Intelligent User Interfaces Become Real. (Hook, 2000). Adaptive Interfaces and Agents. (Jameson, 2008).
January 30	Adam Smith	Discovery Systems and the Automation of Science Computational Approaches to Scientific Discovery - Jeff Shrager and Patt Langley from 1990. Why AM and Eurisko Appear to Work - Douglas B. Lenat and John Seeley Brown, 1983. Other readings from Adam: Robot Makes Scientific Discovery All by Itself from Wired. Proposal: maybe human scientists should work more like the robot does. On the formalization and reuse of scientific research.
February 6	Cynthia Matuszek	The Cyc Project Cyc: Toward Programs with Common Sense - Douglas B. Lenat, Ramanathan V Guha, Karen Pittman, Dexter Pratt, and Mary Shepherd. 1990. Cynthia also sent out these links: The 1994 "Cyc Report" by Vaughan Pratt. Knowledge Begets Knowledge: Steps toward Assisted Knowledge Acquisition in Cyc by Michael Witbrock, Cynthia Matuszek, Antoine Brusseau, Robert Kahlert, C. Bruce Fraser, and Douglas Lenat.
February 13	Richard Newcombe	Intelligence Without Representation - Rodney A. Brooks, 1987. More materials from Richard: Why Heideggerian AI Failed and how Fixing it would Require making it more Heideggerian by Hubert L. Dreyfus.
February 20	Dan Butler	Reasoning with Cause and Effect - Judea Pearl, 1999. Additional reading/watching from Dan: A video lecture by Pearl on causal inference (the first 20 minutes of the lecture), 2008. Causal Learning without DAGs - Duvenaud et al., 2000. Learning STRIPS Operators from Noisy and Incomplete Observations - Mourao et al., 2012.
February 27	Abe-n-Tony	The Unreasonable Effectiveness of Data by Alon Halevy, Peter Norvig, and Fernando Pereira.
March 6	Abe	Is machine learning getting us to artificial intelligence? Bite-sized blog post from Andrew Gelman and main paper by Duvenaud, Lloyd, Grosse, Tenenbaum, and Ghahramani, 2013. Very relevant paper: The discovery of structural form from Kemp and Tenenbaum, 2008. Short blog post about why these questions are good to ask. Another related paper by Grosse et al., 2012.
March 13	John Doe-Smith	What Is a Knowledge Representation? by Davis, Shrobe, and Szolovits, 1993.

Possible Topics

Here are some suggestsions for discussion topics, as well as resources to where good topics may be lurking.

• The Cyc Project
• Debates about knowledge representation and inference: Two Theses of Knowledge Representation
• AAAI's classic papers list
• Tony's classic AI paper list.