Posted on May 25, 2008 by Peter Turney
People like to say that a certain book “changed their life”: The quoted phrase “book that changed my life” gets 61,000 hits on Google. I have some favourite books, but can I honestly say that one of them significantly changed my life? It seems more likely that my life has been influenced by the gestalt of the books I have read; perhaps any individual book could have been skipped without much impact on my life. For the last few years, most of my professional reading has been journal and conference papers, rather than books. Also, there are books that I think are extremely important, such as Axelrod’s The Evolution of Cooperation, yet they have not had any impact on my research, so far. If a reader cannot point to some tangible outcome from reading a book, then the reader may be overestimating the personal impact of the book. Pondering these thoughts, I decided to try to track the major decision points that led to my current research project, to try to recall what books (or papers or lectures) most influenced my decisions.
1. Deciding to do research in AI: I was a teenager when I decided that I wanted to do research in AI. There were several digressions (e.g., a PhD in philosophy), but I eventually became an AI researcher. Science fiction was a major influence on this decision. I don’t think I can pin the credit (or blame) on any individual author or book, but A.E. van Vogt‘s short story Fulfillment stands out in my memory (in the collection The Far-Out Worlds of Van Vogt). I can see that it also had a big impact on Damien Broderick, since the opening sentence of Transcension comes from Fulfillment. I should also mention Asimov’s short story, The Last Question. As a teenager, I was quite troubled by entropy.
2. Deciding to specialize in machine learning: As a PhD student, my interest in AI was rekindled by Machine Learning: An Artificial Intelligence Approach (Volume I and Volume II). These books pulled me away from philosophy and back to computer science. They also led me to specialize in machine learning.
3. Deciding to focus on machine learning applied to semantics: In 1997, I attended a lecture by Geoffrey Hinton, on Unsupervised Neural Networks. He convinced me that I should learn more about unsupervised learning and that I should read A Solution to Plato’s Problem (Landauer and Dumais, 1997). Hinton was clearly very impressed by Landauer and Dumais, and so was I. This paper has had a major influence on my research.
4. Deciding to focus on analogy-making: Michael Littman invited me to try to apply machine learning techniques to SAT analogy questions. Around the same time, I was reading Where Mathematics Comes From (Lakoff and Núñez). Later, I read Metaphors We Live By (Lakoff and Johnson). These two books convinced me that analogy-making was central to cognition, and the SAT analogy questions gave me a tool that I could use to guide my research. Michael and I viewed the SAT analogy questions as analogous to the TOEFL questions that were used by Landauer and Dumais (1997). We were both directly influenced by A Solution to Plato’s Problem.
5. Deciding to attempt to emulate the Structure Mapping Engine without hand-coded representations: My current project is most influenced by (1) The structure-mapping engine: Algorithm and examples, (2) High-Level Perception, Representation, and Analogy: A Critique of Artificial Intelligence Methodology, and (3) Structure-Mapping vs. High-level Perception: The Mistaken Fight Over The Explanation of Analogy. The first paper describes a symbolic approach to analogy-making, the Structure Mapping Engine (SME). The second paper argues that SME is flawed, because it relies on hand-coded representations, instead of accepting raw data as input. The third paper argues that the first two papers are not really in conflict: It may be possible to enhance the SME so that it can use raw data instead of hand-coded representations. My current research is inspired by ideas from these three papers.
What books (or papers or lectures) changed your life (or research)?