Things to Avoid

B. Horn, MIT (1984 approx.)

I found this on a single sheet amongst some old papers in my office. It's aimed at AI researchers, but much of it applies to all research. Horn was Prof. Gennert's advisor at MIT. We should all be so lucky to get such good advice!


Things to Avoid

The early years of any field tend to be characterized by a wide variety of approaches, many false starts and techniques based on inappropriate analogies. We can learn from these mistakes if we wish. Here are some things to avoid:

  1. Using a mechanism-oriented approach, instead of a problem-oriented one.

  2. Applying a known bag of tricks from another field.

  3. Believing that complexity will automatically give rise to interesting behavior.

  4. Hoping that [machine] "learning" will provide a boot-strapping mechanism.

  5. Believing what works in a simple situation can be easily extended to a more complex one.

  6. Suffering from theorem-envy -- introducing unwarranted mathematical hair.

  7. Working only on the "interesting" sub-problem -- often not the weakest link.

  8. Following the latest fad. Create your own instead!

  9. Taking a random path through a maze of possibilities without explanation.

  10. Admiring the King's new clothes. / Tue Jul 25 20:43:07 EDT 2000