I have lived in both worlds. When advising CS students, I stressed doing rigorous studies and trying for full automation and rigorous testing. When advising collaborators, I stressed throwing post-doc labor at the problem and solving such problems manually. Even then, it’s essential to point out the need for testing repeatability and inter-grader agreement.
“Fully automated” is very expensive, when you actually need reliable results yesterday. “Manual” does not always guarantee correctness. One the third hand, measurement error is a common source of noise that can be dealt with by increasing sample size - measurement methods don’t have to be perfect in order to be useful. Finally, both automatic and manual methods should always include an “I don’t know” option. Both humans and algorithms should be aware of their limitations and be willing to say “this problem is outside my range of competence”. -- Kenneth Sloan [email protected] Vision is the art of seeing what is invisible to others. -- ImageJ mailing list: http://imagej.nih.gov/ij/list.html
