Seminar 1 - Department of Information Technology

Seminar 1
Technology, research
and ethics
E. Dyachuk,
K. Haikonen, K. Kovi,
L. Saarinen, L. Sjökvist
Division of Electricity
Dept. of Engineering
Sciences
Uppsala University
Sweden
E. Dyachuk, K. Haikonen, K. Kovi, L. Saarinen, L. Sjökvist
Themes
• Responsibility
• Statistics and Errors
• Conflicts
Responsibility
Prosecution:
Defendants:
• Alledge the defendants
gave a falsely reassuring
statement before the
quake
• Not possible be precise
about the timing of future
events. 30 sec. warning is
possible.
• Did the scientists weight
up all the risks, and
communicate these
clearly to those seeking
advice?
• The best science can do
is talk in terms of risk and
of probabilities
• ”We just wanted to be
warned that we were
sitting on a bomb”
• Seismologists have been
saying since 1998 that it
is a high risk area.
Responsibility
• What is the scientists responsibility?
• -Did they make the right decision or should the have
warned the public, with the risk of crying wolf?
• Communicating scientific results is called ”the third
mission” of swedish universities, but how to communicate
that ”low risk” does not mean ”not dangerous”?
Statistics and errors:
Reproducibility
• Examples from drug research:
– 6 out of 53 studies on cancer were possible to
reproduce
– ¼ av 67 seminal studies
• Examples from machine learning:
– Overfitting
– Estimation: ¼ of results can be reproduced
• Examples from publishing in general:
– Reviewers pick up 2 of 8 deliberate mistakes (British
medical journal)
– An article with several obvious errors is accepted by
½ of the journals it was submitted to (biology/cancer)
Errors and statistical power
Good research practice
• Disclose sources of error
• Use blinded data
• Publish also negative results
• Reproduction studies are important
• Thorough peer review very important
• ”Minimal-threshold journals” which demand good
scientific quality but not that the research is ”new” and
”significant”
Conflicts
• Scandal at Uppsala University, 2007
– The scandal
– Public reaction
Conflicts
• Mail storm
•
•
•
•
Leakage of internal mails onto public domain.
Publishing anonymously on internet.
Open discussion by several others in public domain.
Discussion on the credibility of research models.