Note on Trust
CS73N Note on Trust
Created April 20005 by Gio Wiederhold, renamed 7 April 2006, edited 10 April 2006.
Trust
Why do you trust printed material?
When can you Trust information from the Internet?
External: Credible sources?
- Good citations to authorities
- Reputation of author -- if it's not anonymous as a wiki entry
- Reputation of the institution
- Reputation of the publisher Here is where Internet publications are weak
- [more?]
Internal:
- Well reasoned
- Decent presentation: spelling layout
- There are domain and cultural differences on what is acceptable.
- If it involves predictions into the future, as many of CS73N writings will,
-
- Does it match the present?
- is there a sufficient range of datapoints to allow an extrapolation? Extending a curve out more than twice is always unwarranted unless there are comparable examples.
-
- Does the future make sense? Examples:
- 1970 prediction about number of programmers in 1990.
- 1999 analyisis of toy sale predictions: by 2003 115% of all toys sold would be sold on the web.
If you believe it is true you'll trust it, even if it is wrong.
Much information comes from data via processing that involves assumptions and metrics. Those are not always clear, and sometimes omitted when authors want to make a point and avoid doubt.
The interpretation of factual information may change greatly if other, equally reasonable metrics are used
a) Volvo is a safe car: Do trust that? What causes accidents? Who drives Volvos? Safer than average drivers!
b) Air France is the world's largest airline: Do trust that? Was computed using route-miles, not passenger miles, and Air France had many long routes to Africa and South America that were flown once or twice a week!
c) Flying is safer than Driving: Do trust that? Is computed by mile. Ratio will change by >10 if computed by hour. Would you fly once per month to the East-Coast? Would you drive once-per month to the East Coast. Are those trips comparable by the mile?
|
