Re: Single model or framework for reputation/trust

Ed Gerck <[email protected]> wrote:
> Reputation is usually understood as something from the past. It's
> not about promises.

  Reputation (or past behavior) is a strong indicator of future
behavior.  It shouldn't be ignored.

  e.g. See any book on profiling criminals.  The books lay out exactly
what the criminals are doing wrong, and describe how to use that
information to catch them.  Yet despite having this information,
criminals *can't* change their behavior, because of the exact problems
that make them criminals.

> In addition, you are, perhaps, familiar with the example given by 
> Bertrand Russel, where a farmer would go everyday and walk to a 
> chicken to give food to it. One day, the farmer walked in the same 
> way to the chicken and killed it. If reputation applies here, the 
> farmer should have had a good reputation with the chicken. 

  Very cute, also carefully misleading.  It's trying to pretend that
reputation is perfect, when it's not.  It's also pretending that
positive reputation carries more information than negative reputation,
when it doesn't.

  e.g. 90% of the population doesn't commit crimes more serious than
jaywalking... except for the occasional "surprise".  The positive
reputation they have is useful, and is a good predictor of future
non-criminality.

  The other 10% has a negative reputation, and that reputation IS a
stronger indicator of future behavior than the positive reputation of
non-criminals.

  Alan DeKok.