[
Lists Home |
Date Index |
Thread Index
]
From: Andrzej Jan Taramina [mailto:andrzej@chaeron.com]
>> The choice to choose is always there. The choice
>> of choices to choose from is not.
>He who controls the list, controls all. ;-)
So one chooses authority carefully. The criteria of
that choice is another long topic. We've been there
before on this list during the semantic web debates.
>> That was the price of higher reliability in an open
>> network of choosers.
>Which might be a necessary evil in many cases.
It might not be evil. Having a control hierarchy
is pretty much standard engineering. Two choices:
1. The W3C would be the organization to control
the processes and therefore, to take authoritative
control of markup systems specification. That one
can be debated, but at
the time, it carried the day. I am not disturbed
by the overall results. I can quarrel with the
means and the precedents and have but not in this
thread.
2. A result of one of those processes would be to
limit the syntax (essentially, freeze the SGML
Declaration and remove it from the machine architecture
except insofar as to weld the decisions to the
XML processor).
Then the choice of the customer was SGML with more
options, more responsibilities, greater freedoms,
but more chances to fault systematically, or XML
with a lot of decisions pre-made and makable given
the convergence in hardware, network protocols, Unicode,
costs of subsystems, and the ubiquity of training
given the success of HTML, thus higher reliability.
In other words, the control regime was possible
and likely to work for the most people in the
best way. Greatest good for the greatest number.
A choice was enabled. It worked overall. I side
on keeping SGML around to ensure there is a rollback
strategy for those who actually do realize a requirement
to dink in the syntax and still be a conforming markup
system. Otherwise, I presume (no figures) that the
application of SGML is still declining.
>> The actual authority resides in the exercise
>> of the rule to audit and then to join to a rule for action
>> on audit. This actually probably works fairly reasonably
>> given reasonable humans
>I'm not so sanguine about this. If you need to reach a site for business
>purposes, and an automated 'bot denies access, there may be a human review
>process. However, you cannot ignore the temporal component. If you needed
>the info in a timely manner, running through a manual bypass process can
>easily take too long (ever had to do this with a Fortune 500 company where
>response time to such requests is usually measured in weeks?) for the goal
at
>hand. My experience is that this is the norm, not the exception.
That is a QOS (Quality of Service) argument. I accept it and for many
business systems when choosing a source for a web service of a type, it
will be the critical question. That is the commoditization pressure.
>Humans may at times be reasonable (though the course of human history leads
>me to conclude otherwise). Corporate policies and procedures are rarely
>reasonable.
Again, QOS. At least the human has a higher probability
of being flexible and reasonable. It is difficult and in some cases
bad engineering to generalize a specific program. It is more likely one
will
create a control hierarchy of general policies that can be implemented
as higer order control programs. SMOP.
>Your conclusion depends on so many assumptions that it is more idle
>speculation than anything else. ;-)
I am not concluding that it is always better to have a human in the
loop. I think it system mission specific and a risk analysis problem.
One should not commit zero-sum game decisions for critical decisions
to completely automated means unless it can be determined that the
space of choices and outcomes is finite and within the energy budget
of the system (the entropy problem). I've said this before, but in
the old days of the nuclear blowout, one said simply, 'don't commit
to launch on warning'.
In assessing which business policies are committed to automated
means, we must learn how to assess which decisions are of that
type and which can be defaulted. 2+2=4 is typically safe on
one extreme. On the other end, "launch', analogical pattern matching
is riskier. Send "Amir To Jail Without Council Because He is Named Amir
and Doesn't Have Good Western Social Skills" is among the second
type. "Buy Frobis Because Frobis Is Popular" is in the middle.
len
|
