RE: First Order Logic and Semantic Web RE: NPR, Godel, Semantic W eb

[Jeff Lowery <jlowery@scenicsoft.com>]

> It isn't any harder than working with contracts for services 
> in any other domain.  Identify a reliable provider.  The harder 
> problem is to train an agent correctly if you intend to enable 
> it autonomy.  It is the same problem as choosing management.

That's really my point. With a human in the loop nothing much as changed.
Humans have a deep background in social interactions that give them a
hard-won way to evaluate their sources of information, and even they get
hacked sometimes. An automated process doesn't have that experiential base
to rely on. How does one train an agent? Tell it which sources are reliable,
and which are less reliable? There's going to be a lot of sources out on the
web; you'll need an authority of authorities (ad infinitum?). And how do you
evaluate the impact of actions that those agents can take? You have to
weight both your automated knowledge and your automated actions.

Remember Black Monday? Automated stock agents went into a massive sell-off
because they could see their positions were deteriorating and correctly
tried to cut their losses. They went into automated panic. Consequence: new
rules were developed to halt computer trading in those situations. Things go
wrong fast with agents, slower humans did less damage in more time and could
reflect on their actions.

That was the market. The Semantic web is going to address virtually every
human domain, and impact (through agents and humans acting on agent-derived
information) every human endeavor. Can we avert the dangers? Not all of
them, because we're human and we naturally put too much trust in our own
inventions. We can predict and avert or discover and correct faults, but
it's going to take a looong time until something truly useful and benign is
developed. In the mean time we should focus on prediction of failure rather
than discovery. Let's not build it and wait for all the correction
mechanisms to be sorted out later.

This is all common sense; write me an agent with common sense.

P.S.
On a related note, I was wondering about feedback loops. What happens when
one stereotype is strong formulated on another, and visa versa? There can be
many layers of indirection here, so spotting these self-reinforcing
stereotypes could be difficult. What thought has been given to this? I think
a lot of social pathologies get started this way, as well. Like eugenics,
for instance, where white anglo-saxon people 'proving' that they were
superior by somehow having all the 'desirable' physical characteristics.
Hadn't stopped to think that they were the ones deciding what was desirable.

> -----Original Message-----
> From: Bullard, Claude L (Len) [mailto:clbullar@ingr.com]
> Sent: Friday, May 11, 2001 6:12 AM
> To: Jeff Lowery; John Cowan
> Cc: Danny Ayers; xml-dev@lists.xml.org
> Subject: RE: First Order Logic and Semantic Web RE: NPR, 
> Godel, Semantic
> W eb
> 
> 
> 1. Trust but verify.  Action should require an expected reaction. 
> 
> 2. Contract for right to contest.  
> 
> 3. Limit the time of affect.
> 
> 4. Negotiate results.
> 
> 5. Subscribe to validated and vetted services.
> 
> 6. Only commit resources based on knowable outcomes.
> 
> It isn't any harder than working with contracts for services 
> in any other domain.  Identify a reliable provider.  The harder 
> problem is to train an agent correctly if you intend to enable 
> it autonomy.  It is the same problem as choosing management.
> 
> Len 
> http://www.mp3.com/LenBullard
> 
> Ekam sat.h, Vipraah bahudhaa vadanti.
> Daamyata. Datta. Dayadhvam.h
> 
> 
> -----Original Message-----
> From: Jeff Lowery [mailto:jlowery@scenicsoft.com]
> 
> Well, it seems that the amount of trust one puts in a stereotype or
> authority must be correlated with the weighted impact of 
> consequent actions
> based on the data involved. 
> 
> <snip>
> 
> How does one evaluate
> the consequences of false actions taken; who has the 
> authority to guide us
> here?
> 
> ------------------------------------------------------------------
> The xml-dev list is sponsored by XML.org, an initiative of OASIS
> <http://www.oasis-open.org>
> 
> The list archives are at http://lists.xml.org/archives/xml-dev/
> 
> To unsubscribe from this elist send a message with the single word
> "unsubscribe" in the body to: xml-dev-request@lists.xml.org
>