Healthcare Informatics, the Semantic Web and Public Policy: Balancing Wishful Thinking with Realism
Henry S. Thompson
School of Informatics
University of Edinburgh,
Markup Technology Ltd.
and
World Wide Web Consortium
April 2005
© 2005 Henry S. Thompson
1.
Web Services—Semantic Web
Web Services is the name for a marketing initiative
- The technology has been scrambling to catch up and provide some grounding for the phrase ever since it was invented
The Semantic Web is the name for a vision of the future
- Originally Tim Berners-Lee's attempt to answer the question: What is the full potential of the (World Wide) Web?
They have a common dependency
2.
Web Services
My quick summary of Web Services:
- Loosely-coupled distributed applications
Three key aspects:
- Messages—XMLP (ex-SOAP), XML Schema
- Definition—WSDL: XML->XML function signatures
- Discovery—UDDI (CORBA, oops)
3.
Semantic Web
As vision
- The Web as an information resource for computers
As technology
- A lowest-common-denominator data model
- Relational triples
- With an XML serialisation
Starting to grow
4.
The Origins of the Semantic Web
The information retrieval crisis beginning in the late 1990s led to a widespread interest in what has come to be called metadata.
What is metadata?
- It's just data.
- But it's data about other data
- Data intended for machine consumption
What could metadata do for us?
- Give search engines something to work with (relational triples) that is designed for their needs.
- Give us all a place to record what a document, or any other resource, is for or about.
5.
First Requirements for Metadata
What would we need to make this work?
- A standard syntax, so metadata can be recognised as such;
- One or more standard vocabularies, so search engines, producers and consumers all speak the same language;
- Lots of resources with metadata attached;
- Attribution and trust
- Is this resource really about Pamela Anderson?
6.
Meaning is at the Core
Both SW and WS depend crucially on moving beyond syntax
- XML as such is just ASCII for the 21st century
- Web-appropriate linearisation
- for tree-structured documents with internal links
-
and tree-structured documents are a pretty good transfer syntax for just about anything
- What prospects for moving beyond syntax to semantics?
-
The Semantic Web is committed by its very name
-
Web Services can't succeed without it
-
And this is where the relevance of history kicks in
7.
The place of semantics
Computers 'looking' for 'service providers' are not the same as human beings shopping on the web
Computers 'sharing' each others 'knowledge' are not the same as people reading web pages
So the metaphors underlying both WS and the SW can be very misleading
Negotiation between producers and consumers is the key
- If you can't describe what you want, you can't have it
- If you can't describe what you've got, no-one will use it
- If you can't dicker, you'll always lose
These observations apply equally well to Web Services and the Semantic Web
8.
Those Who do not Study History
are doomed to repeat it
The history of AI is full of examples of two weaknesses:
-
Over-promising by insiders
-
'AI Winter'; Intelligent Agents
-
Over-optimism by outsiders
25 years ago Ed Feigenbaum described Terry Winograd's work as “a breakthrough in enthusiasm”
-
I worry that WS and SW, in their reliance on effective computational semantics, are vulnerable to the same criticism
9.
The History of the Knowledge: Representation Problem
The representation and exploitation of knowledge has been the ultimate grand challenge for Artificial Intelligence since its
inception
Our own human intelligence has sometimes been a real handicap
-
It's too easy to look at a screenshot and see how much knowledge is captured
- for example:
(#$and
(#$isa ?x #$Person)
(#$feelsEmotion ?x #$Fear #$High))
Designing apparently expressive notations is easy
-
Making them do actual work is much harder
10.
The Missing Inference Engine
What we learned in 1978–79 was that designing an approach to KR without first designing an inference engine was a waste of
time
Actually worse than a waste of time
- Because you could invest a lot of work in representing stuff
-
And still end up with nothing to show for it
So we were left with an embarrassing tradeoff:
-
Use (something isomorphic to) 1st-order predicate logic, and get a variety of pretty well-understood inference engines
-
Use something more user-friendly and expressive, but be unable to exploit it
This tradeoff is still with us today
11.
The Semantic Web Today
1 ½ of the four first requirements for metadata I mentioned earlier:
- RDF Model and Syntax gives us recognisable metadata
- OWL gives us a mechanism for defining shared vocabularies, and we have a few
12.
Responsible Public Policy
Top-down argument from need is misguided and even dangerous
- Remember Launch on Warning and Reagan's Star Wars
The (health-care) informatics professional's job is to inject realism into the public arena
- Not simply seek local advantage
It's in everybody's interest to plan based on
- Actual products, not promises
- Proven technology, not vapourware
13.
Augmentation, not replacement
Judgement is at the core of the practice of medicine
- It's not a candidate for automation
Evolution, not revolution
- Greenfield sites are very rare
Look for opportunities to actually improve a process by 10%
- instead of dreaming about replacing it wholesale