Technological Innovation as an Evolutionary Process

Previous communication from John Sowa (2/15/2004 5:27 PM)

http://www.bcngroup.org/Macrocognition/four.htm

Paul,

I know lots of people who work in the general field of AI, but no two of them agree on anything that might be called an "AI mythology".

I think that the most seriously wrongheaded mythology is the one that is being promulgated as the "semantic web".

Following is a note that I circulated to the SUO email list in response to some people who were defending Tim B. L.'s "layer cake" diagram (see attached), which I believe is a disaster.

John Sowa

Danny,

I'll start with your closing comment, which I believe puts everything else in perspective:

Danny

That's certainly an interesting approach, though there's presumably quite a lot of work needed. In the meantime, I suspect the web of semantics is more likely to resemble TimBL's layer cake:

http://www.w3.org/2001/09/06-ecdl/slide17-0.html

I believe that Tim's layer cake (Figure below) is a recipe for disaster. To clarify that prediction, let me summarize some history:

Figure 1: Layer cake

History:

1. The current WWW is the realization of a dream by Vannevar Bush in his 1945 article, which I urge everybody to read (or reread):

http://www.theatlantic.com/unbound/flashbks/computer/bushf.htm

from The Atlantic | July 1945 | As We May Think | Vannevar Bush

2. Three major steps along the way to the WWW came in the 1960s: One was the Arpanet, which became the Internet, another was the General Markup Language (GML), which evolved into SGML, and the third was Ted Nelson's idea for hypertext, unique identifiers, and the "docuverse":

http://www.iath.virginia.edu/elab/hfl0155.html

from Ted Nelson and Xanadu

3. Vannevar Bush was 40 years ahead of his time, and Ted Nelson was just 20 years ahead. But by the late 1980s, all the technology was there. Several major companies -- IBM with GML, Apple with Hypercard, and Xerox with all their innovations -- had a golden opportunity to implement the Bush-Nelson vision, and they let it slip through their fingers.

Tim B. L. deserves a great deal of credit for seizing an opportunity that IBM, Apple, and Xerox ignored. But the vision had been well defined for 25 years, and the technology was there for the asking. Tim put together SGML, hypertext, and the Internet to implement Nelson's docuverse.

The point of this history lesson is that the two bottom layers of Tim's cake are tiny increments beyond what he did over a dozen years ago. He just replaced ASCII with Unicode, URL with URI, and HTML with XML. As vision goes, it has nothing beyond what Nelson wrote in 1965.

To put the upper layers of Tim's cake into perspective, reread Vannevar Bush's article of 1945. In that old article, Vannevar Bush showed an amazing grasp of all the technology of his day, and every one of his predictions has come to pass. But Tim's upper layers show no recognition even of the mainstream computer technology of today. Following are the missing pieces:

1. The most glaring absence is database technology. The first large-scale databases came into use in the 1960s, and relational databases came into use in the 1970s. Today, the entire world economy runs on SQL -- which is a version of logic that is far more expressive than OWL. As I said before, I have no love for SQL notation, and Ted Codd himself said that he wished that he had invented Prolog -- another invention from the '70s.

2. Another glaring absence is integration with computer processing. Sun has been using the motto "The network is the computer" for years, and they designed Java as a language that integrates the WWW with the database, programming, and devices ranging from mainframe servers to PCs, PDAs, and handheld phones. Microsoft has followed Sun by integrating the programming languages, GUIs, and network services in the .NET framework. But none of that is reflected in the layer cake.

3. A third missing point is software design and development. Today, the browser has become the major GUI interface for applications, and anything done with XML for the browser must be integrated with the database, the programming languages, and the tools for application design. For those purposes, the very widely used UML supports multiple subsets of logic. The UML type hierarchy is at the level of OWL, but UML also supports E-R diagrams, activity diagrams (slightly modified Petri nets), and the Object Constraint Language, which supports full FOL. None of this is being considered or accommodated by the W3C efforts.

4. The most glaring absence is the lack of any consideration for what has been happening in AI since the 1970s, but this would require much more discussion. Just one tiny example: Some of my colleagues have been using Prolog as a replacement for XSLT, and it works wonders for that purpose. Prolog is vastly more powerful, more efficient, and more flexible (and another of my colleagues compiles Prolog to Java bytecodes, which run in any browser).

Following are some comments on your other comments:

Danny

OWL's adequacy depends on the purposes to which it's applied - so far it looks adequate for Semantic Web purposes.

John Sowa

If you limit the semantics of the semantic web to what can be done with OWL, then that's a tautology.

Danny

Why? Why not support what you need in a given context - surely that will be simpler.

John Sowa

I agree that you don't need to have every tool use every subset of logic for every application. But what you specify in one subset must be compatible with what you specify in every other subset.

That is a serious limitation of the current version of UML, which does not have a common logic-based semantics for all the diagrams. And that problem will be even more difficult to address if the logic used for the web is unrelated to the logic used for the databases and application programs.