Dr. Mark Humphrys

School of Computing. Dublin City University.

Home      Blog      Teaching      Research      Contact

Search:

CA216      CA249      CA318

CA400      CA651      CA668


Mark Humphrys - Research - The World-Wide-Mind - ISWC-02 summary



Lowering the entry level:
Lessons from the Web and the Semantic Web
for the World-Wide-Mind


Ciarán O'Leary1 and Mark Humphrys2

1 School of Computing, Dublin Institute of Technology, Kevin St, Dublin 8, Ireland
www.comp.dit.ie/coleary
2 School of Computing, Dublin City University, Glasnevin, Dublin 9, Ireland
computing.dcu.ie/~humphrys

poster at 1st Int. Semantic Web Conf. (ISWC-02). See full reference.



The "World-Wide-Mind" (WWM)

The "World-Wide-Mind" (WWM) is a scheme in sub-symbolic AI (numeric and behaviour-based AI, neural networks, animats, artificial life, etc.) for constructing complex agent "minds" (by which we just mean action-taking control systems) through multiple authors. Authors put their (sub-symbolic) agent minds online, and other authors use these minds as components in larger minds. This works in sub-symbolic AI because competition is resolved using schemes like competing numeric weights, rather than through explicit symbolic reasoning. So we can define a sub-symbolic communication protocol based on numbers which avoids (for the moment) the difficult problems of knowledge representation and "agent communication languages" [Martin et al., 2000] of the symbolic AI level. We still want to put these programs online in a communicating network though, and it is here that ideas from Web Services and the Semantic Web will still be useful.

Particular properties of the WWM problem domain

This poster does not discuss in detail exactly what the WWM is for in sub-symbolic AI, for which see [Humphrys, 2001] and [Humphrys and O'Leary, 2002]. Instead it will treat this as another problem domain in which to apply Semantic Web ideas - like other domains of specialised research communities, and groups with programs communicating online. The substance of this poster is taken up with considering how Semantic Web ideas can be applied given some particular properties of this specific problem domain, namely:

  1. We are trying to get existing sub-symbolic AI researchers to "publish" their algorithms online for remote re-use by others (as, essentially, Web Services).

  2. The target audience are programmers, but not network programmers. Any scheme that requires them to become network programmers (or indeed learn any new concepts) will fail (will fail in the sense that most algorithms will remain offline).

  3. These algorithms are often unique, not commodities. Sub-symbolic AI is driven by unique individuals and research teams whose work is often not easily replicable by anyone else. Indeed, one of the problems with this field is how few people try out each other's algorithms [Bryson, 2000; Bryson et al., 2000; Guillot and Meyer, 2000]. Often, if the author does not put his algorithm online then no one will.

  4. As a result, we are likely to be very "forgiving" of whatever the researchers do put online.


The entry level

The Semantic Web entry level

The standard approach with the Semantic Web has been to aim the technology at network programmers, Semantic Web experts, and other specialists, and assume that tools can hide this complexity from non-specialist users [Hendler, 2001]. As a result, the technology is forbidding for the non-specialist, even for programmers. The Semantic Web has made a deliberate decision to raise the entry level. We argue that this will not work in this case.

The Web entry level

The Web itself showed a different approach, where the technology itself could be approached by the non-specialist, at least at the entry level. We adopt this approach.

The WWM entry level

We construct an extremely low entry level, which rejects: (a) local installation, (b) models of programs online that require network programming or complex APIs, and: (c) models of data that are unforgiving - where data must be well-formed or will be rejected. We explain why these ideas will not work in this problem domain. Instead, we construct an ultra-simple entry level of: (a) remote use of server-side programs, (b) simple CGI programs reading standard input and writing standard output, and: (c) the "data" transferred in this world-wide "Society of Mind" being sub-symbolic queries and responses in an XML-like plaintext format, where well-formedness, as in the Web, is not necessary.


Conclusion

This simple entry level does not compromise usage of advanced Web Services and Semantic Web concepts at higher levels, as we explain in the full poster. The discussion of design decisions here may have implications for other areas of the Semantic Web where the target audience may be programmers but not network programmers.

Full poster and further information

Further information about the World-Wide-Mind project is available at:
w2mind.org
The full version of this poster is available at:
w2mind.org/publications.html
or through:
computing.dcu.ie/~humphrys/publications.html

References

Direct links to some of these authors and papers may be found in the full poster and are not duplicated here.

[Bryson, 2000] Bryson, J. (2000), Cross-Paradigm Analysis of Autonomous Agent Architecture, JETAI 12(2):165-89.

[Bryson et al., 2000] Bryson, J.; Lowe, W. and Stein, L.A. (2000), Hypothesis Testing for Complex Agents, NIST Workshop on Performance Metrics for Intelligent Systems.

[Guillot and Meyer, 2000] Guillot, A. and Meyer, J.-A. (2000), From SAB94 to SAB2000: What's New, Animat?, Proc. 6th Int. Conf. on Simulation of Adaptive Behavior (SAB-00).

[Hendler, 2001] Hendler, J. (2001), Agents and the Semantic Web, IEEE Intelligent Systems Journal, March/April 2001.

[Humphrys, 2001] Humphrys, M. (2001), Distributing a Mind on the Internet: The World-Wide-Mind, Proc. 6th European Conf. on Artificial Life (ECAL-01), September 2001.

[Humphrys and O'Leary, 2002] Humphrys, M. and O'Leary, C. (2002), Constructing complex minds through multiple authors, Proc. 7th Int. Conf. on the Simulation of Adaptive Behavior (SAB-02), August 2002, Edinburgh, Scotland.

[Martin et al., 2000] Martin, F.J.; Plaza, E. and Rodriguez-Aguilar, J.A. (2000), An Infrastructure for Agent-Based Systems: an Interagent Approach, Int. Journal of Intelligent Systems 15(3):217-240.


Feeds      w2mind.org

On Internet since 1987.