The Manipulable World Project
Draft Project Description

Slides prepared by Aaron Sloman
partly on the basis of discussions at
a workshop organised by Marvin Minsky,
funded by the Epstein Foundation
St.Thomas Virgin Islands, 14-16 April 2002.

Background and overview

AI has become increasingly fragmented and factionalised since the late 1970s and it is not clear how much of the work done since then can contribute usefully to one of the high level goals of AI (and Cognitive Science) namely understanding how to design a complete human-like agent, or equivalently(!!), how to explain a large subset of the workings of a human mind.

These slides, prepared before, during and after the St Thomas Workshop, partially describe an ambitious project that aims to bring the sub-fields of AI together again. The slides owe much to ideas presented at the workshop in April 2002 led by Marvin Minsky and Push Singh, MIT, though in part they are related to an ambitious multi-disciplinary proposal for a five year project, submitted to the Leverhulme Trust in February 2001, which was shortlisted but not successful. (None of the short-listed projects was successful!)

The slides are available here in PDF and Postscript. (PDF) (Postscript)
Some of the ideas in those slides were later used in the EC-Funded Framework 6 Cognitive Systems project (CoSy) described here (planned to start September 2004)
In November 2002 I got involved in discussions of a proposal for a grand challenge to bring studies of brain and mind together in the context of attempting to design a robot combining many capabilities. Further information about that proposal is here
along with pointers to related material.

A paper arising out of the workshop by Marvin Minsky, Push Singh and Aaron Sloman will appear in the AI Magazine in 2004. A pre-publication version is here:


Browsers for both versions of the presentations are freely available on the internet. See the information in this file

The diagrams in the slides were all produced using the excellent (small, fast, versatile, portable, reliable, and free) tgif package, available for linux and unix systems from here:

The slides use a format suited to fill a typical computer screen which is wider than it is tall. These need to be viewed in "Landscape" or "Seascape" mode (rotated 90 degrees to the left). Your browser should provide such an option if it does not automatically get the orientation right.

A report on the St Thomas meeting, by Marvin Minsky and Push Singh is here, in two formats:

Push Singh's draft thesis proposal, developing some of these ideas is here:


If you send me comments, criticisms, suggestions, let me know if you'd like them to be made public, in which case I'll add them to this file:

Please specify whether you would like your public comments to be anonymous or not.

Related Material

Some notes on the difference between forward chaining and backward chaining research.

The latter starts by trying to identify things we don't understand and working back to ways of explaining and modelling them. The former works forward from what we can do. Some examples of what we cannot yet explain, including the development of perceptual ontologies can be found here

Possible targets and milestones:

Slides for presentation at IBM workshop
March 14-15 2002, IBM Research Center, New York.

"Architectures and the spaces they inhabit" (PDF) (Postscript)

The presentation argues for the need to explore not just one architecture but the space of architectures (designs) and the space of requirements (niches) and trajectories in those spaces, roughly on the grounds that you can't understand a complex system without analysing what difference it would have made had it been different in various ways. That includes looking at alternative designs, at least in some neighbourhood in design space and a related neighbourhood in niche space. (I contrast numerical evaluation functions with structured descriptions of relations between design and niche). In addition, evolutionary history may give clues when we are studying, and trying to emulate, a natural architecture.

The slides include an abbreviated discussion of virtual machines and how events and processes in those machines can be causes. That is discussed more fully in the notes for the philosophy of AI tutorial by Matthias Scheutz and myself at IJCAI01:

The slides end with a (partial) list of things we don't know.

Comments, suggestions, criticisms welcome.
Email A.Sloman AT

There is a growing collection of slide presentations developing related ideas here:

The Birmingham Cognition and Affect project overview is here:

A list of publications and reports and an online book are here

Our software tools (SimAgent toolkit and related software for teaching and research in AI) are described here

Aaron Sloman
Updated: 31 May 2004
Frames-free web site