[ CogSci Summaries home | UP | email ]
http://www.cc.gatech.edu/~jimmyd/summaries/

McCarthy, J., Minsky, M., Sloman, A., Gong, L., Lau, T., Morgenstern, L., Mueller, E., Riecken, D., Singh, M., Singh, P. (2002). An Architecture of Diversity for Commonsense Reasoning. IBM Systems Journal, 41, 3, 530-539.

@Article{MccarthyMinskySlomanGongLauMorgensternMeullerRieckenSinghSingh2002,
  author =  	{McCarthy, John and						
                 Minsky, Marvin and
                 Sloman, Aaron and
                 Gong, Leiguang and
                 Lau, Tessa A. and
                 Morgenstern, Leora and
                 Mueller, Erik T. and
                 Riecken, Doug and
                 Singh, Moninder and
                 Singh, Push},
  title =	{An Architecture of Diversity for Commonsense Reasoning},
  journal = 	{IBM Systems Journal},
  year = 	{2002},
  volume = 	{41},
  number = 	{3},
  pages = 	{530--539}
}

Author of the summary: Jesse Storry, 2007, jstorry@connect.carleton.ca

Cite this paper for:

Comonsense reasoning is too hard a problem to solve using any single artificial intelligence technique. McCarthy et al. propose a multilevel architecture consisting of diverse reasoning and representation techniques that collaborate and reflect in order to allow the best techniques to be used for the many situations that arise in commonsense reasoning. They present story understanding as a task for evaluating and scaling up a commonsense reasoning system.

Basic proposal:

  1. Develop Minsky and Sloman's ideas about a multilevel cognitive architecture.
  2. Develop a system that would exploit existing AI techniques, such as case-based reasoning, logic, neural nets, genetic algorithms, and heuristic search.[p530]

An advantage of the story understanding task is that standardized tests are available to evaluate students on their reading comprehension skills.[p530] A story understanding system should be able to read and understand a story, demsonstrating this by: answering questions about the story, producing paraphrases or summaries, and integrating information the story contains into a database.[p531]

The knowledge needed to solve a commonsense reasoning problem is typically much more extensive and general than the knowledge needed to solve difficult problems. Commonsense reasoning is implicit, whereas the knowledge needed to solve difficult problems is often explicit, making implicit knowledge explicit is a time consuming task requiring a seriuos engineering effort.[p531]

A huge amount of domain knowledge is needed to do even simple forms of commonsense reasoning.[p531]

Story understanding research is blocked on three critical problems: (1) complexity of the structure of natural language, (2) necessity for large commonsense knowledge bases, (3) combinatorial explosion in the understanding process, such that there are multiple interpretations at all levels of language.[p532]

McCarthy et al. propose a three-pronged approach. To combat the complexity of natural language structure, only books for early readers will be used. To deal with the necessity for large commonsense knowledge bases, the domains most frequently used will be identified and addressed before all others. An architecture of diversity will be used to combat the combinatorial explosion problem.[p532]

McCarthy et al. start with a test set and development set of stories. They manually annotate each story in the development set with an informal inventory of what domains of commonsense knowledge and reasoning must be addressed in order to understand the story. They sorted the domains by their frequency and attempted to develop methods to understand which domains occur most frequently. Testing is carried out on the test set. This process is iterated on several higher reading levels, progressing to stories for Grades 1, 2, and 3.[p533]

This problem is too large to solve using any single approach, it must emerge from a large-scale architecture of diversity.[p533]

McCarthy et al. conjecture that the human architecture of diversity is something like the H-Cogaff model (Sloman 2001).
Reactive processes are those in which internal or external states immediately trigger internal or external responses.
Deliberative processes are those in which alternative possibilities for action can be considered, evaluated, and selected or rejected.
Meta-management processes add the ability to monitor, evaluate and to some extent control processes occuring within the system.
The three layers operate concurrently and do not form a simple dominance hierarchy.
Adult humans appear to have all three type of processing, probably rare among other animals.[p533-534]

In story understanding, the meta-management level may control the deliberative level in a number of ways.

Minsky further elaborates the H-Cogaff architecture into the six-level architecture called "Model Six" which can be seen here (Minsky forthcoming).

An architecture of diversity would embed representations from natural language to micronemes, as depicted in Figure 3 (Minsky 2001).

Here are a few of the understandings that a commonsense story understanding system must handle:

McCarthey et al.'s goal is to aim toward a critical "change of phase" that will come when they cross a threshold at which their systems know how to improve themselves. There has not been enough work done on the higher reflective and self-reflective levels that humans use, as they learn to improve their thinking itself.[p537]

McCarthy et al. will try to develop a plan that will incorporate into one system the virtues of many different approaches. Since each scheme has deficiencies, their system will escape this by using higher-level, more reflective schemes that understand what each of the other schemes can do and in what context they are most effective.[p538]

Summary Author's notes: