Talk:ACT-R

What ACT-R looks like
Along those lines, can someone add a picture of ACT-R's buffer/module arrangement? I am not sure I have an image that wouldn't violate copyright agreements.

It might also help spark interest to have a graph with BOLD response predictions. This would also help dispel the unfortunate but lingering impression that ACT-R is a "GOFAI" symbolic production rule system. --24.92.55.159 (talk) 16:41, 22 May 2008 (UTC)

Although ACT-R is certainly not a "GOFAI" symbolic system, it is most certainly a symbolic system and should be clearly classified under the symbolic approach to cognition. The BOLD response predictions are entirely irrelevant to this issue. The statement that "ACT-R has been implemented as a neural network, proving that there is nothing intrinsically "symbolic" in the underlying theory" is a non sequitur and shows complete lack of understanding of what the symbolic-connectionist debate is about. Assuming for a moment that symbolic theories of the mind are correct, this would mean that the brain implements a symbol manipulation system. This is precisely what the symbolic approach to cognition consists in, namely, that the symbolic system of the mind is implemented by the neural network of the brain. Obviously, this does not make the symbolic approach a non-symbolic approach. This is also known as the relegation of connectionism to implementation status (see eg Fodor & Pylyshyn 1988).

ACT-R is a symbolic system because its entities (chunks and productions, which are discrete and unitary representations) operate on the basis of syntactic properties and admit chunks in their slots (and buffer checks) regardless of the semantic content of those chunks. This corresponds to the notion of variables in programming formalism and pretty much amounts to the definition of a symbol manipulation system. In contrast, in connectionist systems, representations are not unitary entities but (usually distributed) patterns of activation over nodes. The activation quantities, which make up the semantic content of the representation, take part in the computation by affecting other units through the connection weights. In connectionism, representations are intrinsically overlapping and similar to a measurable extent, whereas symbolic units such as chunks and productions are entirely discrete. In connectionism it is possible (indeed, unavoidable) to represent an "intermediate" entity not defined by the modeler, when the activation values do not correspond precisely to a predefined pattern, whereas in symbolic systems it makes no sense to speak of intermediate entities -- eg in ACT-R one of the available chunks or productions must be selected as a unit, or a new unit must be created. —Preceding unsigned comment added by 194.177.192.8 (talk) 13:54, 23 January 2009 (UTC)

Most papers I read about ACT-R characterize it as a hybrid approach to cognition. I think there might be a misunderstanding of the term subsymbolic here. Where some refer to this as a special kind of representation, i.e. non-symbolic representation, others use this term for statistical approaches to cognition. As ACT-R uses both, hard rules and statistical rules, it perhaps is correctly termed hybrid. This is also suggested by the following cite from the homepage of the project:

ACT-R is a hybrid cognitive architecture. Its symbolic structure is a production system; the subsymbolic structure is represented by a set of massively parallel processes that can be summarized by a number of mathematical equations. The subsymbolic equations control many of the symbolic processes.

Because of that I will remove the chapter "The symbolic vs. connectionist debate" as already suggested by others.AdolfEinstein (talk) 14:18, 7 August 2014 (UTC)

Simplifying the article
I think the article would benefit from getting rid of two sections:


 * the theory vs. implementation section: I don't really see the point of this section, and it might be distractive.
 * the connectionist vs. symbolic debate, because, interesting as it might be, is off-topic. It is also one of those issues in cognitive science that can spawn endless debates.

I also think that two additions should be worth it:
 * A figure with an overview of ACT-R's modules.
 * A section on how to make BOLD-response predictions in fMRI experiments.

Gardoma (talk) 16:05, 13 July 2009 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified 1 one external link on ACT-R. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:
 * Added archive https://web.archive.org/web/20080907005223/http://act-r.psy.cmu.edu/workshops/workshop-2002/talks/AnthonyHarrison.pdf to http://act-r.psy.cmu.edu/workshops/workshop-2002/talks/AnthonyHarrison.pdf
 * Added tag to http://terrystewart.ca/papers/2005-Reimplementing_ACT-R_full.pdf

When you have finished reviewing my changes, please set the checked parameter below to true or failed to let others know (documentation at ).

Cheers.— InternetArchiveBot  (Report bug) 01:35, 1 October 2016 (UTC)

External links modified
Hello fellow Wikipedians,

I have just modified 1 one external link on ACT-R. Please take a moment to review my edit. If you have any questions, or need the bot to ignore the links, or the page altogether, please visit this simple FaQ for additional information. I made the following changes:
 * Added archive https://web.archive.org/web/20100802225919/http://www.humanfactors.illinois.edu/Reports&PapersPDFs/JournalPubs/FuPirolli07.pdf to http://www.humanfactors.illinois.edu/Reports&PapersPDFs/JournalPubs/FuPirolli07.pdf

When you have finished reviewing my changes, please set the checked parameter below to true or failed to let others know (documentation at ).

Cheers.— InternetArchiveBot  (Report bug) 03:34, 11 November 2016 (UTC)