6

ACT-R was the first big cognitive model and excels at modelling human behavioral data quite accurately.

Spaun, which is the world's largest functional brain model, took a lot of ACT-R's insights and applied them, most notably in it's model of the basal-ganglia-thalamus loop. More importantly, it did so in biologically plausible manner while ACT-R has been struggling recently to map it's functions onto brain areas and functionality.

That being said, ACT-R still seems to have an extremely active community and Spaun's developers are mostly restricted to the lab from which it originated. An obvious reason for not converting is if Spaun's range of cognitive functionality is missing some components of ACT-R.

What functionality of ACT-R isn't found in Spaun?

Seanny123
  • 8,853
  • 3
  • 25
  • 61

2 Answers2

4

Both ACT-R and Spaun are modular, and could be extended to include capabilities of each other. Comparing the functionalities of the two architectures by a simple checklist is not the most appropriate way to compare them. Here are some points to consider:

1) ACT-R is primarily symbolic, whereas Spaun is a neural network. ACT-R is actually making a theoretical claim by committing to a symbolic architecture. This is one of the most famous debates in cognitive science; the upshot is that the two approaches each have limitations, and they go about solving problems in different ways. For a good starting point in this debate, see Fodor & Pylyshyn (1988).

2) Architecture aside, different algorithms can be implemented to solve the same problem. Maybe both Spaun and ACT-R can do simple declarative memory retrievals, but do they predict the same response times and accuracies for a given retrieval? I suspect they don't.

3) As mentioned in your answer, Spaun does not include a of long-term memory but does include working memory. In ACT-R, long-term memory and working memory are not distinct concepts; working memory can be construed as a subset of items in memory that are highly activated. There is no discrete distinction. This suggests that the structure of components in ACT-R is fundamentally different than Spaun.

Community
  • 1
Jeff
  • 5,488
  • 29
  • 44
  • You're right that comparing them via a functionality checklist is probably a bad idea and I'll make sure to edit my question to reflect that. However, I disagree with your statement that Spaun can be summarized as a neural network and thus different from the symbolic approach of ACT-R. Spaun is more of a hybrid of connectionist, dynamicist and symbolic approaches. It uses neural networks to approximate functions and dynamics, but at the highest level it's still manipulating symbols with the SPA. – Seanny123 Nov 11 '14 at 18:32
  • 1
    I am not very familiar with Spaun, but you're right, I am oversimplifying things. ACT-R is not purely symbolic either. My point is that these are actual theoretical claims about the nature of the mind. – Jeff Nov 11 '14 at 18:39
3

Long-term memory storage and forgetting are the two main features that are found in ACT-R, but are missing from Spaun. Currently, Spaun only has a working memory and a fixed long-term memory; it lacks the ability to store new items in a long-term memory.

This is currently being tackled by trying to create better hippocampal models and encoder learning rules. However, there is clearly still a lot of work to be done.

Seanny123
  • 8,853
  • 3
  • 25
  • 61