Its kind of hard to do the measurements necessary to answer that question without being able to accurately and objectively measure actual human beliefs (which may not be the same as self-reported beliefs.)

Ok. What about using inaccurate, subjective measures? How well does it model self-reported beliefs?

What resolution are you looking for? All models fail to be the thing they're modelling.

The first and obvious answer is that "this is software running on hardware, as apposed to being biologically-based".

It might sound like I'm being facetious but your question is really that broad.

"What sort of predictions that this makes are contradicted by observation"?

This is what I meant. Thanks for providing the clarification I didn't think to include.

I thought the clarification to be self-evident, for what it's worth.