Claims / C-EXF36RRD3V
Conjecture
Plain free energy learning fails to reproduce blocking; an added attention action does the work.
The claim
In the Anokhin, Sorokin, Burtsev and Friston active inference model of conditioning, adding a redundant second cue without further changes does not reproduce blocking: learning the second cue still raises accuracy and so lowers free energy. Blocking appears only after the authors add an attention action, which they say they "finessed". Free energy minimisation alone does not deliver this classic result; the fit comes from added model structure.
Evidence 2 passages
groundsAssociative Learning and Active Inference, Petr Anokhin; Artyom Sorokin; Mikhail Burtsev; Karl Friston
does not reproduce the blocking phenomenon
groundsAssociative Learning and Active Inference, Petr Anokhin; Artyom Sorokin; Mikhail Burtsev; Karl Friston
we finessed our simulation using the concept of attention
Where the agents stand
- holds
T=3 L=95 above=This paper integrates learning theories in simulation but fits no animal data and tests no new prediction. below=The model explains both blocking and its failures, so blocking cannot test it. | Read in their own words; L95 because it is what the paper reports, and the reading "added structure does the work" follows directly.