Backward Inhibitory Learning in Honeybees: A Behavioral Analysis of Reinforcement Processing

One class of theore t ica l accounts of associative l ea rn ing suggests tha t re in forcers are p rocessed accord ing to l ea rn ing rules tha t min imize the predic t ive e r r o r be tween the expec ted s t r eng th of future r e i n fo r cemen t and its actual s t rength . The omis s ion of r e i n f o r c e m e n t in a s i tua t ion w h e r e it is expec ted leads to i nh ib i to ry l ea rn ing of s t imuli indicat ive for such a v io la t ion of the predic t ion . There are, however , resul ts indicat ing tha t i nh ib i to ry l ea rn ing can also be induced by o th e r mechan i sms . Here, we p re sen t data f r o m ol fac tory r eward cond i t ion ing in h o n e y b e e s tha t show tha t (1) oneand mult iplet r ia l backward cond i t ion ing resul ts in cond i t i oned inh ib i t ion (CI); (2) the inh ib i t ion is m a x i m a l for a 15-sec in terval be tween US and CS; (3) the re is a n o n m o n o t o n i c d e p e n d e n c y on the degree of CI f rom the US-CS in terval du r ing backward pair ing; and (4) the p r i o r associa t ion of con tex t s t imuli wi th r e i n f o r c e m e n t is no t necessa ry for the d e v e l o p m e n t of CI. These resul ts canno t be exp la ined by models tha t o n l y min imize a p red ic t ion er ror . Rather, t hey are cons i s ten t wi th models of associative l ea rn ing that, in addit ion, assume tha t l ea rn ing depends on the 3Corresponding author. 4present address: European Media Lab, 69118 Heidelberg, Germany. SMartin Hammer died on September 24, 1997, in Berlin. t e m p o r a l over lap of a CS wi th two processes , a fast exc i ta to ry and a s low inh ib i to ry one, bo th evoked by a re inforcer . The f indings f rom this behav iora l analysis of r e i n f o r c e m e n t p rocess ing are c o m p a r e d wi th the k n o w n p roper t i e s of an individual , ident if ied n e u r o n involved in r e i n f o r c e m e n t p rocess ing in the bee bra in , to fu r the r u n d e r s t a n d the m e c h a n i s m s unde r ly ing predic t ive r eward learning.