Señora

Behavioral Ecology doi:10.1093/beheco/arp102 Advance Access publication 17 July 2009

Why some memories do not last a lifetime: dynamic long-term retrieval in changing environments
a

Aimee S. Dunlap,a Colleen M. McLinn,b Holly A. MacCormick,c Matthew E. Scott,d and Benjamin Kerre Department of Ecology and Evolutionary Biology, University of Arizona, Box 210088, Tucson, AZ 85721, USA,bCornell Laboratory of Ornithology, Cornell University, 159 Sapsucker Woods Road, Ithaca, NY 14850, USA, cDepartment of Ecology and Evolutionary Biology, University of California, Santa Cruz, Santa Cruz, CA 95064, USA, dSt Croix Vineyards and Center for Complex Biological Systems, Stillwater, MN 55082, USA, and eDepartment of Biology, University of Washington, Box 351800, Seattle, WA 98195, USADownloaded from http://beheco.oxfordjournals.org/ by guest on January 29, 2013

Memory is a fundamental component of learning, a process by which individuals alter their behavior through experience. Although memory most likely has explicit costs such as synaptic maintenance and metabolic demands, there are also implicit costs to memory, in particular, the use of information that is no longerappropriate or is incorrect. Specifically, the period of retrievability for memories, or ‘‘memory window,’’ should be sensitive to the rate of environmental change of information stored in memory. Much empirical data suggest that memory length—this period of retrievability—changes with both the age and state of the individual. Here, we use a dynamic programming approach to examine how optimal memoryretrieval might change within the lifetime of the individual learner. We find that optimal memory length varies with both age and state (e.g., energy reserves) of the organism and that features of the environment determine how this change in memory occurs. In our model, retrieval decreases as the environment becomes unreliable but roughly increases with the cost of living. Cost of living interacts with thestate of the organism: with high cost of living, an organism in a very poor state should have a long memory length, but an organism in a very good state with low costs of living should have a short memory length. Finally, we find there are circumstances where it is optimal for memory retrieval to decline toward the end of the lifetime. Because this framework does not incorporate inevitabledegradation of neural mechanisms, this result implies that memory loss with age might actually be adaptive. Key words: dynamic programming, environmental variability, learning, memory, optimality, stimulus reliability. [Behav Ecol 20:1096–1105 (2009)]

L

earning is responsible for many of the flexible behaviors animals use when adjusting to changing environments. Ultimately, learning depends on ananimal’s ability to store and retrieve information about its world, that is, learning is tightly entwined with the memory of that information (Bouton 1994; Shettleworth 1998). Consequently, the value of memory often relates to the value of learning. On a simple level, learning is valuable when information stored in memory (e.g., a good response to a stimulus) remains useful over time (e.g., the sameresponse to the same stimulus continues to be good). However, learning and memory lose value when the environment changes in ways that make learned responses unreliable. Indeed, possessing memory might be costly if it repeatedly leads an organism to employ inappropriate behaviors (e.g., if the environment changes extremely rapidly making remembered responses perpetually inappropriate). This issimilar to the effects of ecological or evolutionary traps, where previously reliable information leads to a maladaptive outcome because the situation has changed (Schlaepfer et al. 2002). If memory is evolutionarily adaptive, one would expect that various properties of memory (encoding, consolidation, retrieval, etc.) would be tuned to the historical rate of environmental change. Several...