hippocampus is broken. Even though the code is sparse, it is still redundant, in the following sense. There are 300,000 CA3 neurons in the rat, and even if there are 2% active at any given point in the environment, this still comes to 6000 neurons. Each neuron encodes a random subset of the elements that define the particular location in the environment, but each element is represented several times, so that any given element is likely still to be represented if only a fraction of the CA3 is preserved. Tsodyks, Skaggs, McNaughton, and Sejnowski have recently suggested what may happen during the peak of the theta cycle: here, as more and more hippocampal neurons are allowed to become active, the influence of collateral connections among CA3 neurons, rather than bottom-up connections from EC, may become more and more important in determining which CA3 neurons fire. Thus the neurons activated in the later stages of the theta cycle may reflect an associative pattern completion process. During spatial navigation, this process is thought to be responsible for that fact that neurons active in the peak of one theta cycle represent locations in space toward which this animal is moving.
As a more general comment, I have always found Pribram's perspective, emphasizing the distributed nature of memory storage, to be an extremely valuable and indeed largely valid perspective. It seems to me that the brain has found ways of capturing the positive features of this idea. Whether it does so by making explicit use of interference patterns remains a controvertial point, but the more fundamental insight is that representation and processing are indeed highly distributed, and this is especially true within the hippocampus.
McClelland, J. L., & Goddard, N. H. ( 1997). "Considerations arising from a complementary learning systems perspective on hippocampus and neocortex". Hippocampus, 6, 654-665.
Vaegha-Khadem, F., Gadian, D. G., Watkins, K. E., Connelly, A., Van W. Paesschen, & Mishkin, M. ( 1997). "Different effects of early hippocampal pathology on episodic and semantic memory". Science, 277, 376-379.
In this scheme of things, valuation is also influenced by more momentary utilitarian considerations provided by desires and the probability that these desires will be fulfilled.
Critical to this aspect of valuation is the case of T.J. presented by Charles Ahem and Frank Wood. Ahem and Wood framed their analysis in the terms used in memory research: the distinction between episodic, skill (procedural), and semantic (reference) types of processing. For reasons that, I hope, are apparent, I equate episodic processing with the current, momentary utilities (desirabilities and probabilities) that influence valuation; and I equate semantic processing with discriminations and preferences.
Given these equations, T.J. is important in that his semantic learning proceeded within normal limits and that his semantic memory is unimpaired, as is his skill in using it. As Zajonc ( 1966) has claimed on the basis of his studies, episodic processing appears to utilize a brain system independent of procedural and semantic processing.
Furthermore, episodic processing is not a necessary antecedent to skill or semantic (reference) learning or remembering. In these respects T.J. confirms the findings obtained for