"The more we know, the less we understand" -Anyone about anything

This sentiment captures how I (and presumably some of my classmates) feel as we’re learning more about the brain. We read papers about how some protein has been implicated in a biochemical process (e.g.). But knockouts only partially interfere with the process because that’s just the way the world works. We can only imagine what other mechanisms in the brain yield the same effective result and how many other pathways are affected by this specific protein.

When we analyze electrophysiology recordings, we generally base their conclusions on changes in the neurons’ firing rates (e.g.). And before I say anything more, I should state the obvious that this strategy has shed light on a lot of things, and I really don’t have any better ideas of how to extract meaning in the temporal information that we end up throwing away. However, I am not rarely preoccupied by wondering about the vastly complex computations that are only incompletely recorded. It’s not like the nervous system thinks “hm, the firing rates in these cells have risen by 80%, and therefore, I am looking at a face.” I can’t even fathom the physical computations that allow us to recognize a person’s face. In my depressed days, listening to lectures about how neural activity relates to decision making in cognitive neuroscience class, I wonder why we even bother, since our minds probably can’t understand the computations going on in our brain (to be fair, that didn’t drive their evolution).

Recently, I’ve come across literature on memory replay. The general idea is that a memory is encoded as a sequence of neuronal activation in the hippocampus. Then, during sleep, this neuronal sequence recurs as a hippocampal-independent cortical representation of the memory is formed (systems consolidation) so that the memory is stable in the long-term. This hypothesis is well supported by studies in animals. Ji & Wilson showed that a sequence of place cells that were activated while a mouse ran through a track was “replayed” during the sleep period that followed running through that maze several times.

“Recording began once … rats ran each trajectory at least 20 times.” Fortunately, we don’t need to repeat our experiences, let alone repeat them dozens of times, in order to consolidate what occurred. So could the observed “memory replay” be explained by another mechanism that necessitates this repetition? Using logic similar to Hebbian theory, if the same pattern of activity happens over and over again, this pattern will become more favored in the future.

"If the inputs to a system cause the same pattern of activity to occur repeatedly, the set of active elements constituting that pattern will become increasingly strongly interassociated. That is, each element will tend to turn on every other element and (with negative weights) to turn off the elements that do not form part of the pattern. To put it another way, the pattern as a whole will become 'auto-associated'..." (Allport 1985, p. 44, via Wikipedia)

I like to think of this as minimizing the energy expended by a system, which seems to explain a lot of things (from particle interactions to human interactions). In this case, the nervous system has changed its (anatomical or functional) configuration such that the energy expended to carry out this sequence of neural activations is less than a comparable sequence that has not occurred dozens of times. This makes evolutionary sense because it would be advantageous if the neural processes used to perform common actions were both metabolically cheap and ready to cascade. This might be completely wrong, but to me, its plausibility is comparable to that of replay’s hypothesized role in systems memory consolidation.

It’s difficult for me to imagine memory replay as the consolidation mechanism for a single event. I remember hearing a professor say “If you needed any more reason not to believe in God, here it is” (in reference to pain signals traveling down and then back up the spinal cord, here). I think that the neural activity present at the time of encoding (in the hippocampus or semantic-associated auditory cortex) was not necessarily similar to the pattern of activity that occurred when I remembered the statement later that day, or when it was eventually consolidated into my long-term memory. I think the background neural activity that was occurring at the time of that experience highly determined the precise neural activation that resulted in my professor’s insight entering into my consciousness, and that this perception was later recalled with different though not independent activity.

It would be convenient to naively assume that a sequence of action potentials present at encoding recurs at its later recall or translation to long-term memory. I just would be very surprised if that was the case. The dynamical systems that underlie our high level cognitive processes (and even relatively low-level ones in the retina) are so complex that I doubt that we can conceptualize them into our intuition. Maybe in the future we’ll engineer our minds to have that capacity.

In related news: this past week, I wrote a blog post as part of our graduate program’s weekly journal club. Basically the paper we went over showed a correlation to the amount of hippocampal-cortical interaction during a rest period and subsequent memory performance. While not convincing on its own, there’s a lot of work that implicates this interaction in memory consolidation, and it seems to be pretty solid. I just don’t think we’re ready to assert any more detailed mechanisms as truth. I wholeheartedly support that we feel free to imagine potential mechanisms, as long as we also anticipate our ideas to be met with skepticism.