Memory for computers is getting pretty large, but it is still based on basically the same system that it was several years ago. They have just gotten better a fabricating them.

It is an interesting question to ask whether we could store memories in alternative substrates such as biological ones. The idea is an intriguing one, particularly when we are talking about neurons, because while the capacity of biological networks isn't infinite, it is pretty damn large. Just think of the system for long-term memories in your brain. The fact that you can remember your first grade teachers name despite all the intervening information is pretty astonishing.

Anyway, Baruchi and Ben-Jacob, from Tel-Aviv University, wanted to see if you could store information for long periods in cultured neurons. Culturing neurons is actually not a difficult thing to do; neurobiologists can even buy kits to extract them nowadays. You just mince up a part of an animals brain -- usually a rat or a mouse -- and culture the cells in a special media for several days that kills off all the other cells.

What you see when you do this with neurons is really cool. At the beginning the cells look like...well...cells. They don't have processes. They look normal.

Over time, however, they will regrow their axons and dendrites and begin to spontaneously talk with the neighboring neurons. Activity in neuron cultures appears spontaneously.

What is this activity like? Is it coherent? The manner of spontaneous activity in neuron cultures can be described as what I suspect sex with an old person is like -- passionate but over with relatively quickly. Basically, a foci develops in the random network of neurons that makes them all go crazy for a second. The authors describe this activity as a synchronized bursting event:

Many investigations showed that the activity of stand-alone cultured neural networks is marked by the spontaneous formation of synchronized bursting events -- short time windows (~200 ms) of rapid neuronal firings separated by long time intervals (seconds) of sporadic firing. From the dynamical systems perspective, the SBEs are collective modes of activity each with its own specific complex spatio-temporal patterns of neuronal firing -- the activity starts at a specific location
the focus or initiating point and propagates along a specific trajectory in space. It was suggested that the spontaneous formation of SBEs reflects the action of innate system level mechanisms that are associated with in vivo mechanisms of memory and learning. The idea is that the spontaneous formation of SBEs in the activity of stand-alone cultured networks implies that they can be regarded as memory templates or precursers of the in vivo memory templates.

Each SBE has a characteristic pattern in space and time. The experimenters wanted to see whether they could make their own SBEs on top of the ones that formed spontaneously by injecting a small amount of drug into the culture.

The apparatus they use is in the Figure at the top (which is from the paper by the way). The culture has electrodes mounted at particular points so that they can decode the activity. A micropipette is used to inject the drug, and the neurons can be observed through a microscope.

They discover that they can create artificial SBEs that are stable for long periods. What I mean by stability is not that they are active for a long time, but rather that when you inject the drug you get see it once, and if you inject the drug again you see that same representation is stably represented:

The results shown...illustrate that by local chemical stimulation it is possible to create several additional distinct classes of SBEs, without erasing the existing ones....[E]ach class of SBEs has its own characteristic spatiotemporal pattern of activity propagation. The activity of each class starts at a specific location, close to the location of the chemical stimulations that created this class
accept for the spontaneous SBE, and then propagates along a specific trajectory. These results indicate that our method enables us to imprint a collective mode with a predesigned spatiotemporal pattern. After the stimulation session, the network continues to spontaneously generate the newly created classes of SBEs (memories) for long periods of time (more than 40 h)...

What does this research say about the likelihood of making microchips out of neuron cultures? Well, I imagine that is far off for several reasons. 1) Neuron cultures are expensive and sometimes unpredictable creatures. Silicon chips may always be cheaper to produce. 2) Whether or not you can trigger a specific representation in the culture, you still need some sort of means to decode that and convert it into a form that you can use. I feel like neuron cultures would always require some sort of hash-table to convert them out of neuron-speak and into computer-speak.

In any case, this mode of storage might be useful for some type of representations or computations. It would also help if we could prefabricate neurons in culture, but that still hasn't been accomplished completely. Still, interesting stuff to follow...

Hat-tip: Eurekalert