Probing the limits to memory

June 3, 2014 by John Hewitt report
Credit: Photo Researchers, Inc.

(Medical Xpress)—There is no mystery in the way that computers store information. It is straightforward to determine how much memory they have and what is contained in that memory. Brains don't really have memory in this traditional sense. What they do have, are various adaptive systems that give the appearance of a limited computer-like memory as a side-effect of their primary functions. Two papers that recently appeared in PNAS describe the use of fMRI to explore the limits of human memory. In particular, they address roles of category and context in determining if something might be remembered, or forgotten.

The first paper looked at performance when subjects were asked to remember several things at once. When the items to be recalled were from different categories—faces, bodies or other objects—the researchers found that the subjects did better. The scans revealed that the degree to which the mixed catagory recall was improved could be predicted by the degree to which the fMRI neural response patterns they obtained were also different.

The second study focused on recall performance when various objects appear in different contexts or scenes. Here the researchers found that when items don't appear in a context that is expected, the memory for that object is weakened. In this case it is hypothesized that the brain automatically prunes what it predetermines is an invlaid memory.

One major difference between our animal memory and say, Google's memory, is that while we constantly forget things, Google never does. On the other hand, while it is impossible for us to intentionally forget something on demand (at least while keeping everything else intact) Google could erase any specific memory it might so choose. Clearly in our case at least, all memories are not created equal. Despite that fact, we are neither jury nor judge of our own memories.

By it's own admission, in a typical search lasting 0.2 seconds Google consumes 0.3 Wh (roughly $0.0003 per search) for the whole job of indexing and retreiving the query. Based on the average adult requirement for 8000 kilojoules of food per day, Google concludes that a search is equivalent to the energy a person would burn in 10 seconds. If we take a simple task, for example the question, "Do you know X ?" what happens next inside our heads would put any Google search to shame.

The key thing in the two PNAS papers for us here is that category and context are not just incidentals of memory, they are memory. Without them there is no memory for us, and for Google, there is only a future of expontentiating storage requirements. Clearly there is more to life than text, audio, and video. The task ahead for computer engineers is to learn not just how to extract the outlines that define an image, but how to extract the outlines that define reality, and the events that occur within it.

If we are shown a face, as in the experiments done above, we can instaneously scan our whole history of existence to estimate whether we know it or not. We can also estimate whether we know their name within a few more micromoments. If the name is not recalled immediately, we would then attempt to summon it by creating additional context, quickly trying out sounds or images until one fits. Failing that, we can dig deeper into emotion and other meaningful contexts that draw forth how the face makes us feel, whether it might be someone favored or disfavored.

Failing all these efforts we might abandon hope at ever remembering the name. Only then we might paradoxically be given the perplexing experience of the name coming immediately to attention as another part of our brain that been earstwhile supressed in the effort had it there for us all along. If we are ever to build so-called memory implants that hope to do more good than harm, we will have to at least have an inkling as to how intact brains remember.

Explore further: New research shows memory is a dynamic and interactive process

More information: 1. Processing multiple visual objects is limited by overlap in neural channels, Michael A. Cohen, PNAS, 2014. DOI: 10.1073/pnas.1317860111

2. Pruning of memories by context-based prediction error, Ghootae Kim, PNAS, 2014. DOI: 10.1073/pnas.1319438111

Related Stories

What happened when? How the brain stores memories by time

March 12, 2014

(Medical Xpress)—Before I left the house this morning, I let the cat out and started the dishwasher. Or was that yesterday? Very often, our memories must distinguish not just what happened and where, but when an event occurred—and ...

Outside the body our memories fail us

March 10, 2014

New research from Karolinska Institutet and Umeå University demonstrates for the first time that there is a close relationship between body perception and the ability to remember. For us to be able to store new memories ...

New study suggests a better way to deal with bad memories

April 18, 2014

(Medical Xpress)—What's one of your worst memories? How did it make you feel? According to psychologists, remembering the emotions felt during a negative personal experience, such as how sad you were or how embarrassed ...

Recommended for you

A turbo engine for tracing neurons

April 27, 2017

Putting a turbo engine into an old car gives it an entirely new life—suddenly it can go further, faster. That same idea is now being applied to neuroscience, with a software wrapper that can be used on existing neuron tracing ...


Adjust slider to filter visible comments by rank

Display comments: newest first

1 / 5 (1) Jun 03, 2014
The Study of Memory Has Two Parts:

(1)The Systems Problem of Memory:
Where in the brain is memory stored?

(2)The Molecular Problem of Memory
How is memory stored at each site?


What have you been writing all these years?
Where have you been?
Where did you look?
You have a lot of catching up to do.
not rated yet Jun 03, 2014

No memory is stored, explanations are created on demand and on the fly.
The question then is how are explanations or answers withdrawn from the structural edifice we assemble as brain as queries are imposed upon it?
Does every ganglion cell in the retina try to represent the whole world for the brain, every neuron try to build each memory? At life's end does each cell release its grip on it's neighbor's in a death throw of futile self survival, or grip ever tighter?
1 / 5 (1) Jun 04, 2014
No memory is stored, explanations are created on demand and on the fly - JH.

...from the structural edifice, no doubt.

Neurons are last on the list to depart before clinicians place the label of 'death' on life forms with neurons. Most of the cells you started out with are long gone. Replaced. Neurons don't have that luxury. So neurons repair what they have - DNA - for better or worse till the bitter end.

The source for the creation of explanations is information. The principle of conservation is the postulate of information for quantum information theory - a physical theory.
Classical information theory has no physical constraints.

not rated yet Jun 10, 2014
@johnhew: Oh, that reminds me of Dennett's "agents all the way down" explanation of the mind. ( )

I don't know if there's been any further work on this hypothesis since then, but it seems right up your alley. While I sometimes disagree with what you choose to emphasize, you're the best writer I've seen at working out the full implications of provocative but sketchy hypotheses.

Please sign in to add a comment. Registration is free, and takes less than a minute. Read more

Click here to reset your password.
Sign in to get notified via email when new comments are made.