Hack87.Boost Memory Using Context

Hack 87. Boost Memory Using Context

Your memories aren't stored discretely like objects in a filing cabinet; rather, they are interleaved with other things in memory. This explains why you're good with faces but not with names, why you should go back to your hometown to better remember your school days, and maybe even why you dream, too.

Human memory is not organized like a filing cabinet or a hard disk drive. In these storage systems, each memory is neatly indexed and stored so that it doesn't affect any other memory. The items in a computer memory don't affect processing unless they are explicitly retrieved, and to retrieve them, you have to consult an index to work out where they are. If you don't know where they are or if you don't have the right tag by which to access the files, you're out of luckyou're stuck with a brute force look through each file, one by one. The same holds for finding related itemsyou do it through some form of indexing system or again resort to a brute-force search. The system is content-blind.

But human memory is even further unlike any filing cabinet or computer memory system. This is the fundamental difference: human memories are stored as changes in the connections between neurons, the self-same neurons that actually do the processing.

So there are no passive storage locations: the processing-storage distinction fundamental to conventional computer architecture1 doesn't hold. Instead, memories about things are stored by the same units that are responsible for processing them. As you look at a face, your brain doesn't need to send away for information on whether you've seen the face before, and it doesn't need to store or index that face so that it can be recognized later. The ease with which that face was processed by your neural units provides a signature that can be used to calculate familiarity [Hack #83] . If you see the face once, it makes it easier for the neurons that respond to that particular combination of features to respond together, effectively acting as a key for recognizing it later.

So it should be clear why recognizing faces is easier than recalling names. When recognizing faces, your brain is presented with some input (a face) and can tell if it is familiar just by checking whether the neurons for representing that face easily coactivate. (If all the neurons representing a face activate together easily, that means they've activated together in the past. That is, you've seen the face before.) For recalling the name, you have to recognize the face and then hope that the association with the word information you heard at the same time (the name) as you first met the face is strong enough to allow that to be activated. It's a different process (recall versus recognition) in a different modality (image versus words); no wonder it's so much harder.

Of course, if human memory were organized like computer memory, then recognizing faces would be an equivalent task to recalling names. Both would involve checking the input (the other person's face) against everything you've got stored. If you've got the face stored, bingo!you recognize it. And the information you retrieve to recognize it would be automatically linked to the name, so recalling the name would be just as easy as recognizing the face. Unfortunately, on the flip side, recalling a face would be just as difficult as recalling a name.

T.S.

The second important consequence of this fundamental difference is that memories are distributed between many neurons, all of which are involved in storing many other memories. This means that memories aren't stored independently of one another; so, learning something new can interfere with your memory of something old (and, of course, the things you already know affect what you remember about new material).

9.8.1. In Action

Forgetting something isn't just a matter of information falling out of your brain, as if your brain were a filing cabinet turned over. Traces remain of any information that is forgotten. This is why relearning things is easier than learning them for the first time. And because memories are fundamentally entangled with on another, remembering or relearning something brings related memories closer to the surface too.

One way of showing this is to relearn a subset of some set of knowledge that you have previously learned and forgotten. The effect of relearning should transfer to other memories in the same set,² benefiting all the associated memories, not just the ones you deliberately relearn.

The vocabulary of another language is a good example of something that you learn and use all together, and then forget. In my case, I've forgotten the French I learned at school. So, to demonstrate to myself that the entanglement of memories would produce a transfer effect I performed the following experiment: I took a list of 20 common verbs and tested myself to see how many I could remember the French for. It turned out I could remember the translation for 8 words. I then found out what the remaining 12 French words were. This was the relearning phase. If I wanted to be more thorough, I could have relearned some nouns and adverbs as well, but I didn't. Next, I tested myself on 20 common adjectives. This time I got 13 English-to-French translations right. After only a few minutes thinking about French again it was coming back to meI was more than 50% better at the second set of words, despite being just as unpracticed at these as the first set. Retrieving one set of French vocabulary from my memory had strengthened the associations required for me to recall more and more.

9.8.2. How It Works

Think of the fundamental currency of memory as being associations, not items. This is core to the design of the whole system. It means that content can be accessed by anything associated with it, not just any single arbitrary tag. Human memory is content-addressable. This is unlike computer memory, where you access stuff only through the arbitrary tags you use to keep track of information and that you decide on at storage time (like filenames). The reason Google is so popular is that it gives us content-addressable memory for the Internet. You can type in pretty much anything you remember about the contents of a web page and it comes up in the results. And, also like the Internet, much of the meaning of memory is to be found in the connections and associations, which shift and recombine independent of the content.

A famous psychology experiment involved divers learning lists of words either on the docks or underwater and then being tested either on the docks or underwater.3 Those who scored highest on the test were those who were tested in the same situation in which they learned the material (i.e., tested underwater if learned underwater, tested on the docks if learned on the docks). Those who scored lowest were those who switched contexts from learning to recall. This demonstrates the automatic encoding of context along with memories [Hack #86] and provides some justification for the advice I was given as a student that if you learned something when drunk you should go into the exam drunk. (It may well be true, but it might also be better not to have been drunk in the first place.) Being able to recall better in the original situation is one consequence of your context being automatically laid down in memory. Another consequence is the transfer effect, as shown in the preceding "In Action" section: memories are tangled up with other memories of the same type and themselves constitute a kind of memory context. Remembering one set of knowledge puts you in the right context, and the associated memories follow more easily.

The need to interleave many different memories in the connections between neurons may be one of the functions of sleep. It's vital to store new memories in the same networks of association as used by old memories, otherwise you'd have no way of moving between them. But at the same time, it's important not to overwrite old memories. There is evidence that the need for this process, called interleaving, may explain some features of our memory systems, and there's also evidence that it may occur during sleep as dreams or part of dreaming.⁴

9.8.3. End Notes

1. The Von Neumann architecture separates processing from data and code (http://en.wikipedia.org/wiki/Von_Neumann_architecture).

2. Stone, J. V., Hunkin, N. M., & Hornby, A. (2001). Predicting spontaneous recovery of memory. Nature, 414, 167-168.

3. Godden, D., & Baddeley, A. (1975). Context-dependent memory in two natural environments: On land and underwater. British Journal of Psychology, 66(3), 325-331.

4. McClelland, J. L., McNaughton, B. L., & O'Reilly, R. C. (1995). Why there are complementary learning systems in the hippocampus and neocortex: Insights from the success and failures of connectionist models of learning and memory. Psychological Review, 102, 419-457