Human Hardware: Working Memory
At the recent SMX show in Santa Clara I had the opportunity to present at a couple of sessions that explored the topic of user behavior. One of the things I said in one of them is that humans are more alike than we’re different. Because of this, there are some behaviors that a determined by hardwired traits. I call these Human Hardware issues. I’ve touched on some of these in past columns, but I’d like to dive a little deeper in this series.
Today, let’s set up the series with a quick primer on inherent traits and a look at the first topic, channel capacity in our working memory.
What Do You Mean I’m Not Unique?
As people start to dive into the human genome, it’s somewhat startling to find the lack of diversity in the human gene pool. As different as we all think we are, we actually are alike in many more ways. We share a remarkable similarity in our physiological and neurological makeup. Added to this is the fact that there are several inherent traits we all share, the result of thousands of years of evolutionary tweaking. There are deviations from the norm, but as a quick glance at any bell curve will tell you, for any given trait or characteristic of humanity, including intelligence, loyalty, physical strength, or the ability to juggle, most of us cluster around the center line, otherwise known as the norm. It’s the inherent limits of the vehicle we inhabit, our body.
And lest you start feeling too superior, we actually share 98.4% of our genetic material with chimpanzees, our closest evolutionary relative. There is more genetic diversity between two breeds of dogs than there is between us and the average chimp. In fact, apes and chimpanzees are genetically more divergent than chimps and humans. Try wrapping your mind around that one on your next trip to the zoo.
As we start looking at our success in predicting behavior, the peak of the bell curve for our target population is where we have to start. It helps to understand the human hardware issues, which form the foundation of our understanding of predicted behavior. From here, we can tilt our strategies to accommodate diversions in either direction from the norm.
Why is the human gene pool so shallow? It’s because we all come from the same place, a relatively small population of modern humans in Africa, some 150,000 years ago. Recent research has shown that genetic diversity lessens as we get further and further from Africa. And one particularly interesting study speculates that all blue eyed people come from the same common ancestor. Our family tree has remarkably few branches if you go back far enough.
All this serves to reinforce the fact that we’re not nearly as different as we think. If that lessens your sense of individuality, take heart in the fact that it makes your job as a marketer a little bit easier.
Working Memory: Standard Equipment
Next, let’s take a look at the concept of working memory. This is a vital piece of equipment on the human hardware list. If we didn’t have it, we wouldn’t be human. Working memory is where we think the deep thoughts that define us as humans. It’s our RAM, the place where we pull in pieces of information, work on them, make decisions, and then move on to the next problem.
Memory is a fascinating process in humans. Psychologists and, recently, neurologists, have spent a good part of their time dividing up memories into various categories, trying to understand how we stuff things into our brains. We have short term and long term memory, semantic, procedural, and episodic memory, explicit and implicit memory, and the lists go on. For the sake of this column, let’s restrict ourselves to working memory and long term memory.
Think of working memory as a white board. It’s a work space where we can quickly gather information for consideration. After we’re done, the slate literally gets wiped clean. The minute we stop thinking of something in our working memory, the neurons in use lose their charge because of various chemical reactions and the whiteboard gets erased for the next task.
We pull things into our working memory from various places. We can pull elements in from our long term memory or we can take in stimuli from our senses (sights, sounds, smells, tastes, or tactile sensations). There we mix and match them to come up with a concept. But it’s a bit more complicated than it sounds. This is not a purely rational process. For each long term memory we retrieve, or each stimuli that we process, there is corresponding emotional baggage. For example, our concept of “mother” is not just an abstract label, it comes charged with a flood of memories and emotions. The same is true for almost everything in our lives that eventually gets called into our working memory. These are often called implicit memories, because most of them bubble just under the conscious surface. These emotional constructs can dramatically color and bias our decisions and often we’re not even aware that it’s happening.
Retrieval from Long Term Storage
There is an interesting aspect to the retrieving of long term memories into the working memory space. As long as a memory is encoded in long term memory (called engrams), it’s inaccessible to us consciously (although, as I’ve said, implicit memories, which are also long term, can continually creep into our working memory without us being aware of it). To use a memory, we have to retrieve it back into working memory. And in that process, the memory gets rewritten. Every time we retrieve it, we in effect rewrite the memory, and when it gets encoded back into long term memory the new version gets stored. The “tone” of the rewritten memory will depend on what we’re doing at the time of retrieval, who we’re with, what mood we’re in, and any of hundreds of other factors. Think of memories more as easily shaped modeling clay rather than snapshots or video tapes.
The Friends and Family Memory Plan
There’s one more type of memory we rely on, and it’s one that sits outside our skull. This type of memory is particularly integral to our use of search engines and other online resources. It’s called transactive memory, a concept from Harvard Psychology professor Daniel Wegner. This comes from our tendency to be social animals. Because we like to travel in groups (families, coworkers, neighborhoods, communities), we start splitting up memory tasks and rely on others in the group to remember certain things. In this way, we don’t have to remember everything; we just have to remember who knows what. In our evolutionary dawn, some tribal members knew the right medicinal herbs to use, some knew the best places for berries and edible roots, some know how to find fresh drinking water, and some know how to hunt big game. Today, it’s more likely that the members of our social circle know how to tame a rampant operating system, program a recalcitrant PDA, find the cheapest airfares on the web, and program macros on Excel. Regardless of our environment, we have adapted to rely on a communal concept of expertise.
The Channel Capacity of Memory
As amazing as working memory is, it has its limits. It can only handle so much information at a time. As we absorb new information, to be considered together with information retrieved from long term memory, we have a limited number of slots to use. It depends on the type of information, but generally speaking, our limits, as discovered by George Miller in the mid 50’s, seems to be 7 “chunks,” plus or minus two. The use of the word chunks is deliberate here. We can handle information in chunks, which increases our ability to juggle complex concepts. The chunks can be condensed version of an idea, a label that represents a collection of thoughts that define the concept. So, if we’re trying to decide on a vacation destination, and we’re considering Kauai, Cozumel, Bermuda, Costa Rica, and Southern Portugal, each of those constitutes a chunk, complete with what we know (and feel) about each destination.
The ability to chunk gives us a way to dramatically extend the limits of our working memory. But, even so, there are limits to the amount of cognitive horsepower we want to expend for any given decision. For example, with the vacation example, there are obviously thousands of destinations we have to choose from. Do we rationally pull each of these possibilities into working memory and weigh the pros and cons? No, life is too short. This is an idea called bounded rationality, courtesy of Herbert Simon, and it goes hand in hand with a behavior called satisficing.
The Satisficed Short Cut
Satisficing means that we use heuristic, or rule-of-thumb, shortcuts to cut our choices down to a number that can comfortably fit in our working memory. And these rules-of-thumb are tremendously influenced by emotions. We have gut feelings, or intuitions, that can rapidly pare a list of thousands of candidates down to 5 or 6 that make the satisficing cut. Let’s go back to our vacation example. We’re sick of a long, cold winter and every time we step out the door, all we can dream of is a warm beach and a tropical breeze. At an emotional level, we’ve established our satisficing threshold. We then start finding the candidates that fit and put them in working memory.
Search as an Extension of Memory
Now, with the psychological and neurobiological foundations out of the way, let’s take our first look at how they impact our use of search engines. Again, going back to our vacation example, perhaps we’ve been to some of the destinations. In that case, we have rich concepts, complete with personal experiences, that we can use to help us make our destination. Perhaps we want to stick with the tried and true, or perhaps we want to go somewhere new. Then again, we might have never been to any of the destinations. Our working memory has precious little to pull into the available slots.
This is where a vitally important intersection with working memory, transactive memory, and search happens. Today, when we don’t have first hand knowledge of an option, we generally turn to a search engine. Our alternatives for collecting new information include word of mouth (asking people who do have first hand experience), tracking down promotional materials (from the vendors), or reading third party information (articles, reviews). Search gives us a short cut to all three.
With 150 Friends, Who Needs Memories?
This gives transactive memory a whole new meaning. Consider that our social circle in the past, a concept known as Dunbar’s number, generally was about 150 people. These are not mere acquaintances, but people we know something about. To use the definition provided by Dunbar himself, these are people who, if you saw them sitting by themselves in a bar or coffee shop, you would have no hesitation going over and joining them. Just like our working memory, our ability to maintain social relationships has finite limits. And, once again, there are evolutionary reasons for this. The average size of the human tribe 40,000 years ago on the African savanna? You guessed it, about 150 people. (By the way, I’ll revisit the Dunbar’s Number later in this series when we talk about online social networks and the social graph.)
So, to use transactive memory, we were generally limited to an extension of 150 people, give or take a few. We had to know people fairly well to know what their areas of expertise were. If we know that Frank goes on a winter holiday every year, chances are he’s a good choice to ask about our upcoming vacation plans. But, even as well traveled as Frank is, chances are he hasn’t been to all the destinations we’re considering. So, we still have gaps in our knowledge. The other problem with transactive memory in this situation is that we’re relying on Frank’s memory, and as we’ve seen, memories are highly subjective experiences, colored by personal tastes and beliefs. Frank’s idea of a great vacation spot might not be our idea of one.
Transactive Memory on the Web
But, with the internet, transactive memory took a quantum leap forward. Suddenly, our available list of resources included everybody who had an internet connection. And we didn’t have to know them personally, we just had to know the right words to search for on a search engine. We could tap into the expertise of thousands of people quickly, getting first hand experience on exactly the destinations we were considering. And, we could read between the lines of the post to see if the poster had similar tastes to us. I suspect that we’re rapidly sharpening the ability to pick up the personality of a poster through the tone of their posts.
Satisficing on the SERP
So, with all this information available and with the known limits of our working memory, how do we navigate through it? Again, the concept of satisficing plays a part. We use search engines to filter our choices. The refinement of the queries we use shows the progression of the satisficing process in our mind. We reach out, looking to access transactive memory through a search query. If our search doesn’t look like it’s tapping into the right expertise, we change the query. We quickly scan the results, seeing if we’ve hit the mother lode. We turn into “berry pickers” (to use Marcia Bates’ terminology) of expertise.
This satisficing behavior can explain a question that we’ve had about internet search since the beginning. Why do our search queries seem so rudimentary? Why don’t we make more use of advanced search functionality and more refined queries?
Again, think of the use of search as a satisficing filter. Remember, the whole point of satisficing is to lessen wear and tear on the brain, to reduce cognitive load. We don’t want to think. We’re saving that for when we short list our candidates. So, why would we spend precious extra seconds structuring exactly the right query? That defeats the whole purpose. Use a quick, rule-of-thumb query and see what you get. It’s all about short cuts.
“Chunking” a Page for Scanning
Satisficing behavior also extends to the actual interaction with the search page in a number of ways. First of all, we have capacity limits to the number of results we can put into our working memory for consideration at any one time. We don’t scan the page, load up all 18 or 20 results in working memory, and then make our optimal choice. Here too, the capacity of our working memory and satisficing play a part. In this case, the heuristic cut off is simply the fact that results show at the top of the upper left. This is where we expect to find the best results, so it makes sense that we take the top 3 or 4 results (this seems to be the typical channel capacity for most people with search results) and load these up for further consideration. A quick 4 or 5 second scan, looking for which of these is the best match to our query, and we either click or select the next group of 4 or 5. But everyone, yes, everyone, scans the same way at the beginning. We all start in the upper left corner and scan the first “chunk” of results. We’ve done hundreds (actually, thousands at this point) of eye tracking sessions, so take my word for it. This is typical behavior.
Here’s another flash. If top sponsored ads appear, we don’t ignore them. We give them a quick scan for information scent. The success of the ads depend totally on the relevance of them based on our intent. By the way, the user will also judge the success of their search experience in large part by the relevance of these ads. They’re the first thing looked at, which makes them very influential in the ultimate judgment of search quality.
Finally, we look to the search page to provide us candidates for our satisficed short list. If brands we expect to find at the top of the results aren’t there, we question the quality of the results set. Alternatively, if a brand we hadn’t previously considered shows up, it is often included in the resulting consideration set. We conducted a study with Google that shows that the spread in likelihood to consider can be as great as 40 percentage points, a pretty compelling reason to ensure present at the top of the search page.
So, for the first of our Human Hardware series, Working Memory, this is how this usually plays out on the search results page. In the next column, I’ll look at men vs women and their interactions online.
Opinions expressed in this article are those of the guest author and not necessarily Search Engine Land. Staff authors are listed here.