Thursday, December 25, 2014

The Myth of Multitasking: Serial Attention

"Achtung!" --U2

Cognitive Science is awesome for (at least) two reasons. First, the field likes to debate all sorts of binary questions (e.g., Does the mind use symbols or not? When reading, do we process surface features or semantic features?). Second, cognitive scientists come up with all sorts of crazy metaphors to better understand the complex inner-workings of the mind. The topic today is awesome for both reasons. Early investigations into attention tried to answer binary questions, such as: Is attention parallel or serial? Does information get selected for deeper analysis early in the process or later? Also, they came up with some pretty cool metaphors to describe attention, such as switches, filters, attenuators, and spotlights. 

Before we dive in, let's make a distinction between information selection and information processing. We've all heard the phrase "selective attention" (or, the close cousin "selective hearing"). It seems that some people have an amazing ability to pay attention to only one thing at a time. For example, if your roommate is texting her friend, she might not even notice when you ask a direct question. When texting, your roommate has decided, consciously or not, to select the information emanating from her phone. Processing information, on the other hand, refers to the analysis and response to that information, such as a replying to a text message. 

Now that we've laid the groundwork, let's look at the fascinating world of auditory and visual attention! 

"Were you listening to me, Neo? Or were you looking at the woman in the red dress?" --Morpheus

So we know that deep down, at its core, the attentional system is massively parallel. It doesn't matter how engrossed you are in a task, you will respond to a very loud siren and a red flashing light. Not much analysis needs to take place because your attentional system is always on high alert to keep you alive. If something threatening comes your way, odds are your attention will be captured and you will respond immediately [1].

Going beyond all the loud noises and lights, the attentional system also has to be designed to help you select and process information. Most evidence suggests that there is a bottleneck somewhere in the attentional system such that, once selected, we can only process one stream of information at a time. In early auditory attention experiments, researchers asked people to listen to a recording where they shadow, or repeat, the message in one ear and ignore the message played in the other ear. You can try it for yourself here. After the task was over, the researchers asked about the information in the ignored ear. Most could say if the voice was male or female, and give other surface characteristics of the sound, but not much more than that (i.e., the content of the message). 

A similar finding has also been demonstrated for visual attention. Here is one of the coolest demonstrations of this phenomena. You need to experience it for yourself.


Simon's Visual Cognition Lab


This is a very powerful demonstration of the effect of a goal (e.g., count the number of passes) on the selection of information. It also demonstrates that we can only process one stream of information at a time. 

A STEM Example

I'll be honest, selectively attending to a single stream of information is one of the most fundamental principles of education. You can't learn what you don't pay attention to! That seems almost too simple to state, but it seems like it's easy to forget. 

Knowing that we are serial processors is useful when evaluating educational applications. A great example is duolingo, which is an app that teaches (or re-teaches) a second language. Each task is presented in a simple interface where the learner concentrates on only one goal at a time. Once that task is finished, the learner progresses to the next task. Again, the app keeps things simple and doesn't try to split the student's attention across too many sources of information. 

Share and Enjoy! 

Dr. Bob

For More Information

[1] The lone example I can think of are some Tibetan monks who are able to get deep into a meditative state where they do not respond to loud noises (see Chapter 1 in Search Inside Yourself for a description).

Thursday, December 18, 2014

Better Than Soup: Chunking

"Sloth love Chunk!" --Sloth

In a previous post, we talked about the severe constraints on working memory. Early estimates of the capacity of working memory started out around seven (plus or minus two) items. That translates into looking up a phone number in the phonebook (remember those?), walking over to the phone, and dialing the number. Unfortunately, seven seems like a very low number. In fact, later estimates put working-memory capacity around four items. Four items?! But that seems crazy low. Fortunately, there is a way to expand your working-memory capacity through a process called Chunking.

Does chunking really work? If it does work, what are the limits? How far can we stretch this strategy? 

Does Chunking Work?

How do we know that the brain is able to aggregate or "chunk" information? What is the evidence? To generate some evidence, this interesting study asked a couple of "volunteers" to memorize the position of chess pieces on a chessboard [1]. There were three types of participants. The first was a world-renowned chess master. The second was an intermediate player, but wasn't anywhere near the ability of the first player. The third person knew how to play chess, but was not ranked in any official capacity. The scientists showed them the configuration of a couple of chessboards that were in mid-game. The twist was that some boards were actual games, while the other boards had the same number of pieces, but they were randomly placed across the board. Before I tell you the outcome, what do you think they found? 

As you probably guessed, the chess master's memory for the position of the chess pieces was vastly superior to the intermediate and novice player's memory. What wasn't totally obvious, however, was how well they did relative to each other on the random boards. It turns out that they were all equally the same. This suggests that the chess master wasn't looking at individual pieces on the actual mid-game boards. Instead, he was aggregating the pieces into groups (e.g., a "castling" position). I love this study because it's an elegant demonstration of the process of chunking.

"Take It To the Limit" --The Eagles

The best answer to the question of limits comes from a study that attempted to train someone to expand his working-memory capacity [2]. Going into the experience, the person that was selected to endure the rigorous training regimen was a runner. That means he was well versed in thinking about numbers in terms of running times. He was able to chunk digits into running times. For example, 4:32:8 is an average time for a men's marathon. The runner worked for many training sessions by adding more and more complex retrieval structures. At the conclusion of the study, the participant was able to correctly recall 79 numbers. Impossible!

What does that mean for us ordinary mortals? First, this person wasn't special in any obvious way. That means that any one of us could also learn to memorize 79 digits if we were willing to put in the time and effort. Second, learning to memorize digits of numbers seemed to apply only to digits. In other words, the participant wasn't able to apply what he learned to memorize state capitals or other forms of information (e.g., letters). Finally, it also means that, although we have severe limits to our cognitive capacities, they can be overcome either by cognitive strategies and/or good, old-fashioned hard work (i.e., "deliberate practice"). 

A STEM Example

I'll be honest. When I took Physics in college, it was brutally difficult. Not because of the math (it was a non-calc version), it was hard because it seemed like each new concept arrived from out of the blue. Rotational kinematics seemed to have nothing to do with linear kinematics  Sure, the form of the equations seemed to have something in common, but they were largely taught as disconnected facts. 

Fast forward several years to my post-doc. I was blessed to work with a real physicist who pointed out to me that Physics is easy because you only need to know a few "first principles." From there, you can derive many other facts That hit me like a bolt of lighting. Once someone took the time to sit down with me and demonstrate the inner-connections, Physics didn't seem so hard. I don't want to trivialize education, especially for difficult topics, but the whole process can be made more simple (and perhaps fun?) if the material is presented as a sequence of ever-expanding chunks of information. 

Let's take velocity as an example. To build up to this advanced topic, it helps to start with our intuitive understanding of speed. Most of us have ridden in cars and talked about the measurement of speed in terms of "miles per hour." Once that gets translated into a symbolical representation (s = d/t), you can then expand it to include the concept of change (i.e., delta). Now the equation becomes s = Δd/Δt. Not a lot has changed, and that's a good thing because the student needs to see the equation, not as something new, but slightly expanded. Then you can expand the notion of the delta: Δd = d_final - d_initial. Plug this back into the equation, and you get a slightly more detailed expression. Again, each step is small and needs to be seen as a single chunk of information. 

Share and Enjoy! 

Dr. Bob

For More Information

[1] The chess study was conducted by a pair of researchers at Carnegie Mellon University (CMU) in the early 70s. The first author, Bill Chase, was my graduate-student advisor's late husband. I never had a chance to meet him, but he is a legend in the field of cognitive psychology. On the other hand, I did have the good fortune to take a course from the second author, Herb Simon. It was a fascinating course, and he gave probably the hardest final exam I have ever taken in my life. It had a single question: "Describe a computationally plausible model of cognition." We then had about three hours to provide an answer. 

Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55–81.

[2] Training someone to expand his working-memory capacity took 230 hours of practice! His training was conducted  by K. Anders Ericsson, who we will hear more about in subsequent posts. The original article can be found here

Ericsson, K. A. (1980). Acquisition of a memory skill. Science, 208(4448), 1181–1182.

Thursday, December 11, 2014

Crunched for Space: Working-Memory Capacity

Mental Scarcity

This week, we're going to talk about something so fundamental to cognition that it is easy to overlook. To demonstrate the concept, let's play a simple game:

I'm going to give you a list of numbers, and your job is to repeat them back to me, in the order you saw them. Okay, maybe that's not so simple, but I know you can do it. Ready? Click "Play" to see the list:

Ok, quick, what was the list of numbers!? Did you get them all? If not, don't beat yourself up. I may have been a little unfair because I threw in twelve separate digits. According to this very famous paper [1], you should have only been able to repeat back 7 digits (give or take two) [2].

In other words, the amount of information that you can cram into working memory is severely limited, and we refer to that as your own personal Working-Memory Capacity. First, the bad news: the amount of information we can focus on and use at any one time is very small. Now, some good news: you can use various tricks to expand your working-memory capacity. 

One of the tricks is called Chunking. When I do this demonstration with large groups of people, there's always at least one person who can repeat back the entire list in order. How do they do it? Are they superhuman? Do they practice memorizing numbers all day long? Maybe. But the most likely explanation is that they don't see each digit as a single thing to be remembered. Instead, they focus on grouping the digits together into larger chunks of information. 

Here's the list again: 

1  4  9  2  1  7  7 6  1  9  4  2

Do you notice any patterns in the data? Let me give you a hint: Think about important dates in American history. How about now? Anything emerge? Instead of trying to remember 12 separate digits, now all you have to remember are three years: 1492, 1776, 1942. That's a lot easier, right?

A STEM Example

How does this play out in education? The most obvious example that I can think of is when a student is trying to learn a mathematical formula to calculate something complex, like the circumference of a circle. When a math teacher introduces the idea, there are a bunch of new concepts to learn along with the seemingly random association to their symbols: C is the circumference; pi is a constant; r is the radius. Not to mention that all of these symbols need to be written with operators between them (there's also a spatial configuration). Depending on how you count, that could be 5 (or seven) items to hold in working memory. 

Once you learn the circumference equation, it becomes a single chunk of information, which makes it easier to remember. However, it's easy to forget what it was like not to know something. It's important to keep that in mind when teaching this or any concept. The first time we encounter new information, it is going to appear much more complex because it contains several small chunks of information. As you become proficient in the domain, it becomes easier to take on additional complexity due to the process of chunking.

Share and Enjoy!

Dr. Bob

For More Information

[1] Here is a link to George Miller's very famous paper: The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information

[2] Like the speed of light, the estimate of working-memory capacity is always changing. It depends on how you measure it! Some estimate that working memory is actually capped at a much lower rate -- around four separate items. One methodology for calculating working-memory capacity is the n-back task, which is notoriously difficult. Your job is to remember a digit (or whatever) that is n turns ago. For example, if n = 2, and I give you: 

3 5 7 8 x

When you see "x" you have to say "7" because it happened two turns ago. As if this isn't hard enough, there's also the murderously difficult dual n-back task. If you are feeling strong, try it yourself!

Thursday, December 4, 2014

Rework the Network: Semantic Networks

It's all about the Semantics

In a previous post, we talked about the power of the Associative Network. It explains several interesting cognitive phenomena, such as reminding, creative thinking, and priming. That's pretty powerful; however, there was a weakness in the Associative Network as a way of representing knowledge. It didn't quite capture the interesting differences between the links. In my previous example, I drew a link from whale to mammal and a second link from mammal to three inner-ear bones. Should we treat these links as the same? Maybe not. 

I was purposefully sloppy in the way I presented the whale/fish example. The nodes themselves came in two flavors. The first type of nodes were concepts (i.e., nouns). They included entities like whale, fish, dog, and cat. Then there was a different breed of nodes that described those concepts (e.g., adjectives). They included such modifiers as 3 inner-ear bones, fur, nurse young, and give birth to live young. The network would be so much more useful if the links between these two types nodes were labeled differently. Why is that the case? 

One reason why is that we can use the network to make some pretty interesting inferences. Going back to our whale example, if I know that a whale is-a mammal, and I also know that a mammal has three inner-ear bones, then I can infer the following fact: "A whale has three inner-ear bones." Nobody has to tell me that fact directly. Instead, I can use the labeled relationships in my network to derive or infer these facts. Thus, a Semantic Network is a node-link representation of knowledge where the links have meaningful labels.

A STEM Example

This is a pretty powerful idea for education because it means (at least) two things: 

Number 1: You don't have to tell your students every little fact. Instead, you can let them discover these facts for themselves. Not only is the process of discovery more enjoyable for the learner, it also leads to more robust learning (a topic for another time!). 

Number 2: Thinking in terms of a Semantic Network might also help structure the presentation of ideas in class. For example, it might help map out all of the relations between geometric objects: 

  • A square has four equal sides. 
  • A rhombus has four equal sides. 
  • Therefore, a square is-a rhombus. 

Mapping out these relationships explicitly can help students visualize and understand the distinguishing characteristics between different entities. It also (implicitly) teaches a meta-cognitive strategy of mapping out information in a hierarchical manner, which is easier to memorize. 

Share and Enjoy! 

Dr. Bob

For More Information

Setting up and maintaining a Semantic Network can be an actual career! I like to refer to this as "knowledge engineering." As a knowledge engineer, you get to think about and explore the various types of objects in the world (e.g., the nodes) and their properties (e.g. has and is-a relationships). The ultimate goal of doing this is to either create a system that can either teach existing knowledge or make new discoveries.

The hard part is figuring out a way to represent the nodes and (labeled) links in a way that a machine can read and understand. We call these "propositions," which can take pretty much any format. Here's an example of a format that I made up:

WHALE [ has("blowhole"), has("fins"), is-a("mammal") ]
MAMMAL [ has("3 inner-ear bones"), has("fur"), has("live birth"), is-a("animal") ]

Once you figure out the machine-readable representation, you can then develop a reasoning engine (also, not a trivial task) that you feed in the propositions. Viola! The reasoning engine can spit out conjectures that you never before considered because it uses chains of logic to derive new concepts and ideas. 

There are several projects that are attempting to make this happen. Check them out:

  1. Viv
  2. Cyc
  3. WordNet
  4. The Semantic Web