Chapter 3 Bottlenecks
In conventional computers, such as a laptop or desktop, most of the calculations are done in the CPU - the central processing unit. Computers also have memory, called RAM. For example, the laptop that I am writing this text on had 16 GB of RAM. That is enough to store a lot of informationm- 2857 copies of the complete works of William Shakepeare.
Unfortunately, the CPU of a conventional computer, such as my laptop, can only operate on a tiny amount of memory at any one time. Thus, there is a bottleneck between the memory and the CPU. Instead of the CPU doing calculations and transformations of all the memory simultaneously, it only does so bit-by-bit.
Imagine you had the complete works of William Shakespeare in your computer’s RAM (just one copy), and wanted to change each lower-case letter in it to upper-case, and each upper-case letter to lower-case. It’s a pretty simple task, but the CPU can only do it for one letter at a time. We want to get all of Shakespeare’s works processed by the CPU, but the CPU is a bottleneck - the data has to sit in memory waiting for the CPU to get to it.
After the image of an object forms on your retinas, or sounds hit your eardrum, neurons carry information about these signals to your brain.
Some of that information you will then perceive. For example, you might perceive that there is a salt shaker in front of you. Some of the information you will also remember and be able to recall a second or so later. But you will not consciously perceive all of the information that reaches your brain, and you will remember even less. In other words, we have bottlenecks in perception and memory. Only a limited amount of the sensory signals coming in from the eye are fully processed for memory or even for perception. Attention refers to the control of which signals are processed the most. That is, if you attend to an object, your brain is trying to ensure that it is fully processed.
For perception, some signals from the senses won’t be perceived unless the associated sensory signals are routed to processes in the brain that can only process a few things at a time. Memory also has a limitation, and not just in how much storage is available. There is also a bottleneck for getting things encoded into memory. That is, many things won’t be remembered unless you attend to them so that the associated perceptual representations are routed into the memory processes that, sadly, can only process a few things at a time. We don’t think the situation in the brain is as simple as a single bottleneck like a CPU. Instead, the different things the brain does are subject to different kinds of bottlenecks.
3.3 Feeling a bottleneck
How many math problems can you do at one time? Although your brain contains more than 80 billion neurons, you can probably only do math problems one at a time. How about other tasks - if I showed you twenty words on a page, how many could you read simultaneously? A large body of psychological research suggests that humans can read only a few words at a time, or possibly only one at a time (White, Palmer, and Boynton 2018). Yet when light hits the back of our eyes, it is greeted by six million cones arrayed across each retina. Each bit of the image is simultaneously processed, as each has its own photoreceptors devoted just to it. Retinal stages, then, process millions of small regions of the retina at the same time.
Explicit thinking like math problems, however, involve a series of steps, and our brains appear to be particularly capacity-limited at doing this.
But given that explicit thought is so limited, seemingly to only one thing at a time, something has to control what it thinks about - attentional selection. The earlier lectures in this unit were about memory, decision making, and problem solving. Attention is what routes information to the memory, decision-making, and problem-solving processes.
That memory encoding and problem solving are so limited is quite intuitive - we’ve all had the experience to know that it’s hard to think about solving more than one problem at a time. A bit more surprising, perhaps, are the limitations on perceptual tasks. But it’s important to realise how pervasive bottlenecks are in the mind - they afflict even perception.
Imagine you were to travel back in time several decades ago as a fighter pilot, patrolling the skies of Cuba to deter the expected American invasion back in the 1960s (or the 70s or the 80s, the Americans were always a big threat!). You’d be flying a Soviet-made MIG fighter. When I visited Havana in 2003, I got to see one of these planes up close.
From the top of the stairs they’ve constructed next to the plane, I was able to take a photo of the inside of the cockpit:
As you can see, the pilot has a lot of dials to monitor. Information overload? Yes, although the large number of dials and swiches don’t overload the initial stages of our brain’s visual processing.. at some stage after that, there is a bottleneck for more extensive processing. We will describe that stage in Chapters 7 and 9.
Already in World War II it became obvious that fighter pilots couldn’t fly and monitor effectively all the displays at the same time. This inspired psychologists to begin studying capacity limits .
The challenge was that there was a wide array of signals a pilot might potentially need to be aware of, more than the pilot could really process simultaneously. Because unfortunately a pilot’s cognitive system can only process a few things at a time, only some signals can be processed at any one time, and this comes at the expense of others.
The naive view of many people is that if our eyes are open, we are aware of everything that hits our retinas. But as will be illustrated in later chapters, this is not the case.
Your eye itself is a lot like an old analog camera. It doesn’t know what it is looking at. You need millions of neurons to work together to process the eye’s signals to identify an object. And our brain isn’t big enough to have enough neurons to identify all the objects currently in your field of view. If you had enough neurons to do all that, your head would have to be much bigger, like that of this cartoon man’s:
To save head space, and energy, the brain has only a small number of neural circuits capable of doing math problems, of identifying faces, of tracing shapes, of reading, and of doing other tasks. This raises the problem of attentional selection.
The problem of attentional selection is that of getting the appropriate signals from the eye to the limited number of neurons available for each task. If you didn’t have attention, there would be no control of what gets fully processed.
Regardless of whether you’re paying attention or what you’re thinking about, the photoreceptors in your retina are transducing the light emanating from me, the ganglion cells are sending spikes representing my image to the lateral geniculate nucleus, and the LGN is passing that information into the cortex. But, unlike these processes, not all visual processing is mandatory- some is optional. For example, your brain probably doesn’t start computing the arm trajectory to grab this pen until you willfully think about grabbing it. When you attend to the pen to think about grabbing it, we say that your attention selects the pen for further processing by the motor system.
Put in computer information terms, Anderson, Van Essen, and Olshausen (2005) depicted the situation with the following diagram:
Their tentative, rough estimate was that the optic nerve can process 100 times as much information as can get through the attentional bottleneck. In other words, only 1% of visual signals get past the bottleneck.
The same problem occurs for other senses such as hearing. As you’ve probably experienced at parties, you can’t comprehend what everyone around you is saying at the same time. You need to listen to one particular conversation.
A major question of attention research is what kind of processing goes on before you attend to something. Caleb told you about this in PSYC1 when he discussed the debate about the early, late or flexible locus of selection. In this class, we will only discuss this with respect to visual search. For what we will focus on, remind yourself by looking back at the learning outcomes ( 2 ) and the chapter titles.
3.4 A bottleneck for object judgments
3.4.1 Simultaneous objects
Duncan (1984) was interested in how many objects we can process at a time.
He asked people to make two judgments about a display that was flashed briefly. His participants could be asked to judge
- whether a box was small or large
- whether a gap in the box was on the left or the right side of the box
- which way a line was tilted
- whether the line was dotted or dashed
Here are two of his displays:
An individual trial in the experiment consisted of a brief presentation of a box with a line passing through it, like so:
This was considered to be two objects, a box (with a gap) and a line. The participant knew they would have to make two of the judgments listed above, such as #1 and #2 or #1 and #3. Those two judgments could be either about one object (the box, or the line) or two objects (the box and the line).
- Judgments about one object: #1 & #2 or #3 & #4
- Judgments about two objects: #1 & #3, #1 & #4, #2 & #3, or #2 & #4
Duncan knew that if he presented the objects for long, everybody would get the judgments right regardless. But he found that if he flashed the objects for a tenth of a second or less, the task became difficult and participants made frequent errors.
The results of the study were that when the two judgments the participant had to make were about a single object (the box or the line), they made an error about 17% of the time. But when the two judgments were about different objects (one judgment about the box and another judgment about the line), participants made an error about 24% of the time. That’s about 50% more errors for two judgments about different objects than two judgments about one object.
This pattern of results suggests there is a bottleneck for processing objects. The area the participants had to attend to was approximately the same in the two-object condition and the one-object condition. So the greater difficulty associated with the two-object condition was not down to having to split or spread attention over a greater area. In the conclusion of his paper, Duncan wrote:
Findings support a view in which parallel, preattentive processes serve to segment the field into separate objects, followed by a process of focal attention that deals with only one object at a time.
In other words, Duncan’s proposal was that the early visual system processes the multiple objects of a scene simultaneously, but making certain judgments requires additional processing that proceeds serially.
3.4.2 Successive objects
Subsequent work has yielded evidence that processing of even a lone object is subject to a bottleneck that prevents perception of a second object when a first object is being encoded into memory. The phenomenon that demonstrates it is called the attentional blink. Here is one demonstration and here is another. People do fine identifying a single item from the stream, but not two, if the second one occurs very soon after the first one. In some cases, the performance decrement is quite a bit bigger than that found in Duncan’s experiment for two objects.
The relationship of the memory encoding bottleneck evident in the attentional blink task to that of the difficulty judging two objects presented simultaneously is still being investigated. We don’t really know whether we should think of these as two distinct bottlenecks or whether the same underlying limitation in the brain is causing both.
Answer these questions and relate them to the learning outcomes ( 2 ):
- What did the discussion of the CPU of a computer illustrate about the brain?
- What problem does the cockpit of a fighter plane present for a pilot?
- Why do we need attentional selection?
- Write out brief answers to the first two learning outcomes ( 2 ).
Anderson, C H, D C Van Essen, and B A Olshausen. 2005. “Directed Visual Attention and the Dynamic Control of Information Flow.” In, edited by L Itti, G Rees, and J Tsotsos. San Diego, California: Elsevier.
Duncan, John. 1984. “Selective Attention and the Organization of Visual Information.” Journal of Experimental Psychology: General 113 (4): 501–17.
White, Alex L., John Palmer, and Geoffrey M. Boynton. 2018. “Evidence of Serial Processing in Visual Word Recognition.” Psychological Science 29 (7): 1062–71.