These lecture notes were not written as a course handout, but as a resource for lectures. Therefore, references and comments will not always be complete.

Lecture 5: Memory, Attention & Learning

Memory

Very controversial stuff.

But very important to human factors work.

Key notions:

Memory is limited in the short-term (around 7 meaningful items). This is taken from Miller's famous paper. But expert's meaningful items are larger.
Memory for future activities (both in the short- and long-term) is prone to lapses and is probably limited
Ability to recall is affected by encoding, retention and retrieval processes
Encoding is the 'laying down' of a memory trace. More meaningful processing at encoding seems to lead to longer, more reliable memories.
Retention is the interval between encoding and retrieval. Activities during retention can cause forgetting -- processing similar items; interruptions; etc. Length of retention interval is important too
Retrieval depends upon the information available to cue recall. Retieval is usually better the more similar recall circumstances are to encoding (even down to temperature, lighting, etc.).
We can remember much more information when it is meaningful and when its meaning is processed at encoding time. Later discovery of its meaning does not help especially. Failure to process meaningful information also leads to poor memory.

Poorly understood, but important questions

Impact of interruptions. Many work environments include numerous interruptions. The effect of these on memory has rarely been studied. What work there is suggests that length of interruption is not very important, but similarity of material processed is.
Other memorial tasks. Most of the literature concerns simply learning a list and then recalling it sometime later. Many work related memorial tasks invole remembering some piece of information for a short time, then replacing it with some similar piece of information ('keeping track' tasks). These have not been studied particularly.
Support for memory. Most of the literature is about processes and architectures of memory -- but our human factors interest is in preventing the need for memory and providing support for memory tasks. Little is known about how to do this -- except from common-sense.

Attention

'Paying attention' is clearly a desirable property of a human operator, and designing systems to attract and maintain attention is important. There are some things we know, but there is plenty of mystery still.

Key principles

We can function without attending to information -- how else would we know to shift our attention to some new information. 'Cocktail-party effect'.

Many people believe that our ability to process information without attention is limited to surface features / syntactic properties / etc. And that we cannot process the meaning of something without attention.

Some believe that we cannot remember something in the long-term without paying attention to it.
Evidence suggests that some meaning can be processed inattentively, but not very much, and that any long-term memory will be of surface features almost exclusively.
However, the study of skilled performance reveals that we can perform many things without paying attention to them. Some of these skills (e.g. driving) are very sophisticated -- some meanings must be being processed, but we do not remember any of it.

Slips of action

Attention and skilled behaviour have a critical role in the occurence of 'slips' of action. There seem to be two modes of operating :

Open loop: behaviour can continue without monitoring: Other activities can be performed at the same time.
Closed Loop: Conscious montoring is required of behaviour: Only activity can be done at once.

Open loop skills have moments where they must become closed loop -- for those parts of the activity that cannot be done unconsciously. It is at these transition points that everyday slips of action can occur.

A variety of things can go wrong when we fail to switch into closed loop mode:-

Information that ought to have been processed or influenced behaviour is overlooked. E.g. Mode error on a word-processor.
Continue performing familiar activity even though intended to do something different. E.g. saying 'Yes' to "Do you really want to delete this file?"
Failure to discriminate relevant objects in world -- perform action on unintended objects. E.g. putting away dirty mug instead of,, or as well as, clean ones.

Although not clear that we can design to pre-empt these slips we can predict the kinds of circumstances when they occur.

We can also recognise the power of open loop behaviours yhat can foil "Do you really want to...?" dialogues.

Rasmussen's model of Skill, Rules and Knowledge driven behaviour captures most important features of human skilled behaviour.

Skilled behaviour

Real human skills are a complex mixture of open- and closed-loop behaviours.

Rasmussen has distinguished between Skill-, Rule- and knowledge-based level of operator behaviour.

Skills level: is automatic, functions in normal operating environment. Much of this is not available to conscious thought, verbalisation etc. (This is an example of where Norman gets it wrong. Such behavior is tied to the environment, but it is also not inspectable in its own right.)
Rule-level: Behaviour becomes a conscious activity and is based on familiar rules -- either dictated or acquired -- this is for situations that are rarer but are known to occur and a rule for what to do can be provided.
Knowledge-level: Behaviour occurs at the knowledge level when there are no rules to inform the operator what to do. These conditions (usually emergencies) require some thought and reasoning about the state of the plant, based on the operator's knowledge (i.e. mental model) of the plant.

For example, the Kegworth pilots were operating at all 3 levels. Some aspects of their behaviour were automatic, for others they refer to rules and procedures, whilst for others they reason on the basis of their knowledge about the plane.

Rasmussen's argument is that good design needs to support all three levels of operation, not just one.

[ One can note a social human factor here -- if there are multiple operators, then at the knowledge level they may have different knowledge, which may give rise to conflict or to strength.]

The knowledge level implies a certain amount of planning activity, but you will find there are those who believe that people do not engage in planning -- arguing that behaviour is situated (e.g. Suchman) in a context. In other words, they argue that we perceive the situation and decide what to do then, not on the basis of some pre-formed plan.

Two responses can be made to this:

one can do both - have a plan and allow the situation to change it (e.g. performing actions in a different order from that planned);
planning seems best related to Rasmussen's knowledge-level, whereas situated action is 'rule-level' behaviour. Again, these are not exclusive options -- some people in some skills may function exclusively at one level, whilst others may switch between levels at high speed.

Learning

Human learning is a poorly understood area, with fair controversy.

Two key ideas:

At least two sorts of learning (implicit and explicit) and (declaraive and procedural).
Learning does not need to be complete.
Reinforcement plays a key role in understanding much learning (even when worrying about higher cognitive processes too). For example, unsafe behaviours are rarely punished, whilst fiddly, but safe procedures are consistently negatively reinforced -- awareness of the danger does not change the behaviour; the reinforcement is too strong.

Implicit learning seems to be automatic, based on practice, is not improved by reflection and produces knowledge that cannot be verbalised.

Explicit learning proceeds with full consciousness in a hypothesis testing way. produces knowledge that can be verbalised.

It is easy to think that if you have learnt to operate a system your learning about that system is complete. Needn't be so, can sometimes perform very competently with very little knowledge.

Process of learning

The process of explicit learning can be seen in many ways as the reverse of Rasmussen's model. The novice begins with knowledge and reasoning, proceeds through rule-based, and ends with skills.

Implicit learning goes from the skill level to the knowledge level -- the knowledge eventually deriving from the observation of one's own behaviour.

Whichever way one goes, some key phenomena survive:

the ability to recognise correct / incorrect items comes before the ability to generate correct items.
knowledge is not acqured in an all-or-nothing way. Novices go through a stage of fragile knowledge, where sometimes knowledge is used, sometimes not.
experts acquire a rich repertoire of representations of their knowledge. It is not that experts know more, what they know is much better organised and more readily available.

References

Here are two references, in addition to Preece, that may provide some interesting reading in this area. Ritter's web area also contains some papers in the area with further references.

Anderson, J. R. (1996). Cognitive psychology and its implications (3rd ed.). New York, NY: W. H. Freeman.

Lindsay, P. H., & Norman, D. A. (1977). Human information processing. San Diego, CA: Harcourt Brace Jovanovich.