Working Memory in Air-Traffic Control

This is an unpublished note reviewing memory data which are not well known but are basic to my thinking.

The data suggest that results from experiments studying people trying to remember arbitrary material which they have little experience with may tell us what are the minimum limits to what people can remember, but do not give us much help with understanding how and what people remember in real tasks with meaningful patterns in and values/ costs on the items in the task.

Working Memory in Air-Traffic Control

Lisanne Bainbridge
Department of Psychology, University of Reading
July 1975

Trying to remember telephone numbers and shopping lists makes us realise that we are not very good at remembering long lists of arbitrary items. However we can do tasks like driving a car and holding a conversation, both of which involve keeping in mind a large amount of information about present circumstances and future intentions, without noticing memory limitations. How much can we remember when doing a complex task, what do we remember, and why is this memory capacity not so limited ?

Air-traffic control is a complex task in which memory has been studied extensively. Air-traffic controllers have to check that there is no risk of collision ('conflict') between aircraft flying through the area ('sector') they are responsible for. Information about each aircraft, its flight level (height), route, etc. is written on paper 'flight strips'. [note : These studies were done before the availability of labelled radar or computer supported radar displays.]

As the controllers have all the information displayed in front of them, memory might not appear to be necessary. However, if the controllers can remember an item then they do not need to search for it on the display, so the task can be done more quickly and with less work load. Controllers generally say that they have a 'picture' of what is happening.

It is difficult to test someone's memory without interrupting their work. Bisseret (1970) used laboratory methods to test how much an air-traffic controller could remember. Each controller was given a set of flight strips and asked to familiarise themself with the situation represented. A new strip was then given, the controller checked for collision risks with this new aircraft, and so on. At some time, not expected by the controller, the flight strips were covered. The controller was then asked to draw a map of the present situation, and to describe ('recall') everything they could remember about the aircraft. Finally the person was given a new flight strip and asked to say from memory whether the new aircraft caused a collision risk. The last method is the most natural way of testing the controller's stored information. The information is being used in the real task rather than reproduced in a separate test. But it is not physically possible to introduce new aircraft which require comparisons with all the possible stored information, so the main memory data come from the map-drawing and recall tests.

Different aircraft were remembered in the different types of test. Interestingly the controllers were not very good at drawing a map. They seemed to reconstruct the map rather than drawing it directly. Example comments were :
'it's flying from Dijon to Rolemport, estimated time of arrival Rolemport 05, so it must be about here',
'I've got one at level 150 which is about to pass beacon RLP and another at level 170 which is about 10 minutes behind so is about here'.

One would think that air-traffic control is a spatial task, but this evidence suggests that the aircraft were not remembered by their spatial positions but in relation to each other. This will be discussed further below.

Figure 1 - the number of items remembered by air-traffic controllers depends on their level of experience and the number of aircraft in the air (data - personal communication from Bisseret, based on Bisseret, 1970).

In finding out how much the controllers could remember, Bisseret tested three groups : highly experienced controllers, those who had just passed the lowest level qualification test, and trainees with 3 to 6 months less experience. These groups were tested at three levels of work load : 5 aircraft at a time, 8 or 11. Figures 1.a, b and c show that the amount remembered by a controller depended both on the work load and on the person's experience. Figure 1.a shows the average number of aircraft remembered. At the highest workload there was a 24% spread in memory performance by controllers with different amounts of experience. Figure 1.b shows the average number of items remembered about each aircraft remembered. Again there was a wide spread (36%) in performance at the highest workload. Figure 1.c shows total items remembered, as calculated from no. aircraft x no. items (the raw data collected in the study are no longer available, so it is not possible to find the total number of items remembered directly). At the highest workload the total number of items remembered by the highly experienced controllers was still increasing, the memory score of the newly qualified controllers levelled off, and the overall performance of the trainees deteriorated in an inverted-U curve. The highest average score achieved by the experienced controllers was 33 items remembered, by the qualified controllers 23 items, and by the trainees 20 items. People cannot remember telephone numbers this long, so it is important to ask why the air-traffic controllers can remember so many items, and why this ability increases with experience of the job.

Memory performance as tested in the classic types of laboratory experiment is rather poor. When repeating back items like telephone numbers, people remember 7 items on average. This is a test of memory for arbitrary and unchanging items. Yntema and his co-workers (1963) did several tests of 'running memory' in which the items to be remembered were changing during the test period. The items and the timing of the changes were arbitrary. They found that even 2 to 3 items are not remembered with complete accuracy in this type of task.

A classic laboratory experiment in which memory performance did improve is described by Miller (1956).
The people tested, when asked to repeat back a random series of 0s and 1s, could remember about 7 items.
They were then taught a number code which groups the 0s and 1s together : for example, with grouping into threes they would learn 010 = 2, 110 = 6, etc.
They were then retested at remembering sequences of 0s and 1s, and could remember 2 to 3 times as many items. Miller suggested that the people 'chunked' the items into groups of three, remembered the related numbers, and translated back again when tested. This experiment shows that memory capacity is improved considerably by learning patterns of organisation in the material.

However the air-traffic control task is not completely parallel to this. In the experiment described by Miller the person could learn alternative versions of the number sequence, both representing exactly the same material. In air-traffic control there are not higher level labels which inherently imply the details. For instance, remembering the height of an aircraft has no simple implications about its position. If patterns of organisation are important here they must be more complex.

de Groot (1965) demonstrated the importance of general patterns of organisation in memory for complex tasks, in a study of chess players. He showed a chess position to a player, then removed the pieces and asked the player to replace them. He found that ability to reproduce chess positions increased with the level of the player's chess mastership. A more recent experiment (Chase & Simon, 1973) repeated these tests with randomly placed chess pieces. In this case there was no difference in the ability of chess masters and non-chess players to replace the pieces. This result supports the suggestion that the chess masters remembered patterns in the real chess positions, rather than individual pieces, and that this patterning ability develops with experience. de Groot comments (p.329) 'The position is perceived in large complexes, each of which hangs together as a genetic, functional, and/or dynamic unit.... A separate process is sometimes needed to integrate them. ....the essential relations between the pieces, their mobility and capturing possibilities, their co-operation or opposition, are often perceived and remembered better than the positions of the pieces themselves'.

It seems that the ability to remember items can be increased by a knowledge of the types of pattern which can occur in a particular task. This is an aspect of memory which is not tapped by most laboratory tests of memory capacity, in which the items are random rather than patterned, and the people tested have no previous experience of the task. If de Groot's remarks on chess players also apply to memorising by air-traffic controllers, then the controllers should remember the items in patterns and according to their importance in the task. The comments made by the air-traffic controllers in the map-drawing task suggest that the aircraft were remembered relative to one another (see examples above). If a controller remembers one aircraft this could act as a cue to other aircraft related to it, so more items would be remembered. This suggests it is important to look in detail at which aircraft and items are remembered by the controllers, and how these items are remembered, rather than simply at the overall number of items remembered. The memory studies which will be described show that aircraft and items are remembered according to the recency of the items and the opportunities for repeating ('rehearsing') them. However, as the repetitions occur as part of doing the task, the items best remembered are those which are most important in the task. Also the way in which items are remembered shows that the controllers do not remember the raw information, but remember items in related groups. This is appropriate, as in checking for potential collisions the controller is not interested in isolated aircraft but in their relative positions and behaviour.

At Orly airport the controllers work for one hour periods. Some of the aircraft controlled are taken over from the previous controller in that sector, others are left for the next controller to take over. Sperandio (1969) did a study in which he asked the controllers, when they were relieved at the end of their control period, to recall all the aircraft they had controlled during the last hour. This was a test after the real task situation, and the number of aircraft during the hour varied from 1 to 24. On average the controllers remembered a maximum of 10 aircraft. (Bisseret's experienced controllers also remembered 10 aircraft.) At low workloads the controllers studied by Sperandio remembered all the aircraft, at higher workloads they remembered the most recent. 100% of the aircraft still under control at the end of the hour were remembered. This finding therefore shows the classic 'recency' effect found in memory studies, though only at high workloads. The aircraft which were in the sector at the beginning of the hour were remembered better than other aircraft which entered the control sector during the first quarter hour. Sperandio suggests that these aircraft are remembered in a different way, which can be retained more easily.

In a second study, at the French Northern Region control centre, Sperandio (1970) asked controllers to describe the current situation by filling in a set of blank flight strips at the end of their control session. Aircraft which were in radio contact with the controller were remembered more frequently than aircraft which had not yet established radio contact (Figure 2.a) This was especially true under high workload. All aircraft which were in conflict (risk of collision) were remembered, whether or not they were in radio contact. Both these findings illustrate the classic effect of 'rehearsal' on memory, as aircraft in radio contact or in conflict will be talked with or thought about more frequently than others. This rehearsal is not a random effect, these aircraft are most rehearsed because they are the most important in the task.

Figure 2 : recall of items about an aircraft depends on task category (Sperandio, 1970)

The number of items remembered about each aircraft was also related to the importance of the aircraft in the task. Bisseret op cit (Figure 1.b) found that an average of 2 to 3 items were remembered per aircraft. Sperandio found the same overall average in the real task, and also investigated the data in more detail. For aircraft which were not in conflict with others, more items were remembered about an aircraft if it was in radio contact than if it was not, see Figure 2.a. Radio contact did not affect memory for attributes of aircraft which were in conflict. At least three items were remembered about each aircraft in conflict. The limited evidence suggests that more items are remembered about aircraft on which action has been taken to remove the conflict, and even more items are remembered if the action has been chosen but not yet made, see Figure 2.b.

Considering which items are remembered about each aircraft, and how they are remembered, the relation between memory and the structure of the task becomes even more evident. When checking for a collision risk (Leplat & Bisseret, 1965) had found that, when the controllers check for a collision risk, they compared aspects of the aircraft in the following sequence : flight level, next navigational beacon (aircraft navigate by flying from one radio beacon to another), route towards beacon, time of arrival at beacon, speed, route after next beacon, time of arrival at second next beacon. This sequence of variables reflects the natural hierarchy of their importance in the task, as if a pair of aircraft do not conflict on one of these variables then there is no need to compare the aircraft on further variables in this sequence. Both Bisseret and Sperandio counted the frequency with which each of these variables was recalled. They found that the first variables in the sequence were remembered most frequently, and so on. Again the frequency of recall reflects the frequency with which a variable is used in the task, which parallels its importance. For each aircraft, either its flight level or its position or both were remembered, plus one or more of the other items. In Bisseret's test, in which controllers had to resolve conflicts from memory, the controllers were always right if the problem involved either level or position, and usually right if it involved both. Thus, remembering these items allows the controller to resolve many conflict situations in their head without reference to the display, which makes the task much easier.

The items remembered are therefore not a random selection but are those items which most useful in doing the task. However they are not just remembered as a result of frequency of use. de Groot suggested that chess players remember patterns of chess pieces. The details of how an air-traffic controller remembers items, and the memory errors they made, show that the controllers did not remember raw data about the aircraft, but remembered the information in a form which was the result of thinking out the relation of an aircraft to the airspace and to other aircraft in it. For example, flight levels were remembered in aircraft pairs or threes :
'there are two flying towards beacon DIJ, one at level 180, the other below at 160',
'there are two at level 150, one passed DIJ towards BRY several minutes ago, the other should arrive at X at 22'.

In the map-drawing test, aircraft positions were sometimes reversed, or drawn in correct relative rather than absolute positions, suggesting they were remembered in relative pairs. Half the route errors were minor confusions between routes converging on the same point, suggesting that the controllers' representation contained simplifications when distinctions were not particularly useful in the task. Most of the information was remembered in relation to the future. In the map-drawing study, most of the errors put the aircraft too far ahead. In the verbal reports, half the correct positions were described by both last and next navigational beacons, three quarters of the others quoted the next beacon. 65% of aircraft timings were quoted in terms of the next beacon.

These studies show that the controllers remembered items which were most relevant in doing the task. They remembered these items partly because they are used most frequently in doing the job, but they also remembered them in a way which was organised by and in relation to the task thinking and decisions. The controllers' knowledge of the present situation would best be structured in a way which indicates the most relevant comparisons between aircraft, so reducing the amount of mental work to be repeated later. It is possible to suggest that a highly experienced controller can remember more items because such a person knows the task operations sufficiently well to recognise patterns between or relations between the aircraft rather than remembering each aircraft individually. An experienced controller may also be less susceptible to the effects of increased work load, as if they can remember more items in a more interrelated way they will need to check their display less frequently, and any new information will fit more easily into their existing knowledge of the situation. This increased patterning ability is an important aspect of advanced skill, and can continue to develop over many years of experience.

Studies of air-traffic controllers therefore show that, in complex real tasks, memory factors are important which are not investigated in experiments studying people doing arbitrary tasks for the first time. As yet very little is known about these important patterning processes by which remembered material is organised.


Access to other papers via Home page

©2021 Lisanne Bainbridge


Popular posts from this blog

Ironies of Automation

Types of skill, and Rasmussen's SRK schema

Complex Processes Review : References