My notes and other stuff


Paper: How a Cockpit Remembers Its Speed

So it's been a few weeks, and I decided to cover a cool classic of cognitive science from 1995, How a Cockpit Remembers Its Speed by Edwin Hutchins. In this one, Hutchins tries to make the point that the unit of analysis for cognition can be moved to systems larger than individuals, and highlight properties that are distinct and can't be reduced to cognitive properties of people within that unit. Or to put it another way, you can think of cognition as it arises in the whole cockpit, and not just by putting together the cognition of the pilots in it. He shows this by focusing on how a major airliners use a small set of tools to set their speed when approaching a landing.

This is an interesting approach because if you want to analyze the cognition of individuals, you have to try to guess what's going on in their minds. If you analyze the cockpit, you can be inside and look at the many representations used and interactions, without having to guess about anyone's thought process. So even though this is a 1995 paper and the calculations and tools for aircraft speed at landing have undoubtedly changed, the contrast between the approaches is what will be of interest here.

In terms of landing a plane (an MD-80 in this case), a few steps are mandated and used in nearly all non-emergency circumstances. Some of the key ones have to do with extending the flaps and slats while slowing down (enlarge the wing to increase lift at slower speeds) while slowing down. Based on the runway length, aircraft weight, braking systems, margin for incidents, weather, company preferences, and crew preferences, a reference speed is chosen as a trade-off between going as slow as possible so the landing is safe, but not so much that you can't do a go-around if something goes wrong.

The possible configurations and guidelines are in the aircraft's manual and look like this: The Flap/Slat configuration and reference speed table as it appears in the MD-80 operating manual

Within the cockpit, there are two pilot stations, where one is considered "flying" (PF) and one is considered not-flying (PNF). The PF deals with piloting the aircraft. The PNF deals with air-traffic control (ATC), deals with checklists, and other systems and duties.

With this background set, Hutchins defines three ways of describe the memory for speeds: a procedural one, a cognitive one outside of the pilots, and a cognitive one inside the pilots.

A procedural description of memory for speeds

The flight is usually at roughly 30,000ft, and has to go down to 18,000ft. To pick the proper wing configuration based on all the factors mentioned earlier, they use a booklet of speed cards, each of which looks like this:

A speed card from an MD-80 speed card booklet, showing various reference speeds and slat/flap configurations for a given weight

The steps are usually, done 25 to 30 minutes before landing:

  1. Determine the aircraft weight (continually computed and displayed near the fuel gauge)
  2. Find the speed card and put it in a visible place in the cockpit
  3. Set up speed bugs on both airspeed indicators

So here is what the airspeed indicators look like: airspeed indicator instrument in the McDonnell Douglas MD-80

and here's an annotated version explaining the speed bugs:

airspeed indicator instrument in the McDonnell Douglas MD-80 with annotations: Starting with the bug at 227 knots and moving counterclockwise, the
bugs indicate: 227-the minimum maneuvering speed with no flaps or
slats extended; 177-minimum maneuvering speed with slats, but no
flaps, extended; 152-minimum maneuvering speed with flaps at 15”
and slats extended; 128-landing speed with flaps at 40” and slats ex-
tended (also called Vref)

The descent is then done in steps. Get to 250 knots or below before going down to 10,000ft (safer for traffic volume), then going down to 7,000 with speeds matching slat and flaps extension. They start using speed bugs and checklists are usually involved. The author points out:

Because it is dangerous to fly below the minimum maneuvering speed for any configuration, extending the flaps and slats well before slowing to the minimum maneuvering speed might seem to be a good idea. Doing so both would increase the safety margin on the speeds and would give the pilots a wider window of speed (and therefore, of time) for selecting the next flap/slat configuration. Unfortunately, other operational considerations rule this out. As one operations manual puts it, “To minimize the air loads on the flaps/salts, avoid extension and operation near the maximum airspeeds. Extend flaps/slats near the Min Maneuver Speed for the flap/slat configuration.” The extension of the flaps and slats must be coordinated precisely with the changes in airspeed. This makes the accurate memory of the speeds even more important than it would be otherwise.

At 1000ft, they pick the final flap setting, at 500ft the final approach speed and altitude changes are called out, and keep calling it out if it varies more than 5 knots from the approach speed.

A cognitive description of memory for speeds outside the pilots

The observable representations in the cockpit processes related to the speed are: the weight display (by the fuel gauge), the speed card booklet, the airspeed indicators with the bugs, the speed selection window in the flight guidance control panel, and the verbal communications between crew members. Hutchins adds two speed representations are not accessible: those in the memories of the pilots.

The speed card booklet is a long-term memory item, and can survive the entire life of the aircraft. The weight/speed correspondences too, and they can't be modified by the crew (and there are backups). To filter the speeds in the booklet, they find the current weight and position the booklet in a manner that hides invalid configurations, and then makes this more permanent by picking the appropriate speed card. But not only that, it makes it an artifact visible to both pilots over time, and plays a social role as well as an aide to repeatedly cross-reference speeds and expected weights to re-evaluate choices made.

Spoken words and speed bugs are extra representations. Spoken words are interesting because while they're ephemeral, they use sound instead of visual cues, and the PF is already very busy looking at lots of things.

As for the speed bugs, they are useful when negotiating the gradual transition between lower speeds and various configurations while also just dealing with the overall approach to landing. As the speed needle approaches the speed bug for the next configuration, the pilot can call for changes in flap configurations. The bugs can provide a bridge between speed and configuration at the current weight. In some airlines, both pilots need to acknowledge the change (both visually and spoken), which gives room to find inconsistencies or mistakes, and to synchronize with their various tasks.

Setting the speed bugs is a matter of producing a representation in the cockpit environment that will serve as a resource that organizes performances that are to come later. [...] I call this entire process a cockpit system’s “memory” because it consists of the creation, inside the system, of a representational state that is then saved and used to organize subsequent activities.

A cognitive description of memory for speeds inside the pilots

Most of the cockpit's memory is is done through people interpreting various symbols in their environment rather than purely based on memory: speed bugs let pilots pattern match the weights and cards and speeds, without having to actually recall what they are (just where they are). Over time, standard or frequent patterns may become learned and more automatic. Still, reading the cards has been optimized: frequently used speeds are in a bigger typeface than rare ones, and key values (like the Vref) are put in a more visible box.

Setting the bugs themselves can be challenging because it is done while piloting the aircraft, changing speeds, reading the values, and so on, and other tasks can mess with the working memory of pilots there. The author suspects that the many redundancies in the process around setting the bugs may help make this process a bit more robust.

The interesting visual pattern of bugs is that they end up being markers around speed indicators that create "a space of speeds", meaning that rather than reading specifics speed and flap/slat settings, you can see the needle move, and just deal with things in higher level sequences without requiring as much focused attention on decoding the data:

The same speed dial as before, but the space between each bug is annotated, counterclockwise: 0/EXT, 11-15, 28-40, and Danger past the last interval

The key insight as I see it is:

In the functional system with speed bugs, some of the memory requirements for the pilot are reduced. What was accomplished without speed bugs by remembering speed values, reading the AS1 needle values, and comparing the two values is accomplished with the use of speed bugs by judgments of spatial proximity. Individual pilot memory has not been enhanced; rather, the memory function has now become a property of a larger system in which the individual engages in a different sort of cognitive behavior. The beauty of devices like speed bugs is that they permit these reconfigurations of functional systems in ways that reduce the requirements for scarce cognitive resources.

Similarly for the salmon bug, you don't have to take the time to acknowledge which speed you're going at as much as jut matching it with wherever the bug is. Small details are relevant here. Since you have to be within 5 knots of a target speed, the salmon bug covers 10 knots. Any time your speed indicator lines up with it, you're in a proper range. A conceptual task becomes implemented by a much simpler perceptual process.

Interestingly however, this was not planned as such:

The engineer who wrote the specifications for the airspeed indicator in the Boeing 757/767 reported to me that the width of the base of the command airspeed pointer (salmon bug) is not actually spelled out in the specifications. The width of the tip of the pointer is explicitly specified, but the width of the base is not. [...] the engineers say it has this width so that it will be easy to find, but will never obscure more than one large tick mark at a time. If it covered more than one large tick mark, it might make it difficult to interpolate and read speeds. That constraint solves a design problem for the engineers that the pilots never notice (because the difficulty in reading the scale that would be caused by a wider bug never arises), and provides a bit of structure in the world for the pilots that can be opportunistically exploited to solve an operational problem that the designers never anticipated.

The speed bugs are interesting because by moving the memory requirements ("remembering the speeds and configurations") into artifacts in the cockpit, they can take computational tasks done at a calmer time and use them to reduce the cognitive work in more demanding periods such as the landing approach.

Hutchins comes back to the comparison between the approaches and states that even having a full understanding of how a person's memory works would be insufficient because so much of it lives in their direct environment. The interactions between actors can be as important as how they work internally. Since a theory's job is to help people look at the right places for answers to questions, he concludes that this socio-technical framework of interactions can help bridge the gap between standard analysis targets—direct cognitive views and larger cockpit views. As any new piece of technology disrupts and influences the flow of information, he supposes this would be useful to track better.