The Mind Boggling Memory Experiment: HCI

Human-computer interaction

Magicians do more than manipulate where the audience's attention is focused to ensure we miss the things that give a trick away. They also give us other things to think about. In the joker in the pack trick we saw that people are more likely to make some kinds of mistake if their working memory load is filled. Experiments have shown that the story is more subtle than that though and it can make a trick go wrong too.

We fill a person's working memory by imposing a cognitive loadon them. One way this can happen is that the task they are trying to do could be difficult in itself. The load is a central part of what they are trying to do. This kind of load is called an intrinsic load.

The magician has that problem with this 'memory experiment' trick. To pull out the right cards you have to do some arithmetic in your head. The intrinsic load is high so you are more likely to forget to do steps in the trick. That's what makes it hard, not that you have lots to memorise!

Cognitive load could instead be just due to the way information is presented. This is called an extraneous load. If the information you need to do the task is hidden amongst lots of other information then there is an extraneous load. For example, suppose in another trick you have to watch as cards were turned from the top of the pack to see when a card appears. (The You said it! trick is like this.) That monitoring is adding an extraneous load.

The extra patter that a magician gives around a trick is partly there to add extraneous load. The information the audience need to work out what happened is there. The trouble is they have to filter it from all the other things the magician is saying and doing. If everything is stripped down to the core steps for the mechanics to work, then tricks are much more obvious.

Why do we care that there are different kinds of loads? Well it turns out that they affect different cues in different ways. We have discussed visual cues: cues that you see, like a reminder message. There are also internal cues. Whereas visual cues are out in the world, internal cuesare in your head. As you practice a series of actions over and over again each starts to cue the next one. You apparently don't need visual cues of what comes next any more.

Jonathan Back of University College London has explored how cues are affected by different kinds of loads. His experimental participants role-played doing fire-engine dispatch. They had to respond to incidents, plot routes on a map for fire engines to take and allocate back- up fire engines, all using a computer interface.

For some participants the possible routes were plotted automatically (giving a low intrinsic load as the task itself was then quite easy) whereas others had to work out the best route themselves (a high intrinsic load: a hard task).

All had to extract information that affected the possible routes (like traffic jams) from a streaming ticker tape of information. For some, much of this was irrelevant to the current route (high extraneous load). Others were just given the information they needed in the ticker tape (low extraneous load).

There were lots of opportunities for people to make errors whilst dispatching fire engines. Only three were common. The first was an initialisation error where a person forgot to do a setup step before doing the main parts of the task. Forgetting to click in a password box before you type is an example of this kind of error. A second was a post-completion error where people missed a final tidying up task (remember the mistake of forgetting to switch off your light on getting something from your room).

Finally a mode error was also common. The error happened when people failed to check a mode indicator that told them what they needed to do next. Was the machine in a state where they would be given route possibilities or was it in a state where they had to work out the route themselves? They assumed the wrong mode so did the wrong thing.

The experiments showed that when the task was hard and there was lots of information to sift through (high intrinsic and extraneous load), visual cues were far less effective and people were much more likely to miss them. They were more likely to make mistakes.

What does that mean to you in a performance? If a trick needs you both to do mental calculations and to watch cards being dealt for the values needed, don't rely on noticing subtle reminders from your assistant to stop you forgetting a step! You might expect that internal cues would be affected in the same way. That isn't true. It turns out that all you need is a high intrinsic load for internal cues to be disrupted, and so make it more likely that people will make errors. In the memorising trick you do have a high intrinsic load, so beware. In pressure situations like a performance you will be prone to forgetting things however much you practice. If you have to do tricky stuff in a trick, keep the rest simple!

Design principles

What does this mean to a computer scientist? Well, if you are designing an interface for a task that is hard, sensory cues need to be very, very obvious. Think of an airline cockpit. When the pilot is coming in to land their task is difficult. The procedures they follow (like changing angle of descent, putting the landing gear down, and so on) should not just rely on the pilot's training. Without strong external cues to help in particularly tricky situations, some things may be forgotten (see the Strasbourg crash). Special care should be taken over steps like remembering the autopilot mode, post- completion steps and initialization steps. Better still, design the difficult steps away if you can.