Design of Everyday Things: To Err is Human

Humans are prone to error, which can be broken into two categories. Slips are errors resulting from automatic behavior and mistakes result from conscious decisions. Slips are typically minor, easily identified errors, whereas mistakes are much more difficult to detect and can be major events.

Slips can be further broken down into capture errors, description errors, data-driven errors, associative activation errors, loss-of-activation errors, and mode errors. Capture errors occur when habitual actions override intended actions, such as finding yourself driving to work on Sunday instead of church. Description errors occur when an intended action is similar to other available actions and a lack of specification results in a slip. This type of error is most frequent when right and wrong choices are in close proximity. Data-driven errors occur when extraneous data influences our actions, for example, typing a phone number instead of a credit card number shortly after calling someone. Associative activation errors when external data triggers an incorrect action - saying "come in" instead of "hello" when answering a phone. Loss-of-activation errors occur when we forget what we are doing. Lastly, mode errors occur when devices have contextual operations and we operate one mode thinking we are operating another. Design can be improved by taking slips into account - minimizing them and providing adequate feedback and correction when they inevitably occur. Minimizing slips can be done by differentiating choices and requiring confirmation. Another good design consideration would be the elimination of irreversible actions.

Mistakes occur from choosing inappropriate goals - poor decision-making, misclassification, or lack of information. Humans make decisions based on expectations and prior experience rather than logical deduction. Previous chapters have already explored the problems associated with memory, so it is no big surprise that people make so many mistakes. Another theory of cognition is the neural net approach. This theory uses the structure of the brain to conclude that human cognition is based on activation and inhibitory signals that travel along neurons through the brain. Thoughts are represented by stable patterns of signal activity.

Tasks can be structured into models for analysis (data structures). Turn-based decision-making games such as tic-tac-toe can be modeld using a decision tree - the author refers to this type of structure as wide and deep. Simpler sets of data, such as a menu, can be represented using a list - this is a shallow structure. An example of a narrow structure is a cookbook recipe, where there are few alternatives, resulting in a decision tree that is narrow and deep.

Behavior is often thought of as conscious, but much of it is subconscious. Subconscious thought is based on pattern recognition and generalization. By analyzing trends the subconscious is able to guide behavior quickly and efficiently, but perhaps not as accurately as we might like. Conscious thought is slow, laborious, and relies on STM, which we know is very limited and subject to flaws.

Mistakes are very hard to identify, especially when they are a result of misinterpretation. Furthermore, understanding of an event before and after is occurs can be drastically different. Another factor to consider is the role of social pressure. The perception of pressure leads to misunderstanding, mistakes, and accidents.

The author discusses 4 things that designers can do to design for errors:

1. Understand the causes of error and design to minimize those causes
2. Make actions reversible whenever possible and make irreversible actions difficult to carry out
3. Make erros easily discoverable and correctable
4. Change the attitude toward errors. Think of actions as approximations of what is desired.

Forcing functions address errors by forcing users' actions. An interlock forces operations to take place in proper sequence (a microwave cannot function with the door open). A lockin keeps an operation active when appropriate (soft functions as opposed to hard functions). Lockout devices prevent actions, such as safety rails that prevent accidental death. Forcing functions are very effective but almost universally hated by users. People don't like constraints, even if they are in the user's best interests.