The intent of this project is to train computers to empathize with human’s limited capacities to multi task so that computers can instruct humans in a way that maintains a person’s focus.
Smart phones, em ail automation, Google Calendar ping individuals thoughts, interrupting their daily procedures unaware of the humans tasks. These actions command attention like an impatient toddler unaware of the context of the individual’s current focus. Today knowledge workers have little control over their focus. As the commands come increasingly from our devices rather than contextually-aware humans, the commands arrive scrambled, leaving us unable sort out our own productivity.
Because it takes a lot of mental effort to switch between tasks of a various types, a computer could begin to pair primary tasks with interrupting tasks that are more symbiotic. For example: searching for a picture of a house while drawing the plan of a house would be more symbiotic, than having to write a sentence about Thomas Edison while drawing the plan of a house. Certain interruptions are more jarring because they command different parts of the brain. This limited capacity is probably best demonstrated in the classic example of patting ones head while rubbing the belly in a circular motion. Its not as easy as it seems. A computer could manage the sequence of tasks a person performs to optimize a person’s focus with tasks that pair well together.
The hope is that as we move into the future our devices will become more empathetic to demands they make of us, thus predicting our next tasks and organizing the cacophony of modern life.
I expect that sometime in the not so distant future I will be prompted to meet certain people, perform design tasks, make specific decisions according to what mental/ physical capacity I have at that given moment. If I am nearby a person whom I intend to meet and have a clear schedule will I be prompted to meet? If I am on a train for a few hours my computer will prompt to write that speech on my to do list and thus automatically silence my text and e-mail communications temporarily. The implications for this sort of machine learning is infinite. It may just make us super human.
The development of motion studies and management science during the industrial revolution broke down human motion into a machine-like process. This mechanization changed standard workflows, such as building a wall, into a series of specific motions that optimized the efficiency of that labor.
Now rather than mechanizing the specific motions of a task, automation will drive what tasks are and when they take place. Computers can detect what type of tasks are most complimentary (i.e. least distracting) when performed together. The study of pairing complimentary tasks goes far back. In the 1980s Christopher Wickens described “Multiple Resource Theory” which described the phenomenon of how workload varied depending on how harmonious various tasks were.
“Wickens’ MRT proposes that the human operator does not have one single information processing source that can be tapped, but several different pools of resources that can be tapped simultaneously. Each box in figure 1 indicates one cognitive resource. Depending on the nature of the task, these resources may have to process information sequentially if the different tasks require the same pool of resources, or can be processed in parallel if the task requires different resources.
Wickens’ theory views performance decrement as a shortage of these different resources and describes humans as having limited capability for processing information. Cognitive resources are limited and a supply and demand problem occurs when the individual performs two or more tasks that require a single resource (as indicated by one box on the diagram). Excess workload caused by a task using the same resource can cause problems and result in errors or slower task performance."
I conducted an experiment that directs a test subject to perform tasks requiring engagement from various parts of the brain. These tasks were randomly mashed together with a script so that some would be more symbiotic and others more disjointed.
The computer selects from a list of instructions, and delivers this request to a test subject. As the test subject is performing this primary task, the computer selects
from a list of interrupting tasks at a 30 second interval.
Primary tasks that were based on tasks that typically take approximately 2 minutes to complete. Tasks were modeled after a PRP (Platelet-rich plasma) study performed by Hawkins, Rodriguez to classify the difficulty of switching between different cognitive functions in the brain.
Tasks were organized from specific to vague, because vague tasks tend to require greater mental focus for decision making. Tasks in Hawkin’s experiment required users to switch between tasks such as naming geometric shapes, computing arithmetic to responding to visual stimuli.
The tasks of The Computer Instructor experiment have been adjusted from Hawkin’s model to reflect tasks of a designer. The tasks span from specific to vague (0. Rote, 1. Logical, 2. Memory, 3. Spatial Design, 4. Creative Design)
Interruptions are categorically different from Tasks in that they are tasks which the subject has no control over. A designer may be instructed a task to pick-up redlines, whereas emails and phone calls can demand immediate attention and are much further out of our control.
The interruptions for ‘The Computer Instructor’ are again modeled after the Hawkin’s experiment from specificity to vagueness, however this time the interruptions are based on
their relevance to the primary task. Interruptions prompt the subject to read, describe, or view material that is either related to the primary task or non-related to the primary task. This is meant to emulate a scenario where a designer is asked to describe an issue from a project she is not currently working on and is then required to mentally switch gears in order to answer immediate interruptions.
On the Rhino screen the subject is prompted to complete the given task in the work area window. During a 30 second interval, the subject receives an interruption task to complete.(fig 8)
Performance is based on the time taken for each task rather than the quality of the work performed. Its important to note that in most cases the quality of the work performed was not heavily impactedby interruptions.
However when vague tasks were paired with interruptions that helped to clarify the task direction, tasks actually took much longer because the subject spent more time trying to remember and implement what was found from the interruption. For example, when the Memory Task, “Draw a cross section of the pantheon” was paired with, the Topical Interruption, “Google search images of the task you are working on,” the subject spent more time trying to accurately draw the pantheon based on Pantheon section images during the Google search (fig 12).
Each task was judged based on how much extra time it took to complete with interruptions versus how much time it took to complete with no interruptions (its baseline). This is called the switching cost or reaction time of a task.
RT = reaction time
baseline time = time it takes to complete the task without interruptions
switching time = time it takes to complete the task with interruptions
RT = baseline time - task with switching time
For a table of that data See Figure 15. For a graph of the reaction time (switching cost) see figure 16.
The best switching costs are shown in green for each task type. The worst are shown in red. With this information we can then give a preference to pair tasks with interruptions with lower switching costs.