Digital Human That Errs: A Spontaneous Digital Human Project
Toru NAKATA
Digital Human Research Center, National Institute of Advanced Science and
Technology (AIST), Tokyo.
April 1, 2004.
Last updated: August 11, 2004.
"Give me a fruitful error any time, full of seeds, bursting with its own corrections. You can keep your sterile truth for yourself." -- Vilfredo Pareto.
Fig. Flow of DHTE Process
| Subject | Digital Human That Errs | Real Human |
|---|---|---|
| Cost | Low. Reusable. | High. (Reward and Time) |
| Stability | Controllable | Drift. (Get tired or learn.) |
| Safety | Simulation. (Allows Hazardous Experiments.) | Cannot hurt. Ethic problem. |
| Quickness | Computer. Run it at night. | Take long time. |
| Sensitivity | Increase error probability in order to detect drawbacks of the interface. | Rarely makes serious errors. |
| Breadth | Provides special subjects and situations. | Limited variety of subjects and situations. |
Disasters have been caused not only by machines, but also by humans. Machines may produce accidents, but we can estimate the type and amount of accidents by analyzing accident data statistically.
Humans may cause accidents, not as intentional acts, but as errors. Can we make models of human reliability?
It is difficult. Already ergonomics has produced huge data bases of Human Error Probability (HEP), but the databases are observational and too abstract. What the data bases record are things such as "Nominal HEP on numeric input task is 0.003," or "HEP of omission of item with check-off provisions is 0.001." [1].
But the human powers of cognition and memory are greatly dependent on the situation. HEP values sounds plausible, but they are determined without detailed background on the situation. Actual human interfaces with machines are varied and complex. The type and condition of human users also differs. What the user thinks and wants also influences HEP.
Imagine you are a designer of human interfaces for a heavy-duty machine like a power plant. You have to validate usability of the interfaces. You have to conduct ergonomic experiments over various design plans with various human subjects and situations.
It is too much. Ergonomic experiments cost money and time. You cannot afford to check all combinations of experimental conditions.
The human cognitive system is very complex, but each component of the system can be represented by a simple model.
There are many common misspelling such as "STOP" as "SOTP". This typical error on ordering items is called the "bath tub effect". Humans can pay attention to starting and ending a sequence, but the middle of a sequence is relatively hard to memorize and complete.
Ergonomics have catalogued various typical human errors quite well. Each of the typical errors is very simple to translate into a mathematical model. Combining them, we will make digital human models that err like humans.
The scope of our challenge includes the followings:
- Modeling of actuation error: Tremor.
- Modeling of perceptional error: Illusion.
- Modeling of cognitive error: Misconception.
- Modeling of problem-solving error: Illogical Intelligence.
- Modeling of feedback and learning: Evaluation of the credibility of information.
These are frequently asked questions against my DHTE virtual user concept.
A: No. Prediction of human behavior and error is almost impracticable.
However, we do not aim at such precise forecasting of human behavior. We think DHTE simulation may have ability to find the weakest link in a man-machine interface.
Observing the behavior of beginners shows the more difficult parts of a human-interface. Likewise, errors made by virtual users suggest the weakest link of a human-interface.
Accelerated Test of Human Interface Human Interface Evaluation like a Weather-forecast Target of analysis Weakest drawbacks Realistic forecast of user behavior Required Preciseness Low (Volatility of DHTE is rather desirable.) Strict Simulation Coverage Broad Narrow. (Forecast on particular person and particular condition.) DHTE Suitable for implementation Very hard to acquire preciseness of forecast.
A: Of course, modeling of human mental processes is hard, but DHTE may be useful for practical problems. Many serious accidents are caused by simple human errors. DHTE, with its low intelligence, may simulate such boneheaded errors.
Psychologists have made many kinds of mental process models. DHTE and other spontaneous digital human models may be improved by adopting them.
A: Affordance, or mental modeling, is a characteristic of objects that evokes typical behavior from users.
We do not want to simulate why and how typical behavior is generated.
In our DHTE concept, the simulator just makes the virtual user undertake a certain behavior when faced with some object.
For instance, when the virtual user is facing a door in the simulation world, he will stretch out his hand to the knob, expecting that the door can swing open. Of course, there is no guarantee that the door is a swing-type door.
Here is another example. Real humans tend to avoid walking in the center of a town road. Car traffic forces humans to move from the center of the roads to the edge. Not knowing the reason, DHTE simulator just commands the virtual user to walk along the edge of the roads.
This is enough to evaluate usability of a human-interface.
DHTE is a user model, but its concepts is different.
DHTE is a user model which is intentionally made error-prone.
- DHTE is expected to recover from errors that it made.
- DHTE simulator generates various errors, including unexpected error modes, and estimates rates of errors by tracing error processes.
- DHTE has a physical body model, in contrast to GOMS.
GOMS is a "one-way" and qualitative estimation. In most of GOMS estimations, behavior of a user model is determined before estimation [2]. Those estimations do not cover behavior for error recovery. User performance such as error rate and execution time are estimated qualitatively or determined beforehand by adopting Fitts's Law or Hick's Law. I do not think those "laws" are always valid enough for all kinds of human interfaces and situations.
However, GOMS and MHP have been developed over decades, and I think some part of their methodology can be utilized for DHTE. (Learn more from "2.2.5 Error Recovery Support" by John and Kieras[3].)
A: All errors are results of cognitive mismatch between subjective understanding of the world and objective, real conditions of the world. A user pushes a wrong buttons when he is thinking that he is touching the correct button. An operator skips a necessary step of a procedure due to mismatch between subjective understanding and the objective procedure.
DHTE simulator must imitate this mismatch. The simulation system should have "dual memory buffers" to register the subjective and objective data of the world.
To be appeared in Conference of HCII 2005 at Las Vegas.
Pending Patent: US-2005-0044450-A1: "SYSTEM AND METHOD FOR EVALUATING USABILITY USING VIRTUAL USER"
[1] David I. Gertman & Harold S. Blackman, "Human Reliability & Safety Analysis Data Handbook", Wiley-Interscience, 1993.
[2] F. J. Lerch, M. M. Mantei, & J. R. Olson. "Skilled financial planning: the cost of translating ideas into action", Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, pp. 121--126, 1989. (Available at ACM Digital Library).
[3] Bonnie E. JOHN, David E. KIERAS. "Using GOMS for user interface design and evaluation: which technique?", ACM Transactions on Computer-Human Interaction (TOCHI) Vol. 3 , Issue 4, 1996. (Available at ACM Digital Library).
Copyright(c) Toru NAKATA. toru-nakata@aist.go.jp
