Simulation acts 7-year-old's age

By Kimberly Patch, Technology Research News

Researchers at Penn State have built a computer simulation that acts like a seven-year old in order to study the strategies children use to learn and to explore better ways to teach them.

In addition, the model may prove to be a patient subject for studying how to make computers and other electronic devices work in ways that are more intuitive to humans.

To build the computer model, the researchers first studied how adults solved a puzzle that involved building a pyramid with different-shaped blocks. They made a computer model of the adult behavior, then changed the model based on three basic cognitive theories: that seven-year-olds can't think as fast as adults, that they can't keep as many thoughts in their heads at the same time, and that they aren't as good at choosing a good strategy.

The changes included limiting the models' working memory so it could only work with a couple of features at a time -- like two sides of a block instead of all eight at once; and changing the way long-term knowledge was represented in the model by limiting the number of clauses an "if" statement could contain.

Changes based on the first two cognitive theories slowed the computer's progress and made it more prone to contradictions, said Frank E. Ritter, an associate professor of information science and technology and psychology at Penn State.

But the single theory that best fit seven-year-old behavior was the third one: that seven-year-olds are not as good as adults at being able to choose the right strategy, Ritter said. With its ability to choose modified, the model "not only did... not know what to start with [it] didn't know how to [complete the task] as well," said Ritter. He cautioned, however, that further study of combinations of factors will likely hone the model further.

The research was the the first step in creating a string of computer simulations of cognition at several key learning ages, said Ritter. Because the behavior of a seven-year old is closer to an adult's than that of a five- or three-year-old, the older simulation was the logical one to do first.

Another step, said Ritter, is adding a theory of motivation to the models. Real five-year-olds, for instance, tend to give up fairly quickly when faced with a task that's too confusing, while the simulated model will run for hours even if it's getting nowhere.

The simulations could eventually be used to find good teaching strategies and to test products designed to help children learn, Ritter said.

Ritter is taking the research in another direction that could eventually help computer designers see what frustrates people when they use computers.

Because the models are based on typical human behavior, what's intuitive to them should be intuitive to the average person, he said. They could eventually help solve a common problem in human-computer interface design: not enough testing is done because of the cost and time involved in using real people, he said. Today's practice of finding and bringing in one or two dozen users to test an interface, analyzing the results and writing it up, costs about $1,000 per user, said Ritter.

In theory, Ritter's models will eventually produce similar information at the touch of a button.

Ritter's initial project is a small model that dials an onscreen telephone like a human. "It predicts how long someone will take to dial a telephone number," said Ritter. The next step is using a "model of someone [to determine] the best order of the task: what's the best screen to give them," he said. That should be possible within a year, he added.

Long term, models like this could be used to test various devices. "I hope in the next five or ten years we will be able to take an adult model and say here's a VCR [and] a laundry list of tasks," and have the model respond that two of the tasks are easy and take 20 seconds to complete with no instructions, but a third task is undoable because after five minutes it got frustrated and gave up, said Ritter.

Using a cognition model like Ritter's to improve the way computers communicate with humans sounds plausible, said John Anderson, professor of psychology and computer science at Carnegie Mellon University. "I think it's a good idea in general to inform computer design with properties of the people that have to use it -- there probably hasn't been enough of that," he added.

Ritter, Gary Jones of the University of Derby in England and David J. Wood of the University of Nottingham in England wrote a paper on the cognitive model of a seven-year-old for the March, 2000 issue of Psychological Science. That research was funded by the UK's Economic and Social Research Council through the University of Nottingham.

Ritter's ongoing work on using cognitive models to improve device interfaces is funded by Penn State and Lockheed Martin.

Timeline:  > 1 year; > 5 years; > 10 years
Funding:   Government, University, Corporate
TRN Categories:  Applied Computing; Human-Computer Interaction
Story Type:   News
Related Elements:   Technical paper "Using a Cognitive Architecture to Examine What Develops" in Psychological Science, March, 2000


July 26, 2000

Page One

Tiny robots flex their muscles

Optical coax takes take turns

Making a caged ion blink

Simulation acts 7-year-old's age

Computer science faces decision theory


Research News Roundup
Research Watch blog

View from the High Ground Q&A
How It Works

RSS Feeds:
News  | Blog  | Books 

Ad links:
Buy an ad link


Ad links: Clear History

Buy an ad link

Home     Archive     Resources    Feeds     Offline Publications     Glossary
TRN Finder     Research Dir.    Events Dir.      Researchers     Bookshelf
   Contribute      Under Development     T-shirts etc.     Classifieds
Forum    Comments    Feedback     About TRN

© Copyright Technology Research News, LLC 2000-2006. All rights reserved.