Donchin's ERP work in the Times

Zeke Koch zeke at reed.edu
Tue Feb 9 21:20:19 EST 1993


Does anyone know anything about Donchin's work on Computer Human
Interaction?  I saw the following article in the times today.

--------
THE NEW YORK TIMES, TUESDAY, FEBRUARY 9, 1993

Computers Are Starting to Take Humans' Wishes as Their Commands

By ANDREW POLLACK

ATSUGI, Japan

  People now control computers with a keyboard, a 
mouse or in some cases with spoken commands But at Japan's largest 
computer company, Fujitsu Ltd, and at several other laboratories 
around The world, researchers are developing ways to control a 
compu~er by merely thinking a command

A New York State Department of Health research Team has developed a
system that allows users, after some training, to move a cursor slowly
up and down or side to side on a computer screen by mental action
alone.  University of Illinois psychologists developed a way of
allowing people to type, albeit at a rate of only 2.3 characters a
minute, by spelling out words in their minds

And at the research laboratories of the Nippon Telegraph and Telephone
Corporation, Japan's main telephone company, researchers have devised
techniques to tell from brain waves, with a fair degree of accuracy,
the direction a person will move a joystick. A similar project is
under way at Graz University of Technology in Austria.

"This is no parapsychological exercise," said Emanuel Donchin, a
professor of psychology at the University of Illinois who led the
development of the thought controlled typewriter.  Rather, such
mind-over-cursor techniques work by having computers analyze electric
signals emitted by the brain as it works The signals are collected by
electroencephalography, or EEG, a technique that involves attaching
electrodes to the scalp. It has long been used to diagnose brain
disorders.

Complete human-brain computer interaction is certainly decades away
and might never move beyond science fiction. But in the next decade,
practical if limited systems for helping severely handicapped people
communicate or operate appliances are seen as feasible. A related
technique in which electrical signals to the muscles are detected and
analyzed is also being explored to help paralyzed people operate
artificial arms or legs.

If we can use a computer without even uttering a sound, it would be
easier," said Norio Fujimaki, one of the three researchers
participating in an experimental program on thought driven computers
at Fujitsu's research laboratory in this city near Yokohama.

Attempts to develop thought input for computers began in the 1970's
with the "biocybernetics" program financed by the United States
Defense Department. One goal was to enable a computer to determine the
state of mind of a fighter pilot so it could better assist him in
operating the plane, said Professor Donchin, who was involved in the
work.

But the program was discontinued. in the early 1980's, and since then
work in this field, aimed mainly at medical uses, has been sporadic,
hurt by shortages of financing and technical obstacles.  Research in
this area often raises concerns about whether technology will be
developed to read minds. But Professor Donchin and others, say that
most of the systems under development cannot eavesdrop on a person's
thoughts.

Indeed, for now and in the near future it is a major challenge to
recognize from brain waves if a person means "yes" or "no," let alone
to understand complex thoughts. That is because there is little
understanding about the connection between any particular thought and
the voltages emitted by brain cells.

Moreover, any one signal may be drowned out by the signals from all
the other brain activities going on at the same time.

Don't Breathe, Please

"It's difficult enough to have a speech recognition device, but there
you know the language," said Erich Sutter, a senior scientist at the
Smith-Kettlewell Eye Research Institute in San Francisco who developed
a system using EEG that can tell where on a computer screen a person
is looking. "With EEG signals, we really don't know the language the
brain uses, and the brain may be doing all sorts of things unrelated
to the thought you are trying to dig out."

Consider the first efforts at thought input by Dr. Fujimaki of Fujitsu
and his collaborator, Prof. Shinya Kuriki of Hokkaido University.

A volunteer sitting in a chair would have 12 electrodes attached to
his or her scalp. Because any movement, even blinking or looking at
the scenery, would generate a brain signal 10 times larger than the
one the researchers were trying to detect, subjects had their heads
locked in one position with a special brace. They were told to stare
at a black dot and to breathe, blink and swallow as little as
possible.

The subjects were told to say the sound "ah" in their mind, without
actually voicing it, when they saw one color of flashing light, but
not to say it when they saw another color. By averaging dozens of
readings, Dr. Fujimaki could detect a difference in brain pattern when
a person was mentally saying "ah"

`It's Far From Practical'

But the need to take so many readings rules out the use of the
technique for computer control. Ideally, a person would want to think
the letter "a" only once and have it recognized. "In our experiment,
10 hours are required to communicate only one vowel " Dr. Fujimaki
said. "It's far from practical communications."

Other researchers have made more progress by using particular signals
that are easier to detect and analyze.

At the University of Illinois, Professor Donchin took advantage of
what; is known as the "oddball paradigm." When someone sees something
that he or she has been waiting for but that occurs only rarely, the
brain emits a detectable signal about three-tenths of a second later.

To develop his brain-activated typewriter, Professor Donchin arranged
the letters of the alphabet in rows and columns that were displayed on
a computer screen. The rows and columns were flashed one by one in a
random order. When either the row or the column containing the letter
a person was thinking about flashed on the screen, the person's brain
would emit the telltale signal. By knowing the row and column, the
computer could then identify the proper letter.

Disciplining the Brain

At the New York State Department of Health's Wadsworth Center for
Laboratories and Research in Albany, Dr. Jonathan R. Wolpaw and his
colleagues get around the problem of having a computer try to guess
what the brain is thinking. Their approach is to train the brain to
emit signals that can be easily understood by a computer. "It's
putting the task on the brain," Dr. Wolpaw said.

Dr. Wolpaw's technique uses mu waves, which are rhythmic signals
emitted by the brain's sensorimotor center when it is in idle mode.
In Dr. Wolpaw's system, electrodes measure the amplitude of the mu
waves and translate large amplitudes into an upward movement of the
cursor and low amplitudes into a downward movement.

In one experiment, four of five subjects gradually learned to control
their mu waves enough to move a cursor from the center of the screen
to either the top or the bottom in about three seconds.

Some subjects found that particular thoughts, say, about weightlifting
would move the cursor down, while thoughts about relaxing moved the
cursor up. After a while, such imagery was no longer needed, Dr.
Wolpaw said.

By using more detailed measurements of the mu rhythms, Dr.  Wolpaw's
team has recently succeeded in enabling people to move the cursor side
to side as well as up or down. But people still cannot bring the
cursor to a particular point and stop, a level of control needed to
develop the mental equivalent of a computer's mouse.

Akira Hiraiwa and his colleagues at Nippon Telegraph and Telephone
have taken advantage of the fact that the brain emits certain voltages
before an action, is taken. They developed a pattern- matching
computer kno


More information about the Neur-sci mailing list