However, unlike most neural-net-
work applications, the algorithm’s in-
put (that is, the brain activity) is also
learning in parallel to the decoding
algorithm. “This co-adaptation pro-
cess can be powerful in that adjust-
ing the decoding algorithm allows
the person to explore new and poten-
tially easier ways to think about mov-
ing that will generate stronger brain
activity patterns and ultimately result
in more robust decoding of the ap-
propriate muscle activity,” explains
Taylor. “In any parallel adapting sys-
tem, finding an appropriate adap-
tation rate is key to ensuring stable
rapid improvement.”
While decoded brain activity has
been used to activate muscle stimula-
tors in a few studies and the ground-
work is now being laid for U.S. Food
and Drug Administration endorse-
ment, suggesting that the techniques
and technologies associated with BCI
systems are finally maturing, Taylor
notes there is still a long way to go. She
cites two main technical challenges,
in particular, as bottlenecks for fur-
ther progress. One is that researchers
working in this area need better brain-
recording technologies to extract high-
resolution brain activity reliably. The
other is the actual physical compo-
nents that go into a complete BCI sys-
tem must be made smaller and more
portable, and must also be simple
enough to use so they can be adjusted
without the aid of a research engineer.

“the brain is an
adaptable learning
machine,” says
Dawn taylor.
“We learn to do
a new dance move
or play tennis
through practice.
Learning to control
the movements of a
device directly with
the brain
is no different.”
Platform standardization