Centre for Cognition & Decision Making
Volgogradsky Prosp., 46B, room 210
Time: 15.00
Presenter: Sergei Shishkin, Head of
the Department for Neurocognitive Technologies NRC Kurchatov Institute
For many years, experts considered brain-computer interfaces
(BCIs) mainly as an assistive and rehabilitation technology. Recently, the
passive BCI approach (Zander & Kothe, 2011, J. Neural Eng. 8:025005) has
become the basis for new prospective BCI applications which can become useful
for healthy people.
Operating a traditional BCI requires that a user performs certain mental tasks
or attends specific external stimuli. The BCI detects correlates of these
activities in the user's brain signals and translates them into commands or
messages. A passive BCI analyzes brain signals during the user’s usual
interaction with a machine, without requiring to perform any additional task.
The information about the user’s current brain state is used to improve the
interaction or for other purposes.
Similarly to other noninvasive BCIs, passive BCIs usually employ variations in
the electroencephalogram (EEG) spectral components or the components of the
event-related potentials (ERPs), especially the P300 wave. However, the range
of passive BCI solutions is now evolving especially quickly (Blankertz et al.,
2016, Front. Neurosci. 10:530). The ERP based passive BCIs were already applied
for the disambiguation of search queries (e.g., Google search for an image
using a keyword “chain” produces images related to its different meanings, but
only relevant images elicit a strong P300), for visual search enhancement
(finding a relevant face in a crowd also elicits a strong P300), for guiding
and teaching robots (when an observer notice the robot is doing something
incorrectly, an error-related potential is produced by his or her brain), for
stopping a car in emergency situations, for fatigue monitoring and even for
studying Libet’s “point of no return”. Gaze fixations are often used to enhance
such BCIs by indicating locations and time points where the target brain
potentials may start. Some applications benefit from the multiuser approach
(e.g., a BCI may use data from several observers who simultaneously watch the
same robot), which effectively compensate for the low accuracy of the
single-user based detection of the EEG/ERP markers.
We recently developed a hybrid passive BCI based system for a typical active
BCI’s task, namely for “clicking” objects on a computer screen (Shishkin et
al., 2016, Front. Neurosci. 10:528). In our system, active object selection is
based on gaze dwells, while the intended dwells are separated from the
spontaneous ones with a new passive BCI. This BCI detects an ERP component that
appears when the user expects the feedback from the interface (intentions are
translated into actions through expectation!). Our “Eye-Brain-Computer
Interface” (EBCI) has been implemented as probably the fastest hybrid BCI +
gaze near real-time system to date, using only 300 ms fixation-related EEG
segments for classification.
Why the development of effective non-invasive BCIs is not an easy task? What
kinds of neuro/psychophysiological, mathematical, engineering and programming
efforts are needed? Can we expect that the new technology of “passively”
connecting brain with machines – and, possibly, brains with brains?! – will
enable us to create human-machine systems with new emergent properties that
could be not only practically useful but also interesting from scientific and
philosophical points of view? We will try to find answers to these
questions.
