IEEE Robotics & Automation Magazine - December 2018 - 118

dynamics, and similar paradigms of the
18th, 19th, and 20th centuries will prevail as long as we are constructing robots
that will exist on Earth and interact with
people. I also argue that we need to
understand the principles of transduction (sensors and actuators) so that we
can properly interpret the data, including their limits. This also addresses the
issue of the role of data-driven (large or
small, which is
different from
measurements)
Although our most
versus modelimportant job is to
driven analysis
and prediction
educate the next
of performance.
generation of future
In academia,
we often limit
roboticists, we must
ourselves to debe honest in our
monstrations of
principles, but
interactions with
we owe it to our
the public.
students to teach
them that they
mu st b e ve r y
careful when generalizing these results.
As Kumar [2] eloquently argues, we
need to pay attention to irrational exuberance [3] as well as fairness, accountability, transparency, and ethics [4]. Our
credibility is on the line. Although our
most important job is to educate the
next generation of future roboticists, we
must be honest in our interactions with
the public. We have to be careful when
reporting our results, especially when
we extrapolate from our demonstrations to robust performance.
Robotics in Service to Industry
Recently, there has been a great push
to bring university results into practice
by creating companies. This is more
prevalent in certain areas of robotics,
such as computer vision, image classification, speech and speaker identification, and language translation, than

118

IEEE ROBOTICS & AUTOMATION MAGAZINE

in others. Venture capitalists are showering midlevel and junior faculty with
investments in these areas. An alarming number of these faculty are leaving
universities for newly established
institutes and laboratories. The attraction is not only salaries but the availability of engineering power and
facility resources. Are these academics
promising to deliver an honest product, or are they being overly optimistic? When I see results claiming 85%
accuracy in visual segmentation, I
wonder how a given algorithm can
survive market requirements, which,
for accuracy, are usually 10-n, where n
is between four and six.
I do believe that some of the new
machine-learning technologies can be
very effective if
● the task is narrow and well defined
(e.g., medical imaging)
● we know what we are looking for
● the data acquisition is well controlled
(e.g., if magnetic resonance imaging
parameters are well defined and kept
constant throughout the entire data
acquisition process)
● the data analysis is transparent and
consistent (i.e., all of the imageprocessing parameters are constant
during the processing of all of
the data).
Unfortunately, in robotics we are several
years away from wide adoption of
industrial applications that use datadriven methods.
Of course, that does not preclude
the research community from exploring where and when these methods can
be applied. Consider robotic surgery: in
this case, we need to account for the
fact that every individual is different;
therefore, we cannot rely solely on previously collected data but must rather
have precise measurements of body
parts in real time to make decisions
about how to act. Preliminary data sets

december 2018

can certainly be our guide, but they are
not sufficient to determine our actions
during surgery.
Our Social and Ethical
Responsibilities to Society as
Scientists and Technologists
First and foremost, we must be honest
about what robotics and artificial
intelligence can deliver today and in
the future. We should very clearly spell
out under what assumptions our technologies will or will not work. We
must clarify the range of parameters
for which the systems we build will be
safe and reliable. Error analysis,
robustness, and privacy guarantees
should be the standards of such a system's descriptions.
Recently, I have observed that we've
adopted quite a bit of marketing methodology instead of adhering to scientific
clarity and sobriety. In the past, 50
years ago, academic presentations were
strictly technical, and marketing was
the preserve of business people. Today,
perhaps we have gone to another
extreme. What we need is balance.
References
[1] R. Bajcsy, "What should we teach our students in robotics to survive next 20+ years?"
Summer 2018.
[2] V. Kumar. Irrational exuberance and the
'FATE' of technology. 20 Aug. 2018. [Online].
Available: https://cacm.acm.org/blogs/blogcacm/230472-irrational-exuberance-and-thefate-of-technology/fulltext
[3] A. Greenspan. The challenge of central
banking in a democratic society. The Federal
Reserve Board. [Online]. Available: https://www
. federalreser ve.gov/boarddocs/speeche s /
1996/19961205.htm
[4] Microsoft. FATE: fairness, accountability, transparency, and ethics in AI. [Online].
Available: ht t ps://w w w.m ic ros of t .c om /
en-us/research/group/fate/

https://cacm.acm.org/blogs/blog-cacm/230472-irrational-exuberance-and-the-fate-of-technology/fulltext https://cacm.acm.org/blogs/blog-cacm/230472-irrational-exuberance-and-the-fate-of-technology/fulltext https://cacm.acm.org/blogs/blog-cacm/230472-irrational-exuberance-and-the-fate-of-technology/fulltext http://www.federalreserve.gov/boarddocs/speeches/1996/19961205.htm http://www.federalreserve.gov/boarddocs/speeches/1996/19961205.htm http://www.federalreserve.gov/boarddocs/speeches/1996/19961205.htm http://www.microsoft.com/en-us/research/group/fate/ http://www.microsoft.com/en-us/research/group/fate/

IEEE Robotics & Automation Magazine - December 2018

Table of Contents for the Digital Edition of IEEE Robotics & Automation Magazine - December 2018