PMRExpo
with only the right information presented; “a nothing
more, nothing less” approach.
• Coevolution – AI should progress in line with user
feedback on performance, which should be continuously
measured.
• Trust – Users should be able to judge the value of
the information and recommendations presented by
AI, and the thresholds on minimum trust should be
context-independent.
Saptharishi gave the example of an AI recommending to
a dispatcher which unit should be dispatched to an event.
While sometimes the closest units might be recommended,
there might be times when units with certain capabilities are
better kept back for a future possible event. This could be
surprising to a dispatcher because it may be contrary to what
they would expect, creating the need for the AI to explain the
basis behind its recommendations to them.
Speaking earlier on the same topics at his company’s media
dinner, Saptharishi said Motorola Solutions deliberately
avoids the use of common architectures for its algorithms,
opting instead to build them from scratch to reduce their
vulnerability to attempts to disrupt them – the obvious
example being T-shirts with designs that stop the wearer
from being recognised as a person. If the algorithm is made
using common architecture, while “not everybody” can figure
out a vulnerability, “a relative expert in neural networks can
figure it out”.
One of the great truisms is that anything that looks easy
often requires a great deal of effort. One of machine learning’s
less commonly known traits is the sheer amount of manual
effort that goes into labelling the data that trains algorithms.
Saptharishi explained the amount of human work required is
being cut by a couple of orders of magnitude because of the
advent of unsupervised learning, clustering and bootstrapping
techniques. These reduce the amount of labelling required.
Synthetic data, which is modelled on real data, is also
increasingly being used to train machine-learning algorithms.
LTE now, 5G later?
There were a number of roundtable sessions that examined
broadband communication issues. Francois Dontot, head
of international carrier sales and enablement at Motorola
Solutions and his group discussed mission-critical push-totalk
(PTT) methods and concepts.
Speaking about his session, Dontot said: “I had a lot
of end-users who are really into the future of critical
communications. Some of them are using analogue or
regular radio, so are looking at PTT over Cellular (PoC) as
a replacement.”
Dontot recalled how one participant, from a highways
agency, discussed the challenge of coverage in tunnels. As the
organisation has a lot of users on analogue systems, it was
evaluating whether it should move all users straight onto a
carrier’s mobile broadband network, or whether they should
instead go in stages and move to DMR first.
The organisation had been in discussion with a carrier that
said it could deliver coverage in tunnels, but the bill for that
project turned out to be “outrageous”. It is now engaging
with another carrier, and it remains to be seen if it can move
directly to mobile broadband. Dontot said organisations
should look to “bring coverage at the right cost”.
Dedicated networks cost
money, and money is of the
essence here
Dr. Torsten Musiol, founder and CEO of MECSware,
dived into broadband applications and solutions. His group
discussed how the first 5G release, which has limited coverage
across Europe, is more “LTE plus” than “real 5G”. For that
reason, it is more important to focus on “what you can do
with LTE today”.
Musiol said one session participant discussed his
organisation’s need for overlapping video, and an associated
requirement for transmission in real time. This led to a debate
around what is meant by real time. For applications involving
machine control, it might mean one millisecond, whereas for
others it can be 100 milliseconds, one second or even longer.
End-users need to specify what they mean when they say
‘real time’, when speaking to those building the applications.
Musiol highlighted that “LTE can do a lot of this now”. For
5G to reach its full potential, participants in Musiol’s session
said 5G needs to be available “without a deadzone”.
Bernhard Klinger, PMeV’s chairman of the board,
focused on regulation and frequencies. He was surprised that
participants were open to the “co-use of public networks
for mission-critical broadband”. He attributed this flexible
mindset to their understanding that “dedicated networks cost
money, and money is of the essence here”.
Cyber threats, military mindset
Elsewhere, Professor Holger Mey, head of advanced concepts
at Airbus, discussed how technology can be turned into a
security superiority. Drawing comparisons between military
missions and the task of securing our digital world, Mey
compelled attendees to start taking cyber security seriously.
In his view, cyber-security missions should take the character
of classic defence missions. Technology cannot be stopped, so
organisations need to keep in mind that attackers will exploit
vulnerabilities and must put the right defences in place.
He said: “Cyber security teams are now confronted with
threats that have a military character. Organisations need
to identify what defences are needed and must learn how
to specify the expense and likelihood of an attack. Risk is a
product of likelihood multiplied by the damage level. You
don’t have fire insurance for the likelihood, but because of the
potential cost of a fire.”
The problem is that “whenever we introduce new
technologies, we don’t think about security. When we
introduced cars, it took 50 years before we had seatbelts.
Many companies look at achieving efficiency in peacetime
but not effectiveness in wartime. The competition between
offence and defence will always be there. That’s why you have
to continuously invest, and we have to bear in mind that
technology is only as useful as the mission or task it serves.”
Mey concluded that, when weighing up what cyber defences
are needed, “you simply need to be better than the other side.
They will be good. You have to be better.”
These are wise words given the rise of AI. While it is clear
from Saptharishi that a great deal of thought has gone into
how AI should be used ethically, we must not forget that
technology quickly falls into the wrong hands.
January 2020 @CritCommsToday 21