Lot's Wife Edition 6 2015

SCIENCE & ENGINEERING 33
fully autonomous. They could easily have autonomous
navigation and mobile capabilities. With advances in AI,
targeting software could even be dynamic. Systems could
use a set of training data and efficiently learn while deployed,
then make their own decisions. Other targeting technologies
include facial and image recognition software. But what of
the triggering system? When, if ever, do we allow the machine
to make the call?
Take facial recognition software for instance. Facebook’s
experimental DeepFace algorithm is able to identify human
faces with 97.35% accuracy regardless of lighting or position.
This is almost as accurate as the human capabilities, which
score 97.5%. While this accuracy rate may seem high, we have
to wonder if we can ever entrust a machine with the task of
deciding whether someone should live or die if there is even a
2.65% chance it will make the wrong call.
Similar questions are currently being raised of the U.S.’s
drone program. Created in the name of eliminating terrorism
by Bush in 2001, Obama dramatically expanded the program.
As autonomous weapons, unmanned aerial vehicles (UAVs)
have the advantage of replacing soldiers, risking fewer U.S.
lives on the battlefield. However, the program is not only
widely criticised for its lack of transparency but also the high
number of civilian casualties. Since 2004, drones have killed
approximately 4,000 Pakistani people alone, 1,000 of them
civilians, including children.
Drones reveal several other flaws in our use of autonomous
weapons. Nobody knows how accurate the intelligence,
the data, or the machines themselves are. Nobody knows
whether or not the machine can then make the right call.
Nobody knows whom the machine might kill. And if and when
that happens, nobody will know whom to blame if meaningful
levels of human intervention and control are not defined. It
might be the programmers who select the targeting criteria,
the operators responsible for the machine, or the one who
signs off on the order. Whether it’s the judge, the jury, or
the executioner, we have to ask ourselves: who will we hold
personally accountable for the taking of a human life?
The use of UAVs and assorted autonomous weapons leads
us to question how they change the reality of warfare. While
their use removes humans from the battlefield, this also
removes the atrocities of war, allowing us to distance, excuse
and desensitise ourselves from what should be recognised
as a human tragedy. Additionally, some worry that machine
initiated attacks will lower our threshold for warfare.
"Personally, I believe warfare needs to stay horrific and
brutal. We need it to be so to ensure we only fight wars
as a last resort," says Toby Walsh, Professor of Artificial
Intelligence at UNSW and NICTA, and fellow signatory to the
letter. "Politicians have to see body bags coming home and
be prepared to justify why they risk the lives of our sons and
daughters."
Without regulating these weapons we face dire
consequences, perhaps even the ‘end of the human
race’ as Stephen Hawking warns. We must maintain
meaningful control. Experts are exploring the possibility
that the weapons might be taught to differentiate between
combatants and civilians. Lawmakers question whether they
abide by international humanitarian and human rights laws,
whether they can function ethically.
Consequently, the possibility that this technology might
be upon us within the coming decades necessitates such
a letter and the consideration of pre-emptive policies.
Campaign to Stop Killer Robots is one such group calling
for prohibition. These issues were also discussed during a
Human Rights Council session in April 2013, where questions
such as: "in what situations are distinctively human traits,
such as fear, hate, sense of honour and dignity, compassion
and love desirable in combat?" and, "in what situations do
machines lacking emotions offer distinct advantages over
human combatants?" were considered.
While many countries raised concerns, a ban was opposed
by the UK. "Is a ban on technology which has yet to be full
developed to maturity an appropriate course of action?
I suggest not," said Dr William Boothby in a separate
interview. A retired RAF air commodore and lawyer, Boothby
was responsible for ensuring that newly acquired weapons
conformed to the UK’s international humanitarian law
obligations.
Aside from weaponry however, AI has many more
applications and moreover, benefits. While fully autonomous
weapons rely on AI technologies, they are not one and the
same. In the simplest terms, an autonomous weapon is
able to make a decision. AI on the other hand, is far more
dynamic, and able to emulate more complex human
cognitive processes such as reasoning, problem solving and
planning, among other things. While a mathematical model
from the University of Wisconsin-Madison suggests that
computers will be incapable of human consciousness and
emotion, Hawking suggests that, "Humans, who are limited
by slow biological evolution, couldn’t compete, and would be
superseded."
As of July, the Nao robot created by French company
Aldebaran Robotics passed a self-awareness test. They
were given a riddle, and in order to answer it, had to show
understanding of the question, recognise its voice as distinct
from other robots, and then link this back to the original
question to show self-awareness. Additionally, in June,
computer programme "Eugene Goostman" passed the Turing
test, albeit mildly unconvincingly. The Turing Test assesses
a robot’s capacity to mimic human thought and speech in
a conversation. It should be noted that each of these tests
are inherently flawed, however. A robot may be specifically
programmed to emulate both of these functions.
Robots may also be used to automate industrial processes,
increase productivity and perform tasks too dangerous or
impossible for people. AI technology from Honda’s humanoid
robot ASIMO may be used in disaster response. Yet despite
the increasing intelligence and application of robotics and
AI, any responsible programmer will ensure their machine
has safeguards. ASIMO is programmed to shut down should
it attempt to build itself. Asimov’s laws of robotics stipulate
that: "A robot may not harm humanity, or, by inaction, allow
humanity to come to harm." So while robots may replace us
and take our jobs, there’s no need to worry about them taking
our lives.
That is, unless they learn how to reprogram and control
themselves.