As AI development continues to ramp up, researchers are
figuring out if ethics and law can be embedded into AI itself.
The use of artificial intelligence (AI) has been a talking point for
militaries, especially in recent years, as they ponder how much of warfare can
be conducted without human involvement. This comes as little surprise given how
AI capabilities have continued to expand. Computers can
read stocks,
translate speech, and even
make medical diagnoses better than doctors. The idea
of autonomous warfare is no longer a hypothetical casually discussed around the
dinner table.
In February, the
US Army launched a new initiative, Advanced Targeting and
Lethality Automated System (ATLAS), to design vehicles with AI capabilities for
increased lethal accuracy and ground combat capabilities. Widespread attention
was given to the ATLAS plan following
reports that
the US defense department had plans to upgrade the current ATLAS vehicles used
on ground combat vehicles to not only help human gunners aim, but also have the
capacity to shoot autonomously.
While the Department of Defense has
policies stating
that humans would always make the final decision on whether armed robots can
fire against targets, commentators like Stop Killer Robots have expressed fears
that the lack of a ban against lethal autonomous weapons would leave room for
the world to enter into a
destabilising
robotics arms race.
"Delegating life-and-death decisions to machines
crosses a moral 'red line' and a stigma is already becoming attached to the
prospect of removing meaningful human control from weapons systems and the use
of force," Stop Killer Robots said.
The state of military AI regulation
There are currently no explicit conventions or laws on a
global scale that define the parameters for how autonomous weapons can be used.
The only legal guidance on restricting the use of autonomous weapons is in
Article 36 of the United Nation's (UN) Convention on Certain Conventional
Weapons (CCW), which only requires militaries to consider whether
weapons—autonomous or not—should be prohibited.
More about artificial intelligence
Factors
considered when determining whether a weapon can be
deployed includes if its operator: Knows its characteristics; is assured the
weapon is appropriate to the environment in which it is deployed; and has
sufficient and reliable information on the weapon in order to make conscious
decisions and ensure legal compliance. But unless there is a clear,
disproportionate misprioritisation of military necessity over humanity, states
will often have leeway under Article 36 to deploy a weapons system.
The question on the table for diplomats has been whether
autonomous weapons should be prohibited. Since 2014, the UN has organised
forums—now called the meeting of the Group of Governmental Experts on Lethal
Autonomous Weapons Systems (GGE on LAWS)—to discuss the use of lethal
autonomous weapons. In these meetings, countries party to the UN have explored
topics ranging from human responsibility in using autonomous weapons to reviews
required for weapons to be deployed, with the aim of ensuring AI does not
overturn the international legal framework for armed conflicts.
At first glance, the creation of a forum focusing on
preserving human control in warfare is a step in the right direction. But
Australia's Trusted Autonomous Systems Defence Cooperative Research Centre
(TASDCRC) chief scientist and engineer Jason Scholz, who spoke with
TechRepublic, said that while the forums have good intentions, they have not
been comprehensive enough in discussing the systems of control when using a
weapon.
"It's not only about what can happen in terms of
selecting a target and engaging it, autonomously or not, but the reliability of
the weapon ... the context in which it's used, training, certification of the
people and technology, authorisation, no one from a confident military sticks a
weapon directly into a target without having an entire system of control,"
Scholz said.
The most recent GGE on LAWS meeting took place in Geneva in
late March, with over 90 countries in attendance. Much like the previous
meetings, little progress was made about the use of lethal autonomous weapons.
The United States, United Kingdom, Australia, and Israel all
put forward the opinion that a better understanding of the technology was
needed before any restrictions are made.
With countries being unable to reach a consensus regarding
lethal autonomous weapons, the chatter has arguably resulted in more questions
than answers. As a result, there is still little explanation for how to
distinguish between autonomous weapons and other AI military technology, and
the extent to which autonomous weapons may be used. In this way, the framework
for military AI shares many parallels with the internet, where there are no
clear rules of engagement in handling online attacks.
There also doesn't appear to be any signs of military AI
spending slowing down, with the United States reportedly spending
at least $2 billion on military AI R&D, and the
United Kingdom also spending
£160 million.
The AI conversation in society
With the AI conversation coming to a standstill in the
military domain, there has been little regulation created on the civil front as
well. Pundits like the
American Civil Liberties Union have been vocal over
their concerns surrounding the AI usage of companies and civilians, warning
against the bias and deception that can come with the use of this technology.
Where action has occurred, it has been primarily driven by
employee protests at tech companies, like the ones at Google and Amazon. Amazon
employees last year
protested the sale of its facial recognition services to
police departments. Meanwhile at Google, thousands of its employees
signed
a letterprotesting against the company's involvement in a Pentagon program
that uses AI to improve the targeting of drone strikes.
Following the protests, Amazon and Google placated to
minimise the damage caused to their reputations. Amazon
announced in a blog post that it would support
government regulation for AI technologies such as facial recognition, while
Google created a set of
AI
principles that aim to safeguard against the creation of AI systems
that lead to bias. As part of the principles, Google said it
would not build AI weapons, with the company then
announcing it would not renew its contract for the Pentagon drone program it
was criticised for taking part in.
Google also created an external AI ethics council in late
March, but it was scrapped in just over a week in reaction to another set of
employee complaints. Thousands of Google employees
signed a petition to remove one of the board members,
Kay Coles James, due to her past comments on transpeople and climate change,
which set off a domino effect of the other board members resigning from the
council. Among the issues the ethics council was set to explore was whether to
work on the military applications of AI.
"It's become clear that in the current environment,
[the AI ethics council] can't function as we wanted. So we're ending the
council and going back to the drawing board. We'll continue to be responsible
in our work on the important issues that AI raises, and will find different
ways of getting outside opinions on these topics," Google
said.
While Scholz did not provide comment about the AI principles
or ethics council of Google, he said the reactionary positions taken by tech
companies has skewed the perceptions around military AI. By inflating fear
around the use of military AI, Scholz said, it has made it more difficult to
have public discussion about how military technology can evolve.
It's become clear that in the current environment, [the AI
ethics council] can't function as we wanted. So we're ending the council and
going back to the drawing board.Google
But even in situations where companies have taken a firm
position, it is unclear whether ethical principles created by companies have
any real, tangible impact. Microsoft, Facebook, and Axon—which
makes stun guns for US police departments—have all
created their own sets of AI principles.
In Microsoft's AI principles, the company does not provide
any explanations regarding its approach towards handling AI technology in
military and weaponry contexts, preferring to use broad, umbrella terms such as
"fairness, reliability, inclusivity" instead. Microsoft, according to
a
New York Times report, had announced in October it would
sell technologies to the United States government to build more accurate drones
or compete with China for next-generation weapons.
In a
report
[PDF] written by the AI Now Institute in 2018, a research group at New
York University, experts questioned whether these actions are just vehicles
used by companies to deflect criticism due to the lack of accountability
mechanisms currently in place.
"These [ethical] codes and guidelines are rarely backed
by enforcement, oversight, or consequences for deviation. Ethical codes can
only help close the AI accountability gap if they are truly built into the
processes of AI development and are backed by enforceable mechanisms of
responsibility that are accountable to the public interest," the report
said.
Is embedding ethics into AI the solution?
Rather than focusing solely on the development of
regulations and principles around the use of AI in military contexts,
University of Queensland Law School associate professor Rain Liivoja told
TechRepublic that an alternative solution could be embedding existing ethical
and legal frameworks into the military AI itself.
Beyond UN conventions such as the CCW and the Geneva
Convention, there are few prescriptive laws and ethics for militaries to
follow. For most decisions, militaries follow armed conflict principles such as
proportionality, distinction, and military necessity. Is the firepower used
proportional to the military objective? Will the military decision sufficiently
distinguish between combatants and civilians? Or is the military objection is
even necessary in the first place?
Liivoja, who is currently in a research team from the
University of Queensland and University of New South Wales—along with
Scholz—are undertaking a AU$9 million study into the application of ethics and
the law into autonomous defence systems. The five-year project, which is the
biggest investment in the world into understanding the social dimensions of
military robotics and AI, will attempt to clarify the legal and ethical
constraints placed on these systems, as well as the ways in which autonomy can
enhance compliance with the law and with social values.
Acknowledging the tightrope that exists between violating
human rights and improving a nation's security, both Scholz and Liivoja told
TechRepublic in separate conversations that there are various use cases for AI
in the military that do not relate to targeting.
Among them is the use of AI to prevent weapons systems from
firing at targets wearing protected symbols, such as the Red Cross, Red
Crescent, and the Red Crystal. Firing at individuals that wear these symbols is
a violation of international humanitarian law. While this type of technology is
still in development, Scholz said, such preventative AI could be applied to any
weapon, regardless of whether it was autonomous or not.
"A conventional weapon with AI that can recognise
symbols may be used to direct it away or self destruct to avoid unlawful harm
on something that has that protected object," Scholz said.
"That's an example of something that is clear-cut and
potentially doable because unlike AI for image netting, which has to tell the
difference between various objects, this would be a lot simpler as it may just
need to determine whether its a Red Cross or not."
A conventional weapon with AI that can recognise symbols may
be used to direct it away or self destruct to avoid unlawful harm on something
that has that protected object.Jason Scholz
The use of AI in military contexts could mitigate against
potential mistakes by humans, Scholz added, explaining that situations like the
MH17 disaster where a passenger plane was allegedly shot down by Russian troops
could be averted in the future through such technology.
So while an autonomous machine gun triggered by heat sensors
is clearly something to be banned, the question around banning all military AI
becomes much more difficult to answer if the tech is merely there to help
soldiers navigate the questions around proportionality, distinction, and
military necessity, rather than to make the decision for them.
While there is no quick fix to the ethical dilemmas
surrounding AI, particularly in determining whether lethal autonomous weapons
have a place in warfare, Liivoja said that now is the time where technology
specialists should use their power to drive the regulatory conversation as AI
development is still in its infancy.
"It's a bit too late to start thinking about the rules
once the technology has been widely adopted—the responsible thing to do is to
consider the impact while the tech is being contemplated and developed and that
is what we are trying to do," Liivoja said.
"Increased focus on law and ethics around new
technologies, irrespective of the walk of life the new technologies are being
implemented, is required."
Also see
