Share this article:
In August of this year, a loose coalition of tech company CEOs and artificial intelligence experts wrote an open letter to the UN Convention on Certain Conventional Weapons. Their concern is the development of autonomous weapons systems--robot war machines capable of locating and killing without human intervention.
Elon Musk, who has become in some ways the public face of this movement, tweeted out "If you're not concerned about AI safety, you should be. Vastly more risk than North Korea." Yet this is a technology that continues to advance at a pace that few people are even aware of.
Most military forces already possess military drone technology and the United States military has made extensive use of unmanned hunter-killer drones to bomb terrorist targets in Asia and the Middle East. The key difference is that the Predator and Reaper drones of the Airforce are controlled by remote pilots on the ground whereas the next generation of military robots may remove the human element from the picture entirely.
At present, human pilots on the ground pilot the airframes, select targets, receive clearance from commanders to use lethal force and then ultimately pull the trigger, just as a conventional fighter pilot would. Despite the fact that current military law also requires this, weapon systems are in development that would select and eliminate targets on their own.
One small such drone under development scans the battlefield for radio signals and when it detects them, immediately dives and detonates a grenade. No human control required. Others operate in swarms of armed drones that communicate with one another to coordinate and select targets.
The Pentagon's plans to merge artificially intelligent machines into drones, tanks, boats and a human-like robots would push strategic and tactical decision making away from human control in what experts are calling the "third revolution in warfare."
Able to make decisions more quickly than any human commander while considering thousands of strategic variables, the use of coordinated AI would confer a vast advantage over less capable armies on the same level that nuclear armed militaries with air superiority outmatch conventional, ground-based armies. As soon as the strategic advantage becomes clear, the rules preventing its use will be abandoned, experts believe.
The United Nations is taking the threat of an arms race serious as well, with a meeting set for November of this year to discuss the ramifications of continued development of artificially intelligence, coordinated weapons systems. A panels of governmental experts on "lethal autonomous weapons systems" has also been convened.
The open letter written by the 116 AI experts called on this new committee to "work hard at finding means to prevent an arms race in these weapons, to protect civilians from their misuse, and to avoid the destabilizing effects of these technologies." But many fear that any ban is "unworkable and unenforceable".
Amir Husain, Founder and CEO of AI company SparkCognition, is a signatory of the letter but warned that an outright ban would "stifle progress" and innovation. He believes that "the solution--as much as one exists at this stage--is to redouble our investment in the development of safe, explainable and transparent AI technologies."
The general manager of SparkCognition's defense division, Wendy Anderson, also reminds us that any ban on development that the United States (or any other country) accepts would put the US at a competitive disadvantage--or simply drive the development underground to be conducted in secret with less controls. The sentiment that "We cannot afford to fall behind" isn't felt just by Anderson.
The possession of powerful artificial intelligence and coordinated, autonomous combat systems would give a clear military advantage, but experts warn that the Pentagon does not fully comprehend the risks that uncontrolled AI represents. "AI is not just another technology," Andy Ilachinski, head research scientist at the Center for Naval Analyses, said in a recent interview. He went on to say how it is on the cusp of transforming the world on the level of the printing press or the Internet.
Unlike with conventional computer systems, the next generations of artificial intelligence develop their own solutions to problems in a way that is opaque to programmers. As the machine evolves itself to be more efficient and capable it also moves farther from human understanding and control.
The dangers of fully autonomous weapons systems are multiple from hackers to conflict escalation and even to extinction level events. Despite the risks, the advantages of such systems are so great that once Pandora's Box has been opened, there may be no way to close it.
Hackers: imagine a team of Russian hackers and that overrides the control stream of a division of combat drones, tanks and robotic soldiers. Weapons could be turned on friendly populations, destroyed or simply integrated into the Russian military. Surveillance drones have already been hacked by Iranian hackers, redirecting them to Iranian airstrips.
Escalation and endless conflict: in democratic societies with conventional weapons, casualty figures are a strong argument for peace. Foreign wars that send thousands of young men and women home in coffins become politically untenable.
With wars fought entirely by robots, there is no such political resistance to conflict. Strong countries would be free to bomb and murder less advanced countries with impunity and two countries fielding robotic armies could become locked in an ever escalating and bloody war with no end.
AI singularity: a far worse result, and which Elon Musk and the other 115 signatories of the letter to the UN consider a very real threat, is the possibility of an artificial intelligence that advances beyond human control. It is frightening to think that Skynet, the antagonist of the Terminator series, now has a very real possibility of becoming a reality. A self-aware, superintelligence in control of the nation's military would be, as analysts warn, an extinction level event more dangerous even than nuclear weapons.
In the past year, AI has succeeded at defeating the best human players in go, the board game of strategic thinking, a game far more difficult for machines than chess. Already better at humans in medical diagnosis (IBM's Watson) and target recognition, AI chatbots tests by Facebook recently were observed to quickly develop their own language more efficient than English but almost immediately unintelligible to researchers.
Once all the world's military is in the hands of artificially intelligent, self-aware machines, humanity could easily become a nuisance to be wiped away just as we would bacterial infection. The letter to the United Nations warns that "Once this Pandora's Box is opened, it will be hard to close. We do not have long to act."