The achievements of recent years in the field of machine learning have made it possible to create autonomous robotic systems, including military ones. In this regard, a discussion about the potential risks associated with the development of artificial intelligence has flared up in the scientific community with renewed vigor. Recent research has shown that decisions made by computers are often viewed as unfair by humans, and control over superintelligence is impossible in the long run due to the fundamental principles underlying it.
From fantasy to reality. The author of the term "artificial intelligence" (AI) is considered to be Alan Turing, a British mathematician, one of the founders of the theory of computer science. He talked about it back in 1950.
For more than half a century, the theme of the rise of the machines was extremely popular, but in general it did not go beyond the scope of science fiction novels and films. Seriously, the existential risks associated with AI began to be discussed relatively recently - after digital technologies have deeply penetrated into all spheres of life.
Without a computer and a smartphone, a modern person is not able to do even a few hours. Even a short-term communication failure or Internet outage can paralyze entire industries, cause a transport and energy collapse.
But the “stopping problem,” as experts call this situation, is not the worst prospect. As long as machines can be turned off or reprogrammed, they remain under human control. The real threats will arise when AI becomes smarter than us. And mankind has come close to this line.
Smarter than the smartest. Today, AI analyzes medical images more accurately than any doctor, optimizes logistical routes better than the most experienced dispatcher, and performs thousands more production and management operations more efficiently than a human. And the power of supercomputers for processing and analyzing information is many times greater than the capabilities of all people on the planet combined.
According to the Swedish philosopher, Oxford University professor Nick Bustrem, known for his work in the field of transhumanism, the era of superintelligence is coming - artificial intelligence, which is not only "smarter than the best human brains in almost all areas", but is also ready to take on creative functions traditionally considered the prerogative of man. But the main thing in superintelligent systems is the ability to act independently, to make their own decisions based on the analysis of external information.
The computer superintelligence is going to be entrusted with the management of individual processes in production and logistics. Purely technically, this is not difficult to implement, but it is not clear to what extent a person will be able to correct the actions of the machine and how great is the risk that AI will get out of control at some point. In addition, it is not clear how to put into the computer the concepts of ethics and justice that affect human decision making.
Justice versus rationality. Although the ethical issues associated with the use of superintelligence remain open, such systems are already actively used, including in the social sphere. German scientists have found out how people react to it.
The participants of the experiment were asked to evaluate on a conditional scale of fairness the decisions that were made in automatic, semi-automatic or completely manual mode in four specific situations: when applying for a bank loan, choosing a candidate to fill a vacancy, litigation, and assigning unemployment benefits. The worst were automatic decisions, and the fairest were those made jointly by a computer and a person.
Artificial intelligence has an undoubted advantage in processing large amounts of data, but does not take into account the accompanying circumstances - the so-called decision context. And in situations that affect the interests of people or concern their health and life, this can become a critical factor.
"Superintelligent" Kalashnikov assault rifles. In 2015, hundreds of the world's leading experts in the field of robotics and artificial intelligence published an open letter to the UN calling for a ban on lethal autonomous systems capable of independently identifying targets and destroying them without human intervention in the Convention on the Use of Conventional Weapons.
The petition to ban killer robots has been signed by some 1,300 scientists and professionals, including entrepreneur Elon Musk, famed physicist Stephen Hawking, Apple co-founder Steve Wozniak, Skype co-founder Jaan Tallinn, and activist philosopher Noam Chomsky. The document was published on the opening day of the International Joint Conference on Artificial Intelligence, the main world forum in this field.
Artificial intelligence technologies have reached a point where the deployment of autonomous systems is practically feasible within a few years, not decades, the letter notes. - And the stakes are very high: they can make a third revolution in military affairs - after gunpowder and nuclear weapons. Kalashnikov assault rifles may become autonomous weapons tomorrow.
Unlike nuclear weapons, robotic systems do not require expensive or hard-to-find raw materials. In theory, almost any country can create them. Therefore, scientists believe that if these types of weapons are not banned in time, they will spread very quickly, fall into the hands of terrorists and lead to destabilization in the world.
Nevertheless, artificial intelligence-based killing machines continue to be developed. Technological prototypes already exist. For example, an autonomous or semi-autonomous mode is incorporated in the South Korean Samsung SGR-A1 guard gun, the British BAE Systems Taranis combat drone, and the American Sea Hunter unmanned anti-submarine ship.
Superintelligence control. For the first time ethical rules for robots were formulated in 1942 by the American science fiction writer Isaac Asimov. He called them the three laws of robotics. First, a robot cannot harm a person or, through inaction, allow harm to be caused. The second is that the robot must obey all the orders of a person, except when they contradict the first law. And the third: the robot must take care of its safety to the extent that it does not contradict the first or second laws.
It was immediately clear that these laws would remain in the realm of fantasy. And a vivid confirmation of this is autonomous weapons systems. After all, their main goal is to harm a person. In addition, Asimov was referring to "ordinary", hard-coded artificial intelligence, such as that of anthropomorphic robots or driverless cars. And the supermind is capable of learning. Any superintelligent system launches a program of analytical processing of information, which implies the formation of a "new experience" in the machine and its application in future actions.
A study by mathematicians and computer scientists from five countries showed that even theoretically, systems for automatic control and correction of decisions made by a computer superintelligence are impossible. The machine learning algorithm embedded in it is based on the principles of neural networks similar to those in the human brain, and the ability to find patterns in large amounts of data is superior to that of humans.
Scientists warn that sooner or later, superintelligence, analyzing the patterns of human actions, will learn to copy them, and then bypass them. Just as we, by observing the habits of a chimpanzee, will eventually be able to outsmart him. Artificial intelligence is already beating people in chess, checkers and other intellectual games, creating works of art based on the analysis of books read and viewed pictures. And this is just the beginning.
You must be logged in to post a comment.