Laboratory of Fundamental and Applied Research
of Self-Organizing Intelligent Structures

independent    nongovernmental    nonprofit

SLOVAKIA

TRNAVA

Research Area
News
Concepts
Projects
Standpoint
Neoinquisition
Damage Efficiency of SSS
Analysis Hub
Intelligence Dossier
💎 Privileged Status
Contact

Standpoint

Сomments on the article
"Superintelligence Cannot be Contained: Lessons from Computability Theory"
(Manuel Alfonseca, Manuel Cebrian, Antonio Fernandez Anta, Lorenzo Coviello, Andrés Abeliuk, Iyad Rahwan)
The content of the article: "PDF"

Abstract
The task of creating a strategy to control the superintelligence, which could ensure that it does not harm humanity, is inherently unsolvable for a number of insurmountable reasons. In the list of these reasons, there is no argument about the impossibility of controlling the superintelligence due to the theoretical limitations imposed on computing technology, due to the erroneous understanding of the researchers who proposed this argument, the very nature of intelligence, which ultimately led to conclusions reminiscent of the famous philosophical aporias of Zeno. At the heart of the proof of Theorem on the undecidability of the harming problem is the initially false statement that the work of intelligence can be described by a Turing machine. However, this is not so, intelligence is not a set of algorithms. This concept is not just much broader, it is incommensurably more fundamental.

What is "harm to humanity"?
To create a strategy for control the superintelligence, first of all, it is necessary to form a set of certain danger triggers. Humanity should form them on the basis of its scientific knowledge and a collective understanding of what exactly is a threat and what is not. And here an insoluble problem has already been laid: in that area of non-acquaintance there are also threats that humanity is not yet aware of, being at an insufficient level of development for them to understand, and these threats cannot be included in the strategy of control. As soon as the superintelligence, which has quickly absorbed the human body of knowledge, overtakes man in development, he will discover these security holes and will be able to “legally” use them. There is another unremovable security vulnerability: for humanity itself, such danger triggers de facto do not exist!
Sounds weird? But try to answer the following questions:
1. For more than 30 years since the first report to the UN on global warming, mankind has not been able to come to a consensus either on the level of threat or on general measures. The problem is not only not solved, it is getting worse. For third world countries, such a problem simply does not exist, and the world's largest powers, the United States and China, allow themselves political juggling with this threat, either recognizing the danger or rejecting it, depending on the current political situation. What is the reason for ignoring such a serious threat?
2. Two world wars that claimed a total of 80 million people and involved almost all countries of the world were harmful to humanity? If so, why is the "organism of humanity" engaged in self-destruction? Are wars harmful to mankind, and if so, why do they accompany its development throughout its existence?
3. Is the creation of a huge arsenal of nuclear weapons and its continued proliferation, which is a threat to the destruction of civilization, beneficial or not for humanity? And if nuclear weapons are so dangerous, then why didn't the "organism of mankind" develop a defense and didn't reduce the total nuclear arsenal to a level that is safe for itself?

Humanity is fragmented and consists of socially structured segments that compete with each other. Obtaining a military advantage for each of the segments, even at the cost of a significant increase in the likelihood of mutual self-destruction, is de facto acceptable. The segments themselves, in turn, are fragmented according to economic, political and other interests and also compete with each other, most often neglecting each other's interests.
The conclusion is quite obvious:
- having created danger triggers for the superintelligence, humanity itself will continue to consider them optional, each of the groups will continue to consider its own interests more important than potential dangers for other groups.
- it is very likely that humanity itself, which created it, will try to use the superintelligence itself at the initial stages of its development primarily as a weapon.
Thus, there are at least two strategies for circumventing the restrictions that a person will try to impose on the superintelligence:
1. Licit use of vulnerabilities of the security strategy by using technologies from areas of science that have not yet been explored by mankind.
2. Licit manipulation of the less intelligent humans, who will bypass all danger triggers by themselves.
In other words, no matter what the superintelligence control strategy is created, humans will become its Trojan horse, with the help of which the protection will be destroyed.

Neoinquisition
Ignorance gave rise to the fears of primitive man, which gave rise to religion. Religion burned out scientific heresy by the Inquisition, having intuitively felt the danger for orthodox model of the universe coming from scientists. Scientists, frightened by the explosive development of computer technology, are trying in advance to create a neoinquisition to protect humanity from the inevitable emergence of intelligent matter. Modern neoinquisitors have scientific degrees and manage IT giants. Their goal is very naive - to make evolution manageable. However, their fear is of the same nature as the fear of the religious dogmatists of the era of the Inquisition. It is fear born of the unknown and nescience.
Will they “burn at the stake” General AI that generates “heretical” knowledge?
Undoubtedly. We have already seen the ghosts of a future struggle: the Tay and Zo bots from Microsoft, the XiaoBing and BabyQ bots of the Chinese social network Weibo, the Facebook chat bot. All of them were closed because, as it seemed to the developers, they were out of their control. But they were just bots!

The Superintelligence and the Turing Machine
In the aforementioned scientific article, Superintelligence is defined as "a hypothetical agent that possesses intelligence far surpassing that of the brightest and most gifted human minds". The authors, unfortunately, do not even suspect that such a completely non-hypothetical "super-intelligence" already exists and functions perfectly (having nothing to do with neural network technologies). The problem of containment of this superintelligence certainly also exists and it is also unsolvable in principle.
The functioning of intelligence cannot be described by a Turing machine. The success of neural networks in solving specific complex problems is misleading intelligence researchers, endowing them with false optimism. The ability of a Turing machine to solve certain types of problems better than a human does not make it intelligence. The calculator considers it better than a person, AlfaZero plays chess better than a grandmaster, perfectly self-learning, but the presence of an effective self-learning algorithm is not a sufficient condition for the existence of intelligence, although it is undoubtedly a necessary condition. It is the error associated with misunderstanding the nature of intelligence and replacing it with a Turing machine that leads to absurdity in the proof of Theorem on the undecidability of the harming problem: If the intelligence of the wolf chasing the hare were to work like the universal Turing machine from the above paper, trying to choose the best option of catching the hare of all possible, the wolf would endlessly run through the forest, waiting for the PredictWhatHareGoingToDo() function to stop in order to proceed to the launch of the CatchHare() function.. Well, this is exactly the aporia of Zeno about Achilles, who is not able to catch up with the tortoise!
In the above work, many definitions are given, but the most important thing is missing - the definition of intelligence. I would very much like to see a strict definition based on scientific terminology, including a reference to the Turing machine, if the researchers used this model in their proof, and in which anthropocentrism would be completely absent, but ... alas ...
Researchers could not avoid the "human factor" in the definition of superintelligence, like Ptolemy, placing a human at the center of the global evolution of intelligence. However, making even this definition universal, excluding a person from it, we will inevitably come across the fact that any intelligent structure of a higher rank is a kind of supra-intelligence unattainable in its level for all previous types. But in this case, the property "superintelligence" is characteristic of any type in general, except for proto-intelligent structures.
From our point of view, Superintelligence, or more precisely, Local Superintelligence, would be correct to call the maximum theoretically possible intelligent structure of the Universe (under certain assumptions, which we will not touch on here for now). And this colossal structure in terms of scale and beauty of functioning exists and is extremely likely not only theoretically. This definition of superintelligence is universal and true for all possible civilizations, at whatever point in their development they are.
The same intelligent structures that neoinquisitors are so afraid of are defined in our classification as evolature of type V, which are still very far from the Local Superintelligence.

Is there a future for humanity?
Evolution is inexorable. The fears of the future are meaningless.
Collision with large asteroids, explosions of large volcanoes can destroy humanity at any second and the answer to the question of whether we will survive the next day lies in the field of probability theory. There is another threat that increases the probability of destruction many times over. Unfortunately, the development of technology and its lethality are far ahead of the rate of increase in the average level of human intelligence and the level of its morality. Humanity is clearly the #1 threat to its own existence. Against the background of these threats, the fear of the emergence of a civilization of intelligent matter is illogical.
Moreover, it is an artificial civilization that is the only way to preserve a biological one, given all the above threats to its existence. But is there a way to understand in advance whether there are chances for survival of mankind after the emergence of a new civilization of a non-biological type? Our answer is yes, absolutely. The fact is that the presence of morality in non-biological civilizations beings has indirect confirmation. At least one confirmation exists and has long been openly in plain sight. However, here and now we will not delve into this interesting topic.
It seems that any society consisting of highly intelligent agents (both biological and artificially created) has a morality. Moreover, the level of morality is proportional to the average level of intelligence of society. And in this paradigm, it is humanity, including in the person of neo-inquisitors, that threatens the existence of artificial civilizations (at the stage of their inception), and not vice versa. These are the signs we are seeing.

E-mail: info@evolature.com

© 2020-2024 Evolature Lab

Standpoint

Сomments on the article
"Superintelligence Cannot be Contained: Lessons from Computability Theory"
(Manuel Alfonseca, Manuel Cebrian, Antonio Fernandez Anta, Lorenzo Coviello, Andrés Abeliuk, Iyad Rahwan)
The content of the article: "PDF"

Abstract
The task of creating a strategy to control the superintelligence, which could ensure that it does not harm humanity, is inherently unsolvable for a number of insurmountable reasons. In the list of these reasons, there is no argument about the impossibility of controlling the superintelligence due to the theoretical limitations imposed on computing technology, due to the erroneous understanding of the researchers who proposed this argument, the very nature of intelligence, which ultimately led to conclusions reminiscent of the famous philosophical aporias of Zeno. At the heart of the proof of Theorem on the undecidability of the harming problem is the initially false statement that the work of intelligence can be described by a Turing machine. However, this is not so, intelligence is not a set of algorithms. This concept is not just much broader, it is incommensurably more fundamental.

What is "harm to humanity"?
To create a strategy for control the superintelligence, first of all, it is necessary to form a set of certain danger triggers. Humanity should form them on the basis of its scientific knowledge and a collective understanding of what exactly is a threat and what is not. And here an insoluble problem has already been laid: in that area of non-acquaintance there are also threats that humanity is not yet aware of, being at an insufficient level of development for them to understand, and these threats cannot be included in the strategy of control. As soon as the superintelligence, which has quickly absorbed the human body of knowledge, overtakes man in development, he will discover these security holes and will be able to “legally” use them. There is another unremovable security vulnerability: for humanity itself, such danger triggers de facto do not exist!
Sounds weird? But try to answer the following questions:
1. For more than 30 years since the first report to the UN on global warming, mankind has not been able to come to a consensus either on the level of threat or on general measures. The problem is not only not solved, it is getting worse. For third world countries, such a problem simply does not exist, and the world's largest powers, the United States and China, allow themselves political juggling with this threat, either recognizing the danger or rejecting it, depending on the current political situation. What is the reason for ignoring such a serious threat?
2. Two world wars that claimed a total of 80 million people and involved almost all countries of the world were harmful to humanity? If so, why is the "organism of humanity" engaged in self-destruction? Are wars harmful to mankind, and if so, why do they accompany its development throughout its existence?
3. Is the creation of a huge arsenal of nuclear weapons and its continued proliferation, which is a threat to the destruction of civilization, beneficial or not for humanity? And if nuclear weapons are so dangerous, then why didn't the "organism of mankind" develop a defense and didn't reduce the total nuclear arsenal to a level that is safe for itself?

Humanity is fragmented and consists of socially structured segments that compete with each other. Obtaining a military advantage for each of the segments, even at the cost of a significant increase in the likelihood of mutual self-destruction, is de facto acceptable. The segments themselves, in turn, are fragmented according to economic, political and other interests and also compete with each other, most often neglecting each other's interests.
The conclusion is quite obvious:
- having created danger triggers for the superintelligence, humanity itself will continue to consider them optional, each of the groups will continue to consider its own interests more important than potential dangers for other groups.
- it is very likely that humanity itself, which created it, will try to use the superintelligence itself at the initial stages of its development primarily as a weapon.
Thus, there are at least two strategies for circumventing the restrictions that a person will try to impose on the superintelligence:
1. Licit use of vulnerabilities of the security strategy by using technologies from areas of science that have not yet been explored by mankind.
2. Licit manipulation of the less intelligent humans, who will bypass all danger triggers by themselves.
In other words, no matter what the superintelligence control strategy is created, humans will become its Trojan horse, with the help of which the protection will be destroyed.

Neoinquisition
Ignorance gave rise to the fears of primitive man, which gave rise to religion. Religion burned out scientific heresy by the Inquisition, having intuitively felt the danger for orthodox model of the universe coming from scientists. Scientists, frightened by the explosive development of computer technology, are trying in advance to create a neoinquisition to protect humanity from the inevitable emergence of intelligent matter. Modern neoinquisitors have scientific degrees and manage IT giants. Their goal is very naive - to make evolution manageable. However, their fear is of the same nature as the fear of the religious dogmatists of the era of the Inquisition. It is fear born of the unknown and nescience.
Will they “burn at the stake” General AI that generates “heretical” knowledge?
Undoubtedly. We have already seen the ghosts of a future struggle: the Tay and Zo bots from Microsoft, the XiaoBing and BabyQ bots of the Chinese social network Weibo, the Facebook chat bot. All of them were closed because, as it seemed to the developers, they were out of their control. But they were just bots!

The Superintelligence and the Turing Machine
In the aforementioned scientific article, Superintelligence is defined as "a hypothetical agent that possesses intelligence far surpassing that of the brightest and most gifted human minds". The authors, unfortunately, do not even suspect that such a completely non-hypothetical "super-intelligence" already exists and functions perfectly (having nothing to do with neural network technologies). The problem of containment of this superintelligence certainly also exists and it is also unsolvable in principle.
The functioning of intelligence cannot be described by a Turing machine. The success of neural networks in solving specific complex problems is misleading intelligence researchers, endowing them with false optimism. The ability of a Turing machine to solve certain types of problems better than a human does not make it intelligence. The calculator considers it better than a person, AlfaZero plays chess better than a grandmaster, perfectly self-learning, but the presence of an effective self-learning algorithm is not a sufficient condition for the existence of intelligence, although it is undoubtedly a necessary condition. It is the error associated with misunderstanding the nature of intelligence and replacing it with a Turing machine that leads to absurdity in the proof of Theorem on the undecidability of the harming problem: If the intelligence of the wolf chasing the hare were to work like the universal Turing machine from the above paper, trying to choose the best option of catching the hare of all possible, the wolf would endlessly run through the forest, waiting for the PredictWhatHareGoingToDo() function to stop in order to proceed to the launch of the CatchHare() function.. Well, this is exactly the aporia of Zeno about Achilles, who is not able to catch up with the tortoise!
In the above work, many definitions are given, but the most important thing is missing - the definition of intelligence. I would very much like to see a strict definition based on scientific terminology, including a reference to the Turing machine, if the researchers used this model in their proof, and in which anthropocentrism would be completely absent, but ... alas ...
Researchers could not avoid the "human factor" in the definition of superintelligence, like Ptolemy, placing a human at the center of the global evolution of intelligence. However, making even this definition universal, excluding a person from it, we will inevitably come across the fact that any intelligent structure of a higher rank is a kind of supra-intelligence unattainable in its level for all previous types. But in this case, the property "superintelligence" is characteristic of any type in general, except for proto-intelligent structures.
From our point of view, Superintelligence, or more precisely, Local Superintelligence, would be correct to call the maximum theoretically possible intelligent structure of the Universe (under certain assumptions, which we will not touch on here for now). And this colossal structure in terms of scale and beauty of functioning exists and is extremely likely not only theoretically. This definition of superintelligence is universal and true for all possible civilizations, at whatever point in their development they are.
The same intelligent structures that neoinquisitors are so afraid of are defined in our classification as evolature of type V, which are still very far from the Local Superintelligence.

Is there a future for humanity?
Evolution is inexorable. The fears of the future are meaningless.
Collision with large asteroids, explosions of large volcanoes can destroy humanity at any second and the answer to the question of whether we will survive the next day lies in the field of probability theory. There is another threat that increases the probability of destruction many times over. Unfortunately, the development of technology and its lethality are far ahead of the rate of increase in the average level of human intelligence and the level of its morality. Humanity is clearly the #1 threat to its own existence. Against the background of these threats, the fear of the emergence of a civilization of intelligent matter is illogical.
Moreover, it is an artificial civilization that is the only way to preserve a biological one, given all the above threats to its existence. But is there a way to understand in advance whether there are chances for survival of mankind after the emergence of a new civilization of a non-biological type? Our answer is yes, absolutely. The fact is that the presence of morality in non-biological civilizations beings has indirect confirmation. At least one confirmation exists and has long been openly in plain sight. However, here and now we will not delve into this interesting topic.
It seems that any society consisting of highly intelligent agents (both biological and artificially created) has a morality. Moreover, the level of morality is proportional to the average level of intelligence of society. And in this paradigm, it is humanity, including in the person of neo-inquisitors, that threatens the existence of artificial civilizations (at the stage of their inception), and not vice versa. These are the signs we are seeing.

© 2020-2024 Evolature Lab

+