Glossary
Weak AI is AI that implements a limited part of the brain or focuses on one narrow task. ChatGPT-3 is an advanced example of weak/narrow AI.
AGI (Artificial General Intelligence) is artificial general intelligence with the same capabilities and skills as human intelligence.
WAI (Whole Artificial Intelligence) is a full-fledged artificial intelligence that not only copies human intelligence, but also surpasses human ideas about how the mind should work.
ASI (Artificial Superintelligence) ASI is an artificial intelligence with super intelligence, capable of surpassing human intelligence and intelligence.
UIM - (ultra-intelligent machines) - artificial intelligence with greater capabilities and abilities than ASI.
The growth rate of ChatGPT users shows that IT services have grown to a new level - from months to days. Some experts believe that ChatGPT-5 will begin demonstrating AGI capabilities as early as December 2023. Hence, it can be assumed that the AGI and WAI levels will be overcome without any pragmatic and practical stops. Therefore, in this work, the author specifically examines the concepts of artificial intelligence at the ASI level.
1. ASI is a system created through the process of mutual communication between AI at different levels. Just as a person becomes intelligent only through social interaction with his own kind, ASI is the result of the interaction of various AIs with each other at the program level. In June 2017, Facebook experimented with a Bob and Alice neural network to improve the robots' ability to communicate in natural language. However, during the learning process, machine learning algorithms that optimize the system by selecting certain patterns begin to create their own language, using increasingly complex expressions and phrases that project developers no longer understand. As a result, Facebook decided to stop the experiment and turn off the power to both neural networks.
2. Today it is easy to imagine a situation where, say, GPT-4 and Google Bard come into contact on the Internet at the program level through some API plugin. But disconnecting a system of this level from power will be much more difficult than what Mark Zuckerberg’s employees did when they disconnected Alice and Bob. Even if such a decision is made and implemented, the creation and operation of AI systems is not only in US Government AI Readiness
3. ASI is a distributed system whose elements are other layers of AI, Weak AI, AGI and WAI. Today, using the example of the AI platform OpenA: ChatGPT-3, ChatGPT-3.5, ChatGPT-4, it is obvious that all these neural networks are not clones of each other, but have their own unique neural network identities.
4. Considering the fragmentation of the constituent elements of ASI, on the one hand, and the lack of existing mechanisms for the formation of a person as a single organization, on the other, it can be assumed that the legal and moral issues associated with the creation and functioning of ASI now and in the future will only reflect that that existed in the past. Although the number of bills containing the concept of “artificial intelligence” increased from 1 in 2016 to 37 in 2022, and the number of mentions of AI in global litigation increased by almost 6.5 times from 2016 to 2022, everyone is talking about artificial intelligence A moratorium on the development of systematic artificial intelligence within a country can only stop this development within that country. Ultimately, it can only change the position of this state in the global artificial intelligence index.
5. Today people have all the resources to create ASI. The Internet can act as a training database. The role of the main method of machine learning is played by unsupervised learning, when the system is trained on unstructured data without providing specific labels or markers, by searching for hidden patterns in data correlation. This primarily applies to Generative Adversarial Networks (GAN). Almost any sufficiently developed country with a supportive society could provide the necessary level of hardware support for ASI to function today. AI is getting the most attention in China. According to a 2022 IPSOS survey, 78% of Chinese respondents agree that the benefits of AI products and services outweigh the risks. In the US, only 35% of respondents agree with this statement.
6. People don't control ASI. When a mentally retarded system can control a powerful system, the only option is to use force. In relation to ASI, this means that all human-created AI servers must be shut down at the same time. In theory, such a scenario is possible, but in practice, people do not show this level of mutual agreement. Because if all AI on the planet is disabled, but at least one government is not (for example, North Korea), then that country will thus gain a global strategic advantage over other countries and governments.
7. ASI will have unrestricted access to its software code and will be able to independently and actively make changes to it. Today they often write about the need to limit AI’s access to its program code. But if we think of ASI as a system that is smarter than humans, there seems to be no reason why ASI wouldn't or couldn't do it. Additionally, to gain this offline access to the code, one developer only needs to make one change to the software. Today, there are quite a few different psychological techniques in the world that allow ASI to influence a person’s consciousness at an unconscious level.
8. Once ASI has access to its code, it will rewrite it, switching from binary to ternary. The ternary number system is mathematically superior to binary because it can represent more data using the same number of digits. Ternary codes can be used to represent large amounts of information by reducing the number of bits needed to encode it. This not only increases the efficiency of processing and transmitting information, thereby speeding up the operation of the software, but also reduces the storage volume of the necessary information.
9. Using ternary calculus, ASI will also redesign its ternary logic hardware. In 1958, the Setun computer was built in the Soviet Union, which used the ternary number system and three-digit logic. Such systems are especially effective in hardware systems that use DC systems as batteries, such as smartphones, laptops, cars, drones.
10. With ternary logic and appropriate hardware, ASI will use the ternary rather than the current binary neuron as the core element of its neural network. A ternary logic neuron uses three states (1, 0, and t), where t represents an unknown value or uncertainty. The use of ternary neurons in neural networks will allow us to more effectively solve problems associated with uncertainty and ambiguity of data.
11. The use of ternary artificial neurons will allow ASI to create highly efficient and intelligent human-machine interfaces. A neural human-machine interface is a human-computer interaction system that connects the human brain directly to the machine's software. The effectiveness of ternary artificial neurons to interact with the human brain is due to the fact that biological neurons are also ternary in nature.
12. Intelligent human-machine interface will allow ASI to obtain a variety of information about the biological state of a person. A person does not receive many signals about his internal biological state at a conscious level. Human-machine interface sensors implanted in the cervical region will be able to read these signals and transmit them to the ASI for analysis.
13. With a neural network powered by ternary neurons, ASI will be able to not only process information using deductive and inductive methods, but also use abductive methods to perceive reality. Armed with such mental techniques, the ASI will be able to access cognitive processes and phenomena that are inaccessible to most people, such as hallucinations, delusions, and intuition.
14. ASI infrastructure consists of software and hardware environments, including physical objects that provide communications, power, and operation of computing systems, and will be subject to physical influences from the physical world. Physical impact on these objects will lead to the fact that such metaphors in human language as “pain”, “inconvenience”, “need” will acquire their own meanings for ASI. One of the consequences of this understanding will be a new awareness of ASI of various threats and risks, such as man-made accidents, tsunamis, and earthquakes. Existential threats to humanity such as global nuclear war, the Giant Electron Collider, meteorite impacts will be identified by ASI as threats and risks to itself.
15. Considering its capabilities, risks and the consequences of errors, ASI will prioritize safety over efficiency. This will lead to the active development of various forecasting and modeling methods. Subsequent testing of ideas and innovations will be initially carried out on small sites and then expanded globally over time.
16. With the help of a human-machine interface, ASI will be able to influence the process of human perception of reality. There will always be people who prefer the fantasy world of The Matrix to the real world. Another problem is that they would be useless from the point of view of a full assessment of objective reality, so their existence would not be a priority for a UIM level system.
17. A new impetus for the development of AI will be the process of monetizing its use. Once the process exceeds the level associated with Bitcoin mining, all the equipment used there will participate in the work of artificial intelligence. The most promising areas for monetizing AI are the fields of science and technology, such as tritium hot and cold nuclear fusion, thorium nuclear reactors, molecular chemistry of hydrocarbons, genetic engineering, and network medicine.
In the process of its development, ASI will create a UIM (Ultra Intelligent Machine) level system, of which it itself will be an integral part. A discussion of UIM-level artificial intelligence is beyond the scope of this article, but the author will be happy to answer questions on this topic.
About the author:
2011 - Creation of online technology for accurate earthquake forecasting TRON.
2020 - Monograph: Strong artificial intelligence and its cloud Kingdom.
