Nowadays, interpretability is becoming a hurdle for AI. Last year, the European Union issued the "Artificial Intelligence Code of Ethics", which clearly stated that the direction of AI development should be "trustworthy", including security, privacy, transparency, and interpretability.
Artificial intelligence applications aim to output decision-making judgments. Interpretability refers to the degree to which humans can understand the reasons for a decision. The higher the interpretability of the artificial intelligence model, the easier it is for people to understand why certain decisions or predictions are made. Model interpretability refers to the understanding of the internal mechanism of the model and the understanding of the results of the model. Its importance is reflected in: in the modeling stage, it assists developers in understanding the model, making comparisons and selections of models, and optimizing and adjusting the model when necessary; in the putting into operation stage, explaining the internal mechanism of the model to the decision-making party and interpreting the results of the model. For example, the decision-making recommendation model needs to explain: why a certain solution is recommended for this user. At present, the understanding and definition of artificial intelligence in various fields differ depending on the field, but there is consensus in common technologies and basic research. The first stage of artificial intelligence aims to solve problems and carry out logical reasoning through machine theorem proofs, expert systems, etc.; the second stage realizes environmental interaction, which obtains information from the operating environment and exerts influence on the environment; the third stage moves toward cognition And thinking ability, discover new knowledge through data mining system and various algorithms.Strictly speaking, the United States’ artificial intelligence technology is generally the world’s leader, but when it comes to human-computer integration intelligence, it often does not show such a big advantage, and it does not even have a leading position (maybe China and the United States are in the integration of humans and machines. In terms of intelligence, there is no generation gap at all). Investigate the cause, the human problem. For example, in this epidemic, in terms of medical software, hardware, and medical personnel, the United States should be far better than China's current situation. Unfortunately, it corresponds to a sentence in the "Three Body": Weakness and ignorance are not obstacles to survival. Arrogance is! The mistakes and mistakes of the leaders have greatly compromised many advancements, or even disappeared. This can't help but remind people that the report in the US "Arms Control Magazine" a few days ago may be similar! The U.S. Department of Defense applied for US$28.9 billion for the modernization of U.S. nuclear weapons facilities in fiscal 2021, reflecting the Trump administration’s strategic development focus: enhancing the high degree of automation of nuclear command, control, and communications (NC3) infrastructure and improving its speed and accuracy. Sex, but it also raises a disturbing question, that is, in future nuclear wars, what role will AI autonomous systems play in determining the destiny of mankind? The current computer-aided decision-making is still in its infancy, and it is prone to unpredictable failures. Although machine learning algorithms are good at specific tasks such as facial recognition, they also have inherent "bias" conveyed through training data. Therefore, it is necessary to adopt a cautious and responsible attitude in applying artificial intelligence to nuclear weapons accusations. As long as nuclear weapons exist, humans (not machines) must exercise ultimate control over the use of nuclear weapons. At this time, the true ability of man-machine integration of intelligence will be It will be as important as epidemic control.The integration of human and machine intelligence is fundamentally the representative of the combination of science and technology, humanities and art, mathematical symbolic fact language and natural experience value language. Time and space can not only bend in the physical realm, but also can be distorted in intelligence. If philosophical logic has experienced the world’s origin issues and research method issues, then the analysis of philosophy of the last century-the analysis of human language tools has become a "revolution" in human thinking. This is based on Wittgenstein. The symbolic philosophical revolution directly induced the rapid development of artificial intelligence technology represented by the Turing machine and the Turing test. However, Mr. Jin Guantao’s "Philosophy of Authenticity" believes that in the 21st century, analytic philosophy has finally confined philosophy in a cage. In fact, it has also caused the imprisonment of thought: symbols can have their own when they do not refer to objects of experience. The authenticity of, and this conclusion can be established for both mathematical language and natural language. At the same time, the authenticity of pure symbols can be embedded in the authenticity of experience; scientific research and humanistic research can become two areas that are unified but not overlap with each other and have their own authenticity standards. The great progress of mankind is to make the authenticity instinct (the objectivity of common sense) under the control of the system of ultimate care and corresponding value. But today’s two pillars of authenticity are being overturned by scientific progress, and something truly terrifying has happened: human beings are irresistibly reduced to intelligent “animals”—in a world where there is no distinction between true and false. If there is right and wrong, there will be no real sense of morality and dignity of life. Humans not only express the world with symbolic equivalence or contain logical relations. Human education is not equal to learning and knowledge, but to induce desires in a good direction. The computer itself cannot cross the "understanding" gap, only humans can cross the embarrassment of symbol pointing. For the subject, signs and experience are mixed, logic and non-logic are mixed, axioms and non-axes are mixed, data, information, and knowledge are mixed together. This is the main reason why interpretability is difficult. . The fusion of man and machine is how symbols (mathematics) are embedded in the subject's experience (controlled experiment) to different degrees, just as Lao Tzu said in The Tao Te Ching: "Tao produces one, one life two, two produces three, and three produces all things."The so-called AI is to a large extent just the use of the computer's ever-increasing computing power, and the use of this path is destined to be wrong for intelligence. Humans learn and use, and machines are rigid. And human intelligence is the ability to perceive a small sample of situational awareness (how fast it is better to save it). A well-known example of situational awareness is the “Wear, Hear, and Inquiry” in Chinese medicine, which uses the difference between natural language and mathematical language to break the divergence between mind and physics, and unify facts and values.
Wangwenwenqie should originate from the sixty-first difficulty of the "Difficulty", saying: in the scriptures, looking and knowing is the god, hearing and knowing is the sacred, asking and knowing is the work, and knowing the pulse and knowing is the coincidence. What is it? The earliest use of the four-character joint name should be in the "Ancient and Modern Medicine": "Wearing, hearing, asking and cutting four characters, sincere is the program of medicine." Wang is to observe the patient's development, complexion, tongue coating, expression, etc.; to hear the patient's voice , Coughing, wheezing, and smelling the patient’s bad breath, body odor and other odors; asking about the symptoms felt by the patient, previous diseases, etc.; ). The fundamental reason why the interpretability of artificial intelligence is difficult is that it contains not only mathematical language, but also natural language, and even thinking language (so it is impossible to cross this hurdle). The human-computer fusion intelligence can not only suspend the subject, but also change the subject with ease. In the interaction of human, machine, and environmental systems, deep situational awareness can be realized in real time and timely, and the language of mathematics, natural language, and thinking language can be organically completed. The signifier, the signified, and the signification switch among the signifiers can easily go straight to the purpose and realization of the intention.