The current artificial intelligence is still computer-centered, and has not achieved the "human-centered" cognition that people hope. How to introduce the human cognitive model into artificial intelligence, so that it can reach the level of human-like intelligence in terms of reasoning, decision-making, memory, etc., is a hot topic, difficulty and focus in the current scientific community.
The core of cognition is intelligence and insight, and the core of intelligence and insight is psychology, and the core of artificial intelligence is mathematics. ! A pure machine, whether it is learning or intelligence, has no emotions, and human rationality is similar to a machine on the surface. In fact, this rationality is built on the bottom layer of emotional will, and it is a kind of rationality that integrates knowledge, emotion and will. For example, many memories of people will become fast and good once they involve "I", and this proximity intelligence triggering mechanism contains emotionality-reasonableness. Since the intelligence of human-computer fusion is the empathy of psychology +As a friend put it: Unless someone shows us with conclusive evidence how to introduce mental consciousness into an artificial system, no matter how similar the observable behavior of the system is to human behavior, according to the principle of delocalization, we cannot assume that The system really has spiritual consciousness. Without spiritual consciousness, no matter how powerful calculations are, it will not be able to generate cognition and insight. In addition, love is a unique interface for human beings, which can infinitely expand from the inside out to the intelligent interface that interacts with the outside world. This is also an interface that cannot be produced by machines.
The things in the world themselves cannot be defined and explained , and can only be defined and explained by other things. However, these other things have inconsistencies in themselves, with similarities and differences, so metaphors , analogies are not exact explanations, but approximate explanations, and it is because of the existence of these approximations that the various concepts, ideas, customs, common sense, representations, communication and exchanges that can be used are constituted. Precisely, the source of current computation, mathematics itself, is approximate, such as those axioms, assumptions, conditions, constraints, boundaries, regulations, etc. However, many scholars or artificial intelligence scholars now ignore these mathematical approximations, deliberately regard mathematics as the embodiment of precision, objectivity, and absoluteness, and use formulas, equations, paradigms, reasoning, Computing to complete impossible technical tasks, and then there is a paradox phenomenon that the higher the degree of automation and intelligence, the heavier the cognitive / psychological load of people. The reason is that the imperfect illusion of finiteness is artificially turned into perfection. of infinite reasons!
People gradually form themselves in the interaction with people and things, including relatives, voices, things, pen and paper, ……. Interaction and integration are the source of intelligence and tools that help us think. From language to mobile phones, they may be part of cognition itself.On the one hand, our cognition is always merging with the world; on the other hand, misused calculations may also affect our cognition. Richard Hamming, the 1968 Turing Award winner , once recognized in one sentence: " The purpose of computing is not data, but insight into things. "
On the surface, AI seems to have a more efficient advantage than humans in search, computing, storage and optimization, but it is not. For example, when one or more targets appear, it will be difficult for you to immediately combine the environment to form correct or effective situational awareness. Only when the situation evolves into an appropriate time, space and degree, can people form a good level of situational awareness. According to this, we might as well divide the cognitive mechanism of situational awareness into pre-start period, development period, realization period, depth period, decline period, and end period ... During this period, the degree of concentration of attention is the main factor for regulating different periods of situational awareness. means. Situational awareness, this kind of cognitive behavior is generally composed of two parts, one is the inorganic part, that is, the formal processing of symbols; the other is the organic part, which involves the intentional analysis of understanding, interpretation, thinking and other aspects of the mind. The inorganic part can be optimized computationally, while the organic part is ideally processed with cognition, such as the following situations, which are difficult to characterize with computation but relatively easy to analyze with cognition: static and dynamic, dynamic and static. , moving to know stillness, feeling stillness to know movement , more state of mind, less state of mind, more state of mind, more sense of less knowledge, less sense of sense, more sense of state, virtual state of reality, state of reality and virtual state, sense of virtuality and reality, sense of reality and knowledge of virtuality. The situation of the situation is the deep situation, the perception of the perception is the depth perception, and the situational awareness of the situational awareness is the deep situation awareness. All human-computer interaction is for everyone to communicate or self-cognition, and the computer is a medium or a tool that makes the interaction between people and others more effective, convenient and comfortable.
Artificial intelligence simulates human thinking, and thinking is basically the psychological activities and processes of people in various interactions. When thinking activities are relatively stable, certain thoughts are formed. Therefore, human cognition in artificial intelligence is more important, essential, and more thorough than computing methods, computing power, and computing data. The source of artificial intelligence is people, not labor. If the current artificial intelligence field is put the cart before the horse, it is a kind of Mechanical intelligence and lazy intelligence are probably not too much!
The relationship between cognition and computation is sometimes abstracted as a descriptive characterization (drawing) relationship or a mapping relationship between facts and symbols , which is actually an aspect of the process of giving meaning to propositional symbols, that is, signification. A propositional sign, until I understand it, is still dead and inanimate to me. Comprehension and signification process are in a sense opposite processes, signification refers to going from facts to ideas, then to propositional symbols; comprehension is from propositional symbols to ideas, and then to facts.
From a philosophical perspective, what is cognition? Perceptual qualities; what is calculation? Rational training. Generally speaking, art is an important means of cultivating and training perceptual qualities, and science and technology are the main ways to develop and extend rational self-cultivation. Most real-world perceptual-rational interactions involve hidden information, which is precisely what most AI R&D ignores. Yoshua Bengio of the Université de Montréal, one of the pioneers of deep learning, wrote in an email: “ There is still a huge discrepancy between the estimation models used by learning and reality, especially when the reality is complex. ” Therefore, the progress of artificial intelligence with mathematical calculation as the core is still a long way to go ... As there is a saying: only calculation can distinguish right from wrong, and cognition has no standard answer.
Instinct is the ability to act without pre-training that can produce a result without prior training. William James, the first president of the American Psychological Society, seems to think that the structural aspects of instinct are modular. Each instinct is independently responsible for some simple behavior, but at the same time they also work together, and the calculation of the machine has so far been far from producing an instinct, and it is impossible to produce an instinct.
Some people think: "At present, the information transfer efficiency between humans and machines is still very low, and it is far from realizing human-machine collaboration and mutual promotion in the true sense. To realize the hybrid intelligence of human-machine collaboration, the first problem that needs to be solved is The interaction between humans and machines." Thinking about it carefully, this is not the main contradiction and core issue of human-machine integration . The bottleneck of human-computer integration is not a simple interaction problem, but the combination of cognition and computing . Edsgar Dijkstra, winner of the Turing Award in 1972 , said: "Program testing intelligence is used to prove wrong. , can never prove helplessness.” Polanyi also asserted: "The acquisition of knowledge, even the acquisition of 'scientific knowledge', requires individual evaluation and evaluation step by step." In the field of physics, the creation of quantum theory has made people's understanding of the relationship between subject and object understanding has undergone fundamental changes. In the quantum world, the scientific subject and the object are no longer as absolutely distinct as in the macroscopic world, but as Bohr said: "We are both actors and audiences." In connection with this, Heisenberg also clearly pointed out that the equation of motion of the probability function includes the influence of the interaction of quantum motion and measuring instruments (in the final analysis, people), and this influence has also become an important factor of uncertainty. What Bohr said about the relationship between the actor and the audience means that there is a process of subject objectification and object subjectification between the subject and the object of scientific understanding. The result of mutual transformation and mutual inclusion of subject and object has what Polanyi calls the relationship of "two-way inhabitation". At the dawn of the intelligent era of human-computer fusion, computing has quietly and actively leaned towards cognition . As Alan Paley ( 1966 Turing Award winner ) said: "Any noun can be turned into a verb". In this regard, John McCarthy also showed a positive agreement: "Like all specialized theories, all science is embodied in common sense. When you try to prove these theories, you return to common sense reasoning, because common sense guides Follow your experiment." From this, we can easily see that common sense in cognition is precisely the essence filtered out by calculation.
Human beings have relatively limited intuition in conventional topology, and it is difficult to establish a specific imagination in high-dimensional situations. The only thing that can be grasped is strict mathematical derivation and calculation plus lively mental abstract cognition. Perhaps only in this way can logical and non-logical spaces coexist and form a joint force to solve the more difficult questions raised by nature, and to deal with those things that we think we understand.
In short, cognition is not calculation, but calculation is a kind of cognition.