Why is AI always difficult to implement?

Why is it always difficult to implement AI? Why is artificial intelligence often criticized? Some people say that this is caused by science fiction movies, science fiction novels, video games, news media, etc. This view has certain reasonable elements, but there is another more important fact that everyone ignores, that is, what is supposed to be the "human-machine environment" System fusion intelligence" is often mistaken for "artificial intelligence (or even some algorithms)".

Although life and machines can both be used as carriers of cognition, the nature of cognition is different. One is the cognition of life, and the other is the cognition of machines, which is the cognition of specific people to specific things. Human-machine intelligence focuses on direction and risk, while ergonomics focuses on process and efficiency. Computing - The construction of computing mechanisms is the key to breakthroughs in human-computer hybrid intelligence. The key to swarm intelligence lies in the construction of coordinated logic for three or more bodies, and the construction of three-body logic has exceeded the scope of formal computational logic, and a formal computational logic system needs to be established.

Whether it is complex or simple, whether it is an automated product or an intelligent system, anything that is down-to-earth and accepted by everyone, if you think about it carefully, it is better in terms of safety, efficiency, and comfort. To have these advantages, most of the human, machine, and environmental systems are relatively harmonious, at least not a simple AI in a certain field or an intelligent algorithm in a certain field. Some time ago, I wrote an article "The essence of intelligence does not seem to be data algorithm computing power and knowledge", emphasizing that the mechanism that generates these is the source of living intelligence. This time I will also talk about the " "Talents" may still be "robots" without "souls". The reason is still caused by dry "algorithms." There is no melon, there is a (kind of) brain but no mind, there is form but no intention, there is no eyeball...it can only spin around in the circle of possibilities, but cannot try to explore the impossible world. Even if there is some exploration, it is only within the scope of family resemblance. Bounce on the riverbed while being far from clueless about actual non-family resemblances. In addition to human-computer environment system interaction, the second aspect is the understanding and digestion of deep situational awareness. For example, in many situations, we only know the registration and correction between time and space, but do not understand the coordination between situation, potential, sense and knowledge. Accuracy and correction; only know the distortion solution of non-coordinated distance, but forget the fuzzy expansion of collaborative distance; only know frequency and variables, and do not think about abnormality, changing potential, changing sense, changing knowledge and flexibility; only know the data chain and information Chain, do not consider the fact chain and value chain, or even the human-machine environment system chain formed by the entangled superposition of state chain, potential chain, sense chain and knowledge chain; only know the single modulation of homogeneous, uniform and sequential situation awareness, and It ignores the more important multi-level arrays of heterogeneous, non-uniform and random situation awareness, as well as the fast mobility of sensing first and knowing later and the accurate flexibility of sensing first and then knowing, as well as the autocorrelation between situation, potential, sense and knowledge. Cross-correlated transformation probabilities; only know the model of the human model, not the model of the machine; only know the simulation verification structure, and do not pay attention to the performance obtained in actual combat. Reason Three: When something happens, we will consciously or unconsciously associate it with things that have just happened or have a deep impression around us from time to time, and establish our own personalized "causal relationship" situation map (not just a map). Indeed, Relevant ones are called objective factual connections, specious ones are called possibility connections, and irrelevant ones are called subjective intentional connections... These connections that often occur in life are all components of intelligent cognition. Part of it, the objective factual correlation part that can be programmed is often called AI, but the possibility correlation and subjective intentionality correlation are filtered out, and these two are important components of the flexibility of personalized intelligence. In short, we want to use AI algorithms to simplify the problem of complex systems in the human-machine environment; we only know situational awareness, but do not understand deep situational awareness; we ignore the wormhole connection between wind, horse and cow; these three problems may also be caused by AI It’s always hard to get off the ground!

DARPA’s drone vs. manned drone-“dog fight” test has just ended. After the excitement, judging from the review after the test, the key to AI’s victory lies in its strong aggressiveness and shooting accuracy. , but the problem mainly lies in errors in judgment. According to US military testers, the AI ​​system under test often made mistakes in basic fighter maneuvers. More than once, the AI ​​​​stirred the aircraft in the direction it thought the human opponent's aircraft would go, but it was repeatedly proven to misjudge the human. A pilot's thoughts. This is not difficult to understand. Human pilots often make mistakes when judging the opponent's intentions. What's more, what the AI ​​system lacks is the ability to understand creative tactics. It is not surprising that such mistakes occur. However, due to its "excellent targeting capabilities" and ability to track opponent aircraft, the AI ​​was still able to maintain an overall advantage over human pilots, and the computer system ultimately gained the upper hand in the entire confrontation.

In short, drone AI has an advantage in the accuracy of "status" and the speed of "feeling", but it does not yet have advantages in the judgment of "potential" and the prediction of "knowledge". It is recommended that future manned aircraft pilots work harder on fake moves (like Jordan, Kobe, and James) and breaking rules (like Sun Tzu, Zhuge Liang, and Su Yu)! Without rules, all algorithms and (mathematical) models will lose their boundaries, conditions and constraints, and all calculations will no longer be accurate and reliable. When probability formulas change from calculations to fortune telling, the advantages of machines may not be as good as those of humans. Bar? !

Humans make value-based decisions—discussing big rights and wrongs rather than just calculating gains and losses; machines make fact-based decisions—discussing the addition and subtraction of gains and losses, not right and wrong. The relationships between state and potential, as well as between sense and knowledge, are all quantitative and qualitative relationships. The "potential" among them is the maximum possibility within a certain period of time. Everything that is in "potential" is first in "state"; everything that is in "knowledge" is first in "feeling". It can be said that a single spark can start a prairie fire. If the goal is clear, in a game with a large system composed of control units and equipment, the opponent should be or can only be the corresponding system, not the person who operates the equipment, or the person who designs and controls the system. In this regard, we have great weaknesses. The key is that dynamic changes in long-, medium- and short-term goals in a development environment can cause unclear or even vague goals.

Today’s artificial intelligence is like a high-speed train, which is very fast, but requires a track. Real intelligence should be like an airplane, as long as it can reach the destination, it does not require a specific track or route. Errors in situation awareness are divided into errors in posture, situation, perception, and knowledge, and can also be divided into factual/value errors. The application of artificial intelligence in weapons is mainly reflected in machine-to-machine tasking and real-time re-aiming of weapons. This prioritization of effects on typical "service providers" will be performed at the tactical level and depends on the capabilities of intelligent machines. Whether to digest and analyze data from across the battlefield. In fact, the focus and difficulty of the future manned-unmanned confrontation will be the mixed/integrated sequencing of factual and valuable data, information, knowledge, responsibilities, intentions, and emotions in the distribution of human-machine functions!

The above is the detailed content of Why is AI always difficult to implement?. For more information, please follow other related articles on the PHP Chinese website!