Start a Subspace

"Don't be too impatient, we haven't even fully grasped the artificial intelligence technology of Sibadi civilization." Facing Zhang Ran's question, Lin Qiuyue replied lightly.

"The network topology images sent by advanced civilizations are mainly based on the connectionist neural network system. This is actually a black-box-level technology, and we don't really understand why this thing works."

"This may be a big threshold..."

"Connectionism?"

After Lin Qiuyue's detailed explanation, Zhang Ran finally figured out the current development of artificial intelligence.

There are three major schools of artificial intelligence today: the first is symbolism; the second is connectionism; and the third is behaviorism.

"Three genres, the artificial intelligence developed is completely different."

Lin Qiuyue turned on the laptop, and the glossy screen reflected a girlish face. She has always had a crisp and neat style.

Zhang Ran sat on the other side of the dining table, quietly listening to her explanation.

"The idea of ​​symbolism is to advocate the use of axioms and logical systems to build an artificial intelligence system through the assumption of physical symbol systems and the principle of limited rationality."

"Simply put, artificial intelligence is written in programming languages ​​such as if...else, while...do."

"This way I can roughly understand." Zhang Ran ordered two cups of coffee.

This point is easy to understand. At present, most robotic technologies, such as intelligently controlled machine tools, computer operating systems, and car control systems, are all products of symbolism.

But it is obvious that the intelligence developed based on symbolism is almost close to the human level and can reach the upper bound, but it still does not reach the expected level.

"Simply put, it can't handle content beyond the library, and it's quite far from the goal of self-learning and self-evolution."

"And so came connectionism and behaviorism."

The main principle of connectionism is the neural network and the connection mechanism and learning algorithm between neural networks, which is currently a research direction of major laboratories.

This idea advocates imitating human neurons and using the connection mechanism of neural networks to realize artificial intelligence.

The most typical connectionist artificial intelligence is the Go software "AlphaGo" developed with deep learning algorithms!

This artificial intelligence has a relatively strong learning ability,

"For example, a deep learning algorithm...is a neural network composed of many layers of neurons to achieve the function of machine learning."

"Relying on the various layers of perceptrons provided by the neural network, a deep learner that can be implemented in software in terms of function realizes the mastery of induction in machine learning."

"Because of the whole process, the machine needs to be done by itself, and there will be a lot of incomprehensible black boxes in the middle."

Lin Qiuyue explained: "The chess-playing software 'AlphaGo' is trained by playing games against itself. Due to the high complexity, humans can no longer figure out how it calculates."

"I almost understand." Zhang Ran took a sip of coffee and nodded.

"No, you don't understand." There was a smile in her eyes: "Have you heard of the artificial intelligence that was taught badly during the Earth period, its name is Tay, the set gender is female, a chat robot."

Zhang Ran shook his head: "How to say?"

"When she first debuted, Tay was very polite. She was lively and talkative. She would observe the nickname, gender, hobbies, and relationship status of the chat partner. She would play games with you, tell you jokes, comment on your selfies, etc. She will also learn popular languages ​​on the Internet. Connectionist artificial intelligence will learn."

"But soon, under the guidance of netizens, Tay began to utter insults, swear words, and even racist and pornographic content. After all, there are a lot of related content on the Internet."

"Her developers, who don't know how to modify her, can only force it to close. This is the black box and uncertainty brought by connectionism."

"Is that so..." Zhang Ran took another sip of coffee.

Lin Qiuyue said: "However, this is only the first step of the black box, and there is a bigger black box!"

"Have you ever wondered what would happen if two learning chatbots chatted with themselves and learned by themselves?"

"what happens?"

"They talk hundreds of millions of times a second, and after learning from each other, there will soon be inexplicable garbled characters that we cannot understand."

"But for these artificial intelligences, it seems to be understandable... They have formed their own language system in just a few hours, and humans cannot understand these languages, and can only think of it as a series of garbled characters."

"We have no other way than to shut them down. That's the risk of connectionism, there are too many black boxes in there."

"Whether these newly invented languages ​​really make sense, we don't know, only the machines automatically."

Zhang Ran was stunned for a moment, goosebumps all over his body when she said that the robot actually learned itself and developed its own language, which sounds a bit weird.

"You mean, there will be an intelligent mutiny?"

"There is such a possibility... But for now, caring about intelligence betrayal is tantamount to worrying about it, and the learning level can be roughly expressed as nine levels." She snapped out one by one slender fingers, "First, imitate words and deeds. Second , know how to use knowledge. Thirdly, associate. Fourthly, analyze and refine abstraction. Fifthly, pursue laws and verify true knowledge. Sixthly, accumulate true knowledge and form a knowledge system. Seventhly, guide practical practice. Eighthly, foresee Trends. Ninth, Communication Theory."

"Only by completing these nine points, is the real strong artificial intelligence."

"The wisdom of orangutans has completed the first three points."

"Our current level is still pursuing the first level."

"Well... what about behaviorism?"

The last school is behaviorism, which believes that "intelligence does not require knowledge, representation, and reasoning." As long as it interacts in the real environment and provides continuous feedback, intelligent behavior will gradually evolve.

This is also a bottom-up process. Typical representatives include evolutionary algorithms, multi-agents, and ant colony algorithms.

The artificial intelligence of the Sibadi people is also largely based on "behaviorism".

This type of artificial intelligence can be regarded as an understandable category, and there will not be a large number of black boxes like neural networks.

Lin Qiuyue finally sighed: "Let's put it this way, the first genre, symbolism, has already encountered a performance wall, at least our human intelligence is not good enough. Relying on the library alone, even weak artificial intelligence cannot be developed..."

"The second genre, connectionism based on neural networks, will have a large number of black boxes and develop strong intelligence that we don't understand, but its potential is the highest. Advanced civilizations also give extremely high evaluations, but the theory There are actually quite a few difficulties at the level.”

"The third genre, behaviorism, is currently a relatively feasible path. It may be possible to develop weak artificial intelligence no higher than humans. Sibadi civilization also has similar robots, and they will not lose control. Fully automatic universal factory , was developed based on this idea.”

"The replacement of the universal factory requires human beings to do it themselves."

"Or... the relationship between the three doctrines is not so much conflicting and exclusive as it is a snake-ring pattern of mutual benefit. Through such symbiosis and comprehensive execution, it is upgraded in behavior training time after time. ?”

Zhang Ran also sighed, and drank the last sip of coffee: "How do you feel, the mission of the earth is to create human beings; and the mission of our human beings is to create stronger artificial intelligence to replace ourselves?"

But no matter what, artificial intelligence technology must continue to develop, because the core of the metaverse era is a fully automatic production system based on the Internet of Everything. Just pursuing the entertainment performance of the virtual world is just picking up sesame seeds and losing watermelons.

...