The second climax of artificial intelligence began in the 1980s.

Artificial intelligence was first proposed at the Dartmouth Conference in 1956. The conference determined that the goal of artificial intelligence is to "achieve machines that can use knowledge to solve problems like humans." Although this dream was quickly shattered by a series of unsuccessful attempts, it started a long and tortuous research process for artificial intelligence.

The first climax of artificial intelligence began in the 1950s. In terms of algorithms, the perceptron mathematical model is proposed to simulate the human neuron response process, and can use the gradient descent method to automatically learn from training samples to complete the classification task. In addition, due to the development of computer applications, some attempts to use computers to implement logical reasoning have been successful.

Theoretical and practical effects brought about the first wave of neural networks. However, the flaw of the perceptron model was later discovered, that is, it can only handle linear classification problems in nature, and even the simplest XOR problem cannot be correctly classified. Many application problems have not been solved over time, and research on neural networks has also stalled.

The second climax of artificial intelligence began in the 1980s. The BP (Back Propagation) algorithm was proposed for parameter calculation of multi-layer neural networks to solve nonlinear classification and learning problems. In addition, expert systems for specific fields have also been successfully applied commercially, and artificial intelligence has ushered in another round of climax.

However, the design of artificial neural networks has always lacked corresponding strict mathematical theoretical support. Later, the BP algorithm was pointed out that there is a vanishing gradient problem, so it cannot effectively learn the front layer. Expert systems also expose problems such as narrow application fields and difficulty in acquiring knowledge. Artificial intelligence research has entered its second trough.

The third climax of artificial intelligence began in the 2010s. The emergence of deep learning has attracted widespread attention. The vanishing gradient problem in the learning process of multi-layer neural networks has been effectively suppressed, and the deep structure of the network can also automatically extract and represent complex features.

Avoid the problems of manual feature extraction in traditional methods. Deep learning has been applied to speech recognition and image recognition, and has achieved very good results. Artificial intelligence has entered its third development climax in the era of big data.