Why Is Artificial Intelligence a Big Lie

In fact, the real connotation is "machine learning". So, really powerful, everyone should be excited about this is not so-called artificial intelligence.

Why Is Artificial Intelligence a Big Lie

Big Data Abstracts

Source: predictive analysis sworld

Compiler: Li Xinyu, Zhu Shuai, Jisu, Jiang Baoshan

Artificial intelligence is a big lie.

Or, to put it mildly, it's a confusing concept. To put it more seriously, it is a popular language used to deceive the public, and its popularity is very high.

In fact, the real connotation is "machine learning". So, really powerful, everyone should be excited about this is not so-called artificial intelligence.

On the other hand, AI does provide excellent material for some interesting passages. So, put on your hat of questioning, it's time for an AI version of the "journey to uncover secrets".

Why Is Artificial Intelligence a Big Lie

Schwarzenegger's two iconic roles are kindergarten policeman and terminator.

Before reading this article, three points need to be made clear:

1. Unlike AI, machine learning is completely reasonable. Although exciting progress has been made. However, these advances are almost entirely due to supervised machine learning, which can only solve the problem of computer learning with many labeled or instance data. This form limits machine learning to a very limited range.

2. AI is nothing. AI is just a brand, a strong but empty commitment. The concept of "wisdom" is entirely subjective and essentially comes from human beings themselves.

Those who support AI have infinite possibilities, including Bill Gates and Eromask, have the same assumption: the development of AI is a one-way street. Technological advances, they say, drive us along this one-way street until computers reach human-level intelligence. In fact, even so, the key point of this one-way street is "tagged data". We are advancing rapidly, but moving in different directions, the only thing we can be sure of is that progress can only be made within a very special and limited capacity.

3. AI won't kill you. The robot disaster described in science fiction is actually a ghost story. The idea that machines will upgrade and eliminate human nature according to their own will is of no value.

Victory Neural Network

In the movie Terminator 2: Judgment Day, the robot said, "My CPU is a neural network processor, a computer that can learn." In fact, the neural network referred to by robots is a machine learning method. More precisely, it is a method of describing complex mathematical formulas.

Why Is Artificial Intelligence a Big Lie

Neural network can really accomplish a lot of amazing things, and it can be used for image recognition. Through machine learning, computers can basically complete the work independently by programming. In itself, it has been able to determine the details of the pattern or visual features to be sought. The ability of machine learning to achieve these goals is awesome!

The latest improvement of neural networks is called in-depth learning. It greatly improves the accuracy of object recognition. Deep learning makes learning deeper by adding more layers of neural networks.

Why Is Artificial Intelligence a Big Lie

Architecture of Simple Neural Networks with Four Layers

Machine Learning Needs Marking Data

So, as machines perform better and better in humane tasks, does that mean they are becoming smarter and smarter?

No It can really and excellently accomplish certain tasks, but only if the correct data is available for learning. For the target recognition discussed above, this can be achieved by learning from a large number of labeled photos. Because the target has been correctly marked.

In-depth learning requires these labeled data. This is called supervised machine learning: when there is pre-marked training data, the learning process is guided or "supervised" by the marked data.

It constantly adjusts the neural network to make it better on these data, and then improves one by one. This is the learning process. The only way to improve or "learn" a neural network is by testing the labeled data. Without labeled data, the neural network can not know whether the improvement is correct or not, nor what improvements should be adhered to in this process.

Why Is Artificial Intelligence a Big Lie

Watson in The Edge of Danger

This is another example. In 2011, IBM's Watson beat two human Championships in the TV quiz show Dangerous Edge.

To be precise, computers can't really understand problems, but convert each problem into text input. However, considering that the problems in Dangerous Edge are based on human habits, using complex and ingenious language and covering a variety of dialogue scenarios, Watson can answer one question after another at a stretch, and has to admit that it is a very good "intelligent-like" object.

However, it should be noted that Watson is so smart because he learns beforehand from a large number of labeled sample questions: 25,000 questions collected from TV Q&A programs over the past years, and each question gives the correct answer.

In fact, the core technique is to translate each problem into a "yes/no" prediction. For example, "Is so-and-so the right answer to this question?

Please read the Chinese version for details.