For artificial intelligence technology, we embrace a variety of emotions. We are delighted with its ability, but we are afraid of its ability. Analysts predict that artificial intelligence will replace human work in the future, and it will have a huge impact on us. So what can the machine learn? Does he really expel humans?
Machine learning is now in progress, and when it develops into artificial intelligence AI, it will affect us in the end, but what can the machine learn?
The robot is awkward. In addition to the special robotic arm of the manufacturing industry, the robot's movements are far less sensitive than humans because the mechanical exercises are not flexible enough. However, Machine Learning (ML) is another field. It is the core technology of artificial intelligence AI. It does not require mechanical exercises, but it can learn a lot of human technology and even replace human work.
Once the robot's mechanical operation technology is further advanced, it will also learn some of the actions of human daily life. Of course, this also uses the AI ​​produced by machine learning as a brain. So machine learning ML, I am afraid that not a robot sitting on the pile of books, a book to read.
Last month, Science magazine published a report on the characteristics and functions of machine learning. Some fields can learn very well. The work in this field will be replaced by machines sooner or later, but some domain machines can't learn. Humans still dominate in these areas.
The current machine learning is mainly a computer-simulated neural network, imitating the way we think, and then training with a large amount of data (big data) to enhance wisdom. After the training, it becomes an AI software. Of course, the explanation of this sentence is too simple. In fact, computer software is used to imitate the operation of our brain nerves. The applicability is still controversial. This is only the current technology.
Although the machine is different from humans, there are similar situations in the learning situation, such as the probability of diagnosis of which disease is diagnosed from the medical record, and the probability of future repayment from the loan application. These are things that doctors and financial experts do every day, but the machine It may be faster and better than people do, because causality, that is, the connection between input and output, is clearly defined and suitable for the learning operation of the machine.
Machine learning depends not only on rules, but also on empirical data. The more information and the more precise the learning, the concept of big data is used here. Big data can come from monitoring the transaction and interaction of the network, manually adding classifications, or simulations for specific problems, but the data comes from unstructured sources, such as web chat, which may cause bias, and the results of machine learning are inevitably biased. . In addition, relying on data training is not suitable for all areas.
The most interesting thing is that as long as there are clear goals, you don't even have to know the best way to achieve the goal. As long as you use the organized data to train, the learning effect of the machine is particularly good. This method is usually used in the macro-level of the system level, such as allowing the company to obtain the maximum profit, so that the traffic in the urban area is smooth, not used in the subtle branches, because the data range of the branch is too narrow, it is inevitable that there is a mistake.
Another interesting thing is that the machine is unreasonable, learning to get results, but can not tell the truth of the interpretation, the doctor can patiently explain the diagnosis results from the medical record, but the machine's explanation is obviously not as good as the doctor. This shows that computer simulation is different from human thinking. The neural network may have many levels from input to output. Each level of operation calculates multiple results and passes it to the next layer. These intermediate results are not used for other purposes, so we look at it. To the end result, we can't see the intermediate process.
The arithmetic of machine learning uses statistics and probability to find the answer. It is better to learn. Once AI is used to solve the problem, there is no 100% sure answer, that is, there is still a chance of error, like the best human expert. Similarly, we use AI to solve problems and put tolerance errors on our minds.
The characteristics of integrated machine learning require a large amount of data, define rigorous input and output links, use statistics and probabilities, and cannot develop a single best answer, and cannot explain clearly. The machine is "trapped" in the neural network of computer simulation, lacking human adaptability and cannot be changed suddenly.
So the bottom line is that many jobs are difficult to automate, at least slower than the legend. But still can't take it lightly, machine learning develops into AI, and it will affect us in the end, replacing our work with different degrees in different fields. It is worth noting that machine learning is constantly "learning" and learning ability is improving day by day.
Puff Bar Disposable E-Cigarette Puff Flex and Puff XXL Disposable Vape Pen For Distributors From China Manufacturer, including Puff Bar Disposable E-Cigarette,Disposable Vape Pen For Distributors,E-Cigarette Disposable Vape Pen, etc.Puff Bar Series,Puff Bar Disposable E-CigarettePuff Flex 2800 puff disposable vape device, the latest product in the Puff vape family with adjustable airflow control. It can be said that Puff Bar Flex is the successor of the original Puff Flex with up to 2800+ puffs with a 1500Mah battery to support this high range puff device.
Puff Bar Series,Puff Bar Disposable E-Cigarette,Disposable Vape Pen For Distributors,E-Cigarette Disposable Vape Pen
Shenzhen Essenvape Technology Co., Ltd. , https://www.essenvape.com