Machine learning (ML) is a field of computer science that uses data to extract algorithms and learning patterns, and applies “learned” generalizations to new environments, including performing tasks without direct human programming. With more and more use cases in modern data analysis, unlocking common misconceptions about machine learning will help understand how we can take advantage of the powerful potential of machines.
Misunderstanding: Machine learning and artificial intelligence will replace humans
The machine will not take away your work. In fact, machine learning and artificial intelligence (AI) are more likely to help you get the job done better and allow you to focus more on the realization of your own ambitions and the vital human elements of your role-including creativity and strategy.
Although ML and AI are useful in many applications, the performance of these technologies in many fields is not particularly powerful and requires human influence, intervention, or supervision. These include：
- Long-term planning
- Abstract or creative thinking
- Understanding causality
- Make decisions that require domain knowledge or background
Human judgment is also necessary to eliminate the inherent biases in ML algorithms. Even if technology advances, we may never see a perfect algorithmic solution. Interpretability and transparency are very important for people to trust the output and suggestions of the machine; we must understand what is happening in the “black box” of the model in order to fully trust and integrate the decisions affected by the machine in our work and life.
Misunderstanding: Machines learn from experience
Contrary to popular belief, machine learning does not rely on experience, but on data. You can’t expect a problem to be solved as long as you turn on a computer-the machine needs to learn from the data and create algorithms to apply algorithms to future environments, including：
- A method of classifying or depicting the components of a data set
- Indicators for scoring or evaluating success
- Optimize the model parameters of the data
This is achieved by extracting a broad explanation (such as an abstract story) from the data set, because the data set may contain complex patterns or hidden laws that are difficult for humans to recognize. Today, many financial institutions use ML to analyze transaction data to identify and mark violations that may be fraudulent charges, or to assess risks and provide recommendations for lending operations.
In simpler terms, upload a photo of a donut to a machine, and it can determine whether a new photo contains a donut. First, we will upload photos with donuts and photos without donuts separately, and tell the image classifier which one is which. This provides the machine with data, and the machine will use this data to build a model for deterministic prediction, so as to distinguish between photos with “donuts” and photos without “donuts”. Then, we can upload a new image, and the machine will apply its algorithm model and make a decision-is there a donut in this photo?
Misunderstanding: Machine learning and artificial intelligence are the same
Artificial intelligence and machine learning are two different but related concepts. Regarding the relationship between AI and ML, one way of thinking is: the former is a problem, while the latter is a solution that tries to solve this problem. If the ultimate goal is that machines can use the cognitive abilities of (human) intelligence to solve problems, then this process of applying algorithms derived from data to a new and larger environment is a solution to achieve this goal.
In order to help people distinguish between the two, it can be understood that artificial intelligence can solve problems that are easy for humans and difficult for computers, such as computer vision. We continue to extend the example of donuts, and now we introduce a new challenge: teaching computers to distinguish between bagels and donuts. For humans, this is an easy task, but for computers, it is somewhat challenging. At this time, AI refers to the ability of a machine to successfully distinguish between bagels and donuts (that is, to solve a problem), while ML refers to the method by which a computer learns to draw conclusions when it sees a new photo (that is, a solution).
On the contrary, ML performs well and is often used in situations that are simple for machines but difficult for humans, such as performing complex mathematical algorithms or using probabilistic calculations. The computing power of machines helps to quickly perform more challenging tasks, or discover patterns that humans may ignore.
