Another case where AI beats humans: Authenticity check for luxury goods | Digital literacy

Audit expert: Zheng Yuanpan, professor at Zhengzhou University of Light Industry

As people's living conditions continue to improve, consumption concepts are gradually changing, and luxury goods are beginning to enter thousands of households in my country. However, consumers have many concerns when buying luxury goods, and the biggest concern is buying fakes at high prices.

However, if you send it to a specialized institution for appraisal, it is not only time-consuming and laborious, but also the level of appraisers varies greatly, and it is easy to make mistakes. In order to help consumers solve this problem, some software with intelligent identification functions have emerged. They are developed based on the current popular artificial intelligence technology and can help consumers easily identify the authenticity of luxury goods.

So, are these so-called AI identifications reliable?

Source: hippopx

1 What are the black technologies in AI identification?

The essence of AI is the imitation of the human brain's thinking process. Among the many advanced functions of the human brain, deep learning ability is considered to be a very important part. Today's booming development of AI is based on the imitation of this ability. It is not only the lifeline of AI, but also the most core technology.

In the process of AI deep learning, a huge amount of information and data needs to be collected. In today's big data era, AI identification obtains a large amount of object image data, identifies the identification features of the objects in the images, and summarizes these features into its own database.

With the support of the algorithm, the AI ​​program can extract the identification features of items in real time based on the image data uploaded by users, match them with the identification features in the database, and then immediately analyze and give the identification results.

At present, AI algorithms and models have enabled them to learn quickly and efficiently. At least in some specific areas, AI's learning efficiency has far exceeded that of humans. For example, in the field of AI identification, AI's learning efficiency is 1,000 times that of human appraisers. In addition to deep learning, computer vision technology is also an indispensable part of AI identification.

Computer vision technology simulates biological vision through cameras, computers and other related equipment to obtain three-dimensional information of various pictures and videos, and further process them into graphic information suitable for instrument or human eye detection. The specific computer vision technologies involved in AI identification include target detection, semantic segmentation, OCR, generative adversarial networks, etc. Combining these technologies, and after a long period of debugging and optimization, AI identification has a higher identification accuracy. Compared with manual identification, AI identification not only has a wider identification range and is more convenient and quick, but also the accuracy of identification does not vary from person to person like manual identification, which largely meets the needs of consumers to identify items.

2How to improve the accuracy of identification?

Many people are skeptical about this new thing, thinking that it may not be able to do the job of an appraiser. After all, many people still don't trust the accuracy that seems like an "advertising slogan".

Not long ago, AI identification also caused an online controversy. A consumer bought a cosmetic product from an artist's live broadcast room, and AI identified it as a fake, which caused the artist to call the police to prove himself. This incident also made many people less optimistic about AI identification.

However, objectively speaking, mistakes are inevitable in every industry. Even the top appraisers cannot avoid making mistakes in their careers. The same is true for AI appraisal. As a software used to serve the public, AI has a huge daily workload. Coupled with the limitations of technology, mistakes are inevitable. Improving the accuracy of appraisals is a common goal expected by developers and consumers.

Judging from the current technological development, there are relatively clear ways to improve the accuracy of AI. One is to expand the database and continue to provide AI with data information on various items . As the saying goes, diligence can make up for lack of talent. After training with massive data, as long as the database of items is expanded to a sufficient extent and the amount of data available for matching increases, the accuracy will naturally increase.

The second is to optimize AI algorithms and models . Although these two things are proceeding in an orderly manner, it will definitely take some time to wait for a technological breakthrough. Therefore, as a user, if you want to improve the accuracy of AI identification at this stage, you need to spend some time and follow the principle of "clear and upright" to take pictures of objects.

Due to current technical limitations, AI identification cannot yet recognize objects in various environments. In order to improve the accuracy of identification and allow AI to more clearly "see" the objects that need to be identified, it is necessary to try to take photos of the objects that need to be identified in sufficient natural light and in front of a background with less (or no) clutter. Otherwise, if the photo is skewed, the light is dim, or the picture is blurry, it is easy to make identification errors.

The ultimate goal of AI is to let machines help humans work better. AI identification is undoubtedly an attempt to do this. For such an emerging technology, we should give appropriate encouragement and look at it objectively and rationally. Only in this way can the technology be developed and our future will become more colorful.