Shulou(Shulou.com)06/02 Report--

People always have more or less stereotype, such as nationality, skin color, education, gender and so on. In people's minds, local groups are often platitudes. For example, women should be gentle and obedient, engineers and technicians are mostly male, whites have higher intelligence than blacks, and so on. Everyone is an individual. It is normal for people to have different attitudes and opinions towards others. Unexpectedly, artificial intelligence also has a "stereotype".

Recently, an interesting new study has been conducted by Facebook's artificial intelligence laboratory. They found that artificial intelligence bias exists not only within countries, but also between countries.

The researchers tested five popular object recognition algorithms, including Microsoft Azure, Clarifai, Google Cloud Vision, Amazon Rekognition and IBM Watson, to understand how each program works. The data sets used for the test come from households in different countries, including the poor and the poor, including 117 categories, from shoes to soap to sofas.

In this study, the researchers found that when asked to identify low-income families, the target recognition algorithm was 10% less accurate than high-income families. As the monthly income increases, so does the error. Especially for Google, Microsoft and Amazon, the accuracy of the low-income home identification algorithm is very low.

Compared with Somalia and Burkina Faso, the accuracy of this algorithm in identifying American products is improved by 15-20%.

There are many reasons for this phenomenon. The most common reason is that when algorithms are created, they are often influenced by the lives and backgrounds of the relevant engineers. Most of these people come from the elite of the high-income class. Their ideas and concepts are based on their ideas and concepts. Artificial intelligence algorithms are also based on their ideas and concepts.

Artificial intelligence has powerful computing power. On the basis of this algorithm, it is easy to obtain massive data and identification information. However, it is worrying that its prejudice is biased not only in terms of targeting, but also in terms of racial and gender discrimination.

The researchers tested the face recognition capabilities of Microsoft, IBM and Facebook. They found that white men had more than 99% accuracy, while black women had an error rate of 47%.

Amazon has developed an artificial intelligence program to block resume selection, which also shows discrimination against women. After choosing resumes, there are more men than women. Resumes with female roles will be given less weight, and even graduate scores at the two women's colleges will be lowered. It can be seen that artificial intelligence also has self-choice and indelible discrimination. On the other hand, artificial intelligence seems to be extremely equal. For example, the recently launched artificial intelligence robot, it can completely achieve that everyone's attitude is equal. It won't change your attitude towards you because you have money and your sex. He will ask and answer questions as gently as a person and accept criticism in a timely manner. Is the phone robot easy to use? Many companies are scrambling to answer this question.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.