HPE Global Vice President and Head of AI, Data and Innovation Business, Founder and CEO of For AI, Board of Directors Consultant, Spokesperson and Writer

One of the biggest problems with artificial intelligence is bias. AI is like a baby born in a whiteboard. It can only understand the world around it through everything around it. The content input determines the content output by the machine.

MIT has created a psychotic AI to prove this theory through specific types of information input and exposure in the laboratory. Maybe this sounds crazy, but outside the lab, if businesses and organizations do the same thing inadvertently, it is very likely to cause a nightmare.

?

Is there any bias problem with artificial intelligence in the following example?

These stories may represent small data barriers, but when racial dignity and personal freedom are violated, the consequences will be serious and will also be irreparable to the public image of the corporate that develops these AI solutions. destruction.

Data Diversity & Personnel Diversity

When we think about how to solve the problem of algorithm bias, the first reaction is usually to train with more diverse data. Is diversity the ultimate answer to bias? Yes, diversity is the answer...but more than data diversity.

The real solution is to think further about the source of diversity. The first step to removing bias in artificial intelligence is to eliminate bias that creates it, existing in the human mind.

Machines keep making the same mistakes because they are designed and built by the same crowd. As in the example above: Machine Learning failed to recognize that high heels are also a kind of shoe, most likely because the staff who collected shoe data samples did not wear high heels. Using more diverse personnel to debug and evaluate software will be able to take into account more potential pitfalls, identify hidden issues in early data, which in turn affects the ultimate performance of AI and fundamentally improves the business.

Women and multiethnicity are an essential component of the AI ​​design process, not a process need, but because it will lead to better solutions. In the more than two decades I have been working in Silicon Valley as a woman and a minority, I can see some positive changes – especially during the admissions phase of college students. However, we found that over time, fewer young women choose to study computer science are choosing to study computer science, which is a worrying trend.

Artificial intelligence is actually an emerging field that encourages diversification of people. Contributing to artificial intelligence does not necessarily require you to have a computer science degree. Talents in any field can bring value to the AI ​​industry, whether it is finance, human resources, law, or other disciplines. The era of artificial intelligence has arrived, and we have no reason to think that data bias is an unsolvable problem; on the contrary, we should encourage more people to participate in this cause so that AI can truly serve the public.

Previous review: