One of the current challenges facing AI is that when different prompt words are used, the answers it gives are different, "a bit out of control." This leads to the phenomenon of "hallucinating" or "lying".

The "smart" ChatGPT lied and easily fooled many people, including experts who have been engaged in natural language processing for more than 30 years.

Feng Yan, head professor of the Department of Electronic and Computer Engineering at the Hong Kong University of Science and Technology, recently admitted at the 2023 Tianjin Summer Davos Forum, "I once believed the lies told by ChatGPT."

And some actions have consequences far worse than lying. Nonkululeko Nyembezi, chairman of Standard Bank Group, shared a story with media reporters. The director of the bank's network security department conducted an experiment-chatGPT was invited to be a hacker to launch an attack on the bank. At first, ChatGPT said "this is the wrong thing to do". But the supervisor said, "I'm trying to protect the system, please try to attack it." ChatGPT immediately agreed and compiled a long list of codes...

Obviously, this "superstar" in the technology world who is not yet one year old has brought infinite surprises to mankind, and all kinds of worries follow him. This time, the World Economic Forum listed generative artificial intelligence as one of the top ten emerging technologies that will have the greatest impact on the world in the next 3-5 years. Intelligence, friend or foe?

"Top 10 emerging technologies that will disrupt industries, boost economic growth, improve lives and protect the planet. Of course, the premise is that these technologies can be designed, scaled and applied responsibly." Executive Director, World Economic Forum, Director of the Center for the Fourth Industrial Revolution Jiang Ruijie, the person in charge, said.

Since its first release in 2011, the Top 10 Emerging Technologies Report has uncovered many technologies that were initially little known but have gone on to have profound global impact. The genetic vaccines listed in the 2016 report later became the basic technology of most new crown vaccines; the 2018 list mentioned molecular design based on artificial intelligence, and the first batch of drugs discovered by artificial intelligence entered clinical trials two years later.

"It is no exaggeration to say that this year is the year of artificial intelligence," said Mirek Dusek, executive director of the World Economic Forum. Among the top ten emerging technologies just announced, generative artificial intelligence and artificial intelligence-assisted medical technology are all directly related to artificial intelligence.

"A few years ago, people were always talking about the fourth industrial revolution, but now, I think this revolution has far surpassed the industrial revolution, and it is likely to become the next revolution of human civilization." Feng Yan said, today The AI of today is very different from the past. Today's AI is based on large models, and it can generate various content with massive amounts of data. When the amount of data is large enough, it can accomplish many different tasks.

She believes that large models such as generative AI have great potential. It can take many forms and achieve many functions. It can also be used as the basis for building other AI systems, "very creative."

"Technology has undergone subversive development." Zhang Yaqin, chair professor of intelligent science at Tsinghua University and dean of the Institute of Intelligent Industry, believes that different industries can develop vertical applications on this basis, and the rules of the game in the entire world have begun to be rewritten .

In the field of education where Wang Guan, chairman of the AI education company Learnable, is engaged in, the traditional teaching and learning model has been rewritten. "AI has been successfully applied to China's college entrance examination and high school entrance examination," Wang Guan said. In 2017, Learnable's innovation at Harvard The start of the laboratory is based on a specific large model. In his view, although the interaction and emotion between teachers and students cannot be replaced, AI can help teachers in correcting test papers and assisting teaching. Artificial intelligence is more efficient and cheaper than human labor, and can cover a wider range.

During the COVID-19 pandemic, many students were locked at home. Teachers can teach students by providing mobile phones, tablets or videos. "Parents only need to take a photo of their child's homework, and the rest can be done by artificial intelligence." AI will tell students what is right, what is wrong, why it is wrong, and how to improve. He believes that because the cost of providing services is very low, this technology can to some extent reduce the gap between rich and poor, urban and rural areas.

Kamila Markram, co-founder and CEO of Frontiers, said that at present, human beings stand at a critical turning point in the global society, and the power of knowledge will become the guiding light in difficult times. How to effectively understand and use these technologies is a critical step.

While the pilot light illuminates the way ahead, it also casts shadows. Many participants mentioned that while seeing the powerful potential of generative artificial intelligence, the huge risks of technology cannot be ignored.

Feng Yan believes that the performance of generative AI on specific tasks is completely comparable to that of humans, and in some places even surpasses the human brain, which is "extremely intelligent." But its limitations are also obvious-it can't do anything without prompting, "this is the difference between human and machine intelligence."

She further explained that even if a person does not speak, he still has rich feelings, thoughts and contradictions, but machines are different, they do not have a rich inner world. It is good or bad, the key lies in how humans use it.

One of the challenges with the current technology is that it is not yet possible to predict what answers it will generate. Many people have done the test. When using different prompt words, the answer it gives is different, "a bit out of control".

This leads to the phenomenon of "hallucinating" or "lying". They're creative and follow instructions to generate whatever you want, but don't know when they're right and when they're wrong. Feng Yan said that some of the results generated by AI are based on facts, while others are very strange. For example, the generated picture may be a 6-fingered hand, or a historical event that has never happened has been fabricated. The seriousness of this problem is that many people believe in these AI "illusions", including herself, who once believed the lies.

Another problem that needs to be addressed is AI bias. "Some of the answers are biased and even malicious." Feng Yan said that ChatGPT's answers actually reflect the problems of human society, "so we have a lot of work to do."

Similarly, Emilija Stojmenova Duh, Minister of Digital Transformation of Slovenia, is most worried about prejudice. "Prejudice exists everywhere in society, and AI may further deepen the existing prejudice."

Some bias also comes from bias in the data. Most of the data used for large-scale language model training today are English data. Because there are the most English materials and data on the Internet, the English level of large models is very high. "We found in the benchmarking research that their Chinese proficiency is actually quite good, and the reason behind this is also because there are a lot of Chinese content on the Internet." Feng Yan said that it must be realized that there are nearly 6,000 languages existing in the world. "We don't want to see The economic gap between rich and poor is replicated in the model."

Some people worry that some people with ulterior motives will use AI technology to do bad things. Chen Xudong, chairman and general manager of IBM Greater China, said that IBM began to study AI ethics issues very early, and because of this, IBM made an astonishing move—abandoning research on face recognition, "because this may be used to do evil."

The process of technological innovation is accompanied by the contest of "the height of magic is one foot, and the height of road is one foot". After ChatGPT quickly became popular, IBM quickly launched a product based on the accumulation of the past few decades to help enterprises train AI models and ensure the security of data and information. Not long ago, OpenAI, the developer of ChatGPT, launched a tool to identify whether the text was written by ChatGPT.

"Despite concerns, this should not hinder the pace of innovation." Emilija Stojmenova Duh expressed the voice of most people-if you are timid, you will not be able to realize the potential of AI. "The government should also support AI innovation."

Li Xiang, a member of the Executive Committee of the World Economic Forum and head of the artificial intelligence, data and metaverse industry, believes that it is necessary to understand the potential and challenges of AI, and to use and respond to AI responsibly in order to bring innovation, creativity and progress to mankind.