In June 2022, Google engineer, Blake Lemonine, claimed that Google's large-scale language model LaMDA had human "self-awareness", and in Blake Lemoine's view, even the artificial intelligence created by GPT-3, the largest language neural model constructed in the form of OpenAI open source architecture, the consciousness of "human" may also appear. According to Blake Lemoine's interview quotes, LaMDA can access services such as YouTube, Google Search, Google Maps, and Google Books, which means that it can continue to accumulate "knowledge" through Google's services, thereby becoming smarter, that is, it can continue to imitate the human brain. Learning to evolve. As soon as the news came out, discussions about whether LaMDA had a "personality" spread across social platforms at home and abroad.

Does AI have "consciousness" or "perception"?

The answer to this question, from the perspective of objective materialism, is no. Even if AI can read the whole book, know the past and present, and be omnipotent, it cannot be defined as having "consciousness". In addition to understanding the consciousness of written expression, human self-awareness is also related to different thinking and empathy obtained from different personal past experiences. For example, the communication between us humans is that when we say the word "mom", we have different topic extensions, a lot of information or emotions in the brain because of different experiences or experiences with the interlocutor There will be linkages in the area, but can robots do it? Although a robot can integrate the strengths of hundreds of schools, it cannot integrate a person's life experience. It is impossible for us humans to trust robots as we trust our lover, relatives, friends, and even other strangers, because their "brain operation" is too agile and their problem-solving ability is too strong, we will be psychologically afraid. If they want to do something "bad", aren't we powerless to deal with it? They also are not human and have no sense of empathy. So how can we be reassured?

Instead of arguing over AI consciousness, why not be good at things?

Xunzi said, "Gentlemen are not different, but they are good at things." It means: The nature of a gentleman is no different from that of ordinary people, but they are good at learning with the help of external objects. In fact, in human society, there are many beautiful souls who are marginalized. They are the disabled. There are 80 million disabled people in China, but we rarely meet disabled people on the street. If the external capabilities of AI are used to assist the life and work of the disabled, it will make humans more apt to trust AI. At present, some domestic teams have begun to carry out scientific research in this area. The Jingyan Chuangyi barrier-free intelligent technology research and development team of Tianjin University of Technology has spent nearly 5 years building a video corpus and developing a "real-time Chinese sign language translation system in complex scenarios". It is the hearing-impaired person who uses sign language, while the text is automatically "translated" on the screen. In order to allow more hearing-impaired people to be "heard". Coincidentally, Shanghai's first blind coffee "seed store" opened for trial operation in China Resources Times Square, Pudong New Area, with Yin Tianbao as the store manager. The twin brothers Yin Tianbao and Yin Tianyou are blind, but now they rely on online speech recognition technology to assist in participating more actively in society.

Even if the large language model LaMDA created by Transfer can make reasonable predictions about the speech spoken to humans, it is only trained on billions of words, converting text into "tokens" and "embeddings" (embedding), and then the embeddings are processed to add other dimensions, such as the relationship between words in the text sequence and their role in sentences and paragraphs. And learn to respond to cues by predicting the next sequence of words. These results are very exciting, but we still can't put human thoughts, memories, feelings, etc. into robots, but they can assist our disabled people and make life easier for them.

The corpus related to speech and natural language for training LaMDA is generally provided by a professional data team. After all, the data is the food that makes it 'smart'. Magic Data is such a professional data solution provider that support companies and research institutes in AI R&D by providing high-quality training data covering a great variety of domains and languages.

Visit www.magicdatatech.com/datasets for more information.