Chinese scientists have built an AI known as Tong Tong, which they claim can “interpret human emotions.” Bigai refers to its new creation as an “AI child,” but in reality, this is just software trained to perform certain tasks on its own without human agency.
Bigai revealed Tong Tong, which means “Little girl” in English, at a Frontiers of General Artificial Intelligence Technology Exhibition held in Beijing from January 28 to 29.
The Chinese entity – Bigai – was established in 2020 by Song-Chun Zhu, who spent 28 years living, working, and studying in the US before abandoning his professorship at UCLA to start a company back home.
Interpreting human intentions
Tong Tong, was programmed to clean up her environment and bring tidiness around her. Ideally, she can “interpret human intentions,” according to the South China Morning Post (SCMP), because of her ability to know what a human being would do next. For instance, if Tong Tong sees a crooked frame, she fixes it.
Whenever she sees spilled milk, she cleans it herself without being asked.
“Tong Tong possesses a mind and strives to understand the common sense taught by humans. She discerns right from wrong, expresses her attitudes in various situations, and has the power to shape the future,” says Bigai in their video post.
According to SCMP, Tong Tong depicts the behavior of a three- or four-year-old and is expected to improve on her skills, values, and knowledge through continuous interactions.
Also read: AI Researchers Discover AI Models Deliberately Reject Instructions
The Little Girl has feelings too
The firm sees the development as a huge step towards achieving artificial general intelligence (AGI), a state where machines can think or reason like human beings.
According to the Bigai video, Tong Tong can express feelings as she “has her own joy, anger, and sorrow.” Apart from emotions, researcher and Bigai director Song-Chun Zhu further explained that having common sense like humans is a critical component of the general aspect of intelligence.
Additionally, the AI child should not only be able to complete an “infinite array of tasks but should also define new ones autonomously.”
“To advance towards general artificial intelligence, we must create entities that can comprehend the real world and possess a wide range of skills,” explained Zhu.
Zhu, a renowned scholar who has specialized in AI’s areas of research, includes the general AI field, autonomous robots, and computer vision, to mention a few. He has won several top awards in various fields, such as the ONR Young Investigator Award from the US Naval Research Laboratory, and served at key institutions in the AI sector.
The Tong Test
Bigai also exhibited a platform for AI testing known as the Tong Test, which Zhu and his team published in the journal Engineering that the Chinese Academy of Engineering hosted last August.
According to the SCMP, traditional AI tests, which focus on task orientation or human identification and “virtual environments, have their limitations.” For instance, the Turing Test only assesses an AI’s communication level with humans.
Although they create realistic experiences, virtual environment tests “tend to oversimplify physical environments.”
But the Tong Test assesses five dimensions, according to the researchers, and these are cognition, language, vision, motion, and learning.
“With nearly 100 specialized tasks and more than 50 general tasks, the Tong Test offers a complete testing regime for the development of general artificial intelligence,” the institute said in a statement on its website.
“For general AI to integrate seamlessly into human environments, it must learn and execute tasks in complex settings, driven by values and an understanding of causality,” said Zhu in a statement.
For this reason, Zhu said, his team proposed the Tong Test, which is a new direction for testing AI with a special focus on practical abilities and values.
“Our research will guide general AI in learning and improving its capabilities more effectively and safely, ensuring it serves human society better.”
and then