China: World’s first AI child can now negotiate, throw tantrums like a 6-year-old

TongTong evolves further

TongTong 2.0 builds on its predecessor’s capabilities, advancing in language, cognition, movement, learning, emotion, and interaction. Initially introduced in early 2024, TongTong 1.0 had intelligence equivalent to a three-year-old child. The new version now demonstrates cognitive abilities comparable to a 5- to 6-year-old.

TongTong 2.0’s value-driven conduct is one of its noteworthy features. It displays a developed “cleanliness” value system by cleaning toys, wiping stains, and organizing objects on their own in simulated situations. Training in over 100 fully realistic digital environments leads to this increase in values, knowledge, and abilities.

Chen Hao, Executive Director of the Advanced Technology Center at the Beijing Institute for General Artificial Intelligence, described TongTong 2.0 as a digital persona with her own values and worldview. According to 1ai, the AI engages in dialogue based on her understanding of intent while considering her values and personality. Coherence, realistic conversation management, and value-regulated dialog generation have all enhanced the system.

In everyday encounters, TongTong 2.0 can display human-like traits like negotiating or delaying tasks. For example, when told to sleep or eat something it doesn’t want to, it might justify its actions or look for rewards before complying.

An important advancement in AGI development is this capacity to control dialogue and mimic emotions. The launch of TongTong 2.0 demonstrates China’s quick progress in AI, which is fueled by open-source creativity.

Adaptive AI advances

Unlike its predecessor, which struggled with task completion when faced with environmental changes, TongTong 2.0 can reflect, re-plan, and adjust in real-time. It can pause, resume, modify, or cancel tasks as needed while dynamically creating new plans to meet goals more intelligently and flexibly.

According to 1ai, TongTong 2.0’s autonomous learning is one of its main features. It converses with humans in multiple languages and gains knowledge from real-world situations instead of depending only on offline material. It mimics human-like introspection and development by updating its knowledge base, value functions, and abilities through this ongoing interaction.

In its demonstration, TongTong 2.0 demonstrated its capacity for problem-solving. TongTong retaliated by taking off her shoes, raising herself up with sofa cushions, and grabbing the remote control when technicians placed it out of her reach. She was instructed that items must be plugged in after she found out the air conditioner wouldn’t turn on. Her knowledge base grew as a result of this experience, which also helped her remember and use the abilities she had learned.

Researchers constructed virtual habitats, such as a kindergarten and a six-member family, to aid in its development. By introducing a variety of intelligent beings with unique personalities, skills, and values, they created a social network of “family and friends” for TongTong. Her continuous development is aided by these situations, which further hone her social and emotional intelligence.

The developments of TongTong 2.0, which prioritizes intelligent interaction and adaptive learning, are a prime example of China’s AI achievements, reports 1ai.