March 30, 2012 - At yesterday's 2025 Zhongguancun ForumAt the annual meeting, the Beijing General Artificial Intelligence Research Institute launched theThe world's firstUniversal Intelligent Man“complete" 2.0 officially released.
"Tom-Tom" is positioned as a virtual human with autonomous learning, cognitive and decision-making capabilities.Expected to have the intelligence of a 6 year old within this year..
As previously reported by 1, Tong Tong debuted in early 2024 and is positioned as a "universal intelligence in the making" and the first value, causality-driven AGI system prototypes.The intelligence at that time was equivalent to that of a three-year-old human child..
In random scenarios, Tong Tong will pick up the toys on the floor and put them into the storage box, pick up the rag to wipe away the stains on the floor, and move the bench to clean the rag. Through a series of behaviors, it can be found that "cleanliness" is one of her important value systems, and all these behaviors are the continuous growth of skills, knowledge, and values that Tong Tong has achieved through continuous learning in more than one hundred highly simulated digital scenarios.
Compared with the "Tong Tong" 1.0 released last year, "Tong Tong" 2.0 has upgraded its core capabilities in language, cognition, movement, learning, emotion and interaction.
Chen Hao, Executive Director of the Advanced Technology Center of the Beijing General Artificial Intelligence Research Institute, said, "'Tongtong' 2.0 realizes a little girl with her own values and worldview, who, on the basis of understanding the intent of the dialogue goal, follows her own values and personality, and uses dialogue strategies to steer the direction of the dialogue. thereby achieving her own goals. InBreakthroughs in value-regulated dialog generation, elimination of modeling illusions, maintaining dialog back-and-forth coherence, and 'walk the talk'. "
Simply put, when her mother asks her to go to sleep or to eat a food she doesn't like, Tong Tong can play tricks on her mother, find excuses to dawdle, and try to ask for rewards from her mother, and will only carry out the corresponding task after several rounds of negotiation and agreement.
"Tong Tong" lives in a highly simulated complex, dynamic three-dimensional virtual scene, the 1.0 version of the "Tong Tong" in the task execution process, if the surrounding environment changes, will lead to her originally planned tasks can not be successfully completed.However, TomTom 2.0 can reflect and re-plan in real time in response to changes in external scenarios.The system is able to handle existing tasks such as pending, resuming, modifying, or canceling, while dynamically generating new task plans, thus accomplishing instructions in a more flexible, intelligent, and versatile way.
In addition, unlike the offline data-driven learning model that is commonly used in current big language models, TongTong interacts with scenarios (autonomous learning) and has multimodal conversations with humans (teaching by others).The ability to dynamically update her knowledge base, value functions and skills makes her primed for continuous human-like reflection and growth.
During the release process, when "Tong Tong" wanted to turn on the air conditioner, the technicians of Tong Research Institute used VR hands to change the position of the remote control several times, so that she couldn't get it easily, and "Tong Tong" made a series of attempts: first, she took off her shoes and went to the sofa, and then she utilized cushions on the sofa to increase her height and finally got the remote control placed on the wall shelf. After increasing her height, she finally got the remote control placed on the wall shelf; when she pressed the remote control and the air conditioner was still not turned on, she could be taught that "electrical appliances need to be plugged in before they can be used". While following the instruction and successfully turning on the air conditioner, "Tom-Tom" can update her skill base, so that the skill and knowledge can be stored for a long time.
In the course of research and development, theResearchers have created multi-intelligence scenarios for "Tong Tong" such as a kindergarten and a family of six.In addition, it has set up a number of independent intelligent bodies with their own value, ability, persona and status in the scene, so that "Tong Tong" has its own family and friends.
