The ABC brain model provides a theoretical basis for the birth of industrial brain networking.
"With the artificial intelligence general large model as the technical base, the industrial brain network will realize the upgrading of the five capabilities of independent design, independent process, independent processing, independent assembly and independent service, and bring new changes to the intelligent manufacturing system." Gan Middle School, Dean of the Institute of Engineering and Applied Technology of Fudan University and Chairman of Zhichang Group, made the above statement at the 8th China Robot Summit.
Founded in June 2016 by Dr. Gan Zhongzhong, a well-known expert in the field of intelligent manufacturing in mainland China, Zhichang Group is an innovative group company of industrial brain networking with ternary control technology as the core and Sanbee robot cluster as the carrier. Since its establishment 7 years ago, the company has always adhered to the deep integration of a new generation of information technology and industrial manufacturing, and continuously promoted the upgrading and innovation of cluster robot applications.
At this robot summit, Zhichang Group released the Zhichang Bee Brain Industrial Internet Platform. The biggest feature of the platform is the deep integration of intelligent robots and the industrial Internet to make industrial manufacturing more intelligent. At present, self-replicating robot technology has become the frontier technology in the field of intelligent robots, which can achieve autonomous growth, independent evolution, and independent emergence, and if it is applied to the industrial Internet platform, it will bring profound changes to the industry.
As a senior expert in the field of intelligent manufacturing, Gan Middle School made a report entitled "ABC Brain Model and Self-replicating Robot" at this summit, introducing the development direction and theoretical framework of artificial intelligence, as well as the technical framework, implementation path and significance of self-replicating automata proposed by John von Neumann in the robot industry in 1948, providing forward-looking strategic thinking for the development of artificial intelligence and robotics.
Communication Industry Network, Zan 73
What is the ABC brain model
The ABC brain model is the beginning of the story of self-replicating robots. In 2017, Fudan University accepted a major science and technology project in Shanghai and set up a special team to study robotics and artificial intelligence, when the research and development team proposed the concept of the ABC brain model.
As the core of the research team, Gan Middle School experienced the whole process of the development of the model. He said that the ABC brain model describes a heterogeneous swarm intelligence system that combines artificial intelligence (AI), biological intelligence (BI) and swarm intelligence (CI) to form a structure similar to the human brain, and continues to promote the improvement, evolution and emergence of AI through the cycle of ABC.
ABC Brain has seven thought chains. First, the multimodal sensor information of the robot group is used as input, through holographic decoding, visualization, and intuitive decision-making links, without passing through the limbic brain and cortex, forming a rapid closed loop.
The second is driven by the general basic model of the robot, forming a pathway link from the whole system coding to attention, reasoning, and then to decision-making.
The third is the multiple decision-making pathway of CI in the loop, which gives full play to the general scene recognition and reasoning capabilities of CI, and collaborates with the robot ontology decision-making.
The fourth is the introduction of the ABC slow cycle process of BI, on the one hand, the introduction of BI to improve the ability of boundary scenarios, on the other hand, to promote the alignment of the values of AI and CI models with humans.
The fifth is the continuous learning process of the general basic model of robots, which accumulates and sorts out the knowledge of the biological brain into knowledge representations with causal relationships, and promotes the iteration and evolution of the robot world model through the paradigm of supervision/semi-supervision, long-term and short-term incentives.
In addition to the existing supervised learning, the interaction and closed-loop between GPT and the real world can be studied in the future, such as reinforcement evolutionary learning of multimodal large models.
The seventh is the long-term and short-term stimulation of the ABC brain, which is stimulated by the intention of the biological brain, and obtains long-term and short-term stimulation through interaction with the environment. Long-term and short-term incentives can further change the motivation paradigm of the robot world model and GPT.
The proposal of the ABC brain model has a direct impact on the evolution of robots, and then provides a theoretical basis for the birth of industrial brain networking.
Let robots replicate and evolve themselves
Although robotics has made great progress, there is still a certain gap from the era of self-replicating robots.
Gan Middle School said that today's robots are still in the stage of automatic control, and they do not yet have growth mechanisms similar to organisms, such as self-growth, self-evolution, and self-emergence, but in the field of language, with the emergence and development of ChatGPT, the above mechanisms have been realized, and the development of the robot industry has a clear direction, that is, the future robots must be able to self-grow, self-evolve, self-emerge, and even self-replicate and self-develop.
Of course, this assertion is not without purpose. More than half a century ago, von Neumann, the "father of computing," argued that a self-replicator should include a coded representation of the individual, a mechanism for copying the code, and a control that influences and shapes the environment in which the individual lives.
Following this line of thought, von Neumann proposed a theory of self-replicating automata. In his theory, a self-replicating automaton (R) was born like this: the universal coding machine (Y) compiles the feature code of R (φ), and the universal controller (Z) first orders Y to copy two copies of the φ, and then orders the universal constructor (X) to actually make R according to the φ, and remove one of the φ copies, and Z will bundle X with the remaining φ copy and separate them from the combination of X+Y+Z, thus creating the combination of R and φ. Thereafter, with φ, R can infinitely self-make and self-replicate.
Gan believes that von Neumann's theory makes logical sense, but has flaws, that is, where did the world's first self-replicating automata come from? Machines are inanimate and cannot evolve from single cells like living things. Therefore, the world's first self-replicating automata still needs to be made by people, and self-replicating automata cannot be completely separated from people to achieve self-replication, at least not in von Neumann's time.
But if the ABC brain model is combined with von Neumann's theory, it will be possible for robots to replicate themselves. According to Gan, the cycle of the ABC brain model can realize three characteristics of self-replicators, in which holographic representations (robot language) are used to complete the coded representation of robots; Use ABC loop to iteratively automate the code generation and replication process; Use the closed loop of the machine and the environment to obtain negative entropy and achieve self-replication.
"The circular structure of the ABC brain model provides a possible paradigm for robots to replicate themselves, which also embodies the next generation of AI for engineering design ideas." Gan Middle School emphasized.
The new future of robotics
Relying on von Neumann's self-replicating automata theory and ABC brain model, Gan imagined a new future of robots, that is, the construction of future intelligent self-replicating robots.
Speaking of this future, MOSS, a conversational large-scale language model developed by the Natural Language Processing Laboratory of Fudan University, plays an important role in it. It is reported that MOSS is based on open Chinese and English data training, currently has 20 billion parameters, has the ability to talk to humans, and can achieve iterative optimization through interaction with humans.
Gan Middle School said that the introduction of MOSS as an intelligent brain for universal controllers; Robots and 3D printers serve as general-purpose constructors, and this new combination greatly increases the possibilities of robots to make robots. However, there are still some challenges to overcome in the process.
"At present, in the field of industrial manufacturing, the complexity and performance of the industrial mother machine is much higher than that of the product being produced, and if the self-replication of the machine is to be realized, then the complexity of manufacturing equipment and products will be the same, which requires the collaboration of the complete industrial chain to achieve." Gan Middle School said frankly: "This also means compressing the elements of the entire industrial chain into universal self-replicating automated machines." ”
In Gan's conception, this process would require three steps. The first step is to compress the main part of the production process to a country and region with a complete industrial chain; The second step is to compress the robot production process to one or several limited enterprises through vertical integration of the industrial chain; The third step is that as the robot's capabilities become larger and more flexible, the robot's own process is compressed into a single robot.
At present, some progress has been made in this work. According to reports, the beginning of machine self-replication - product design, part design, processing process design, assembly process design and other product feature gene coding has been constructed, and relying on the ABC brain model and MOSS learning training, the iterative evolution of product feature gene coding has been gradually realized.
Gan Middle School pointed out that it is currently trying to use the modular building block method to automatically frame the design of robots, and use big data models to promote the self-programming of universal controllers. Moreover, the use of 3D printers can realize the additive manufacturing of some modular parts of self-replicating robots, the additive and subtractive manufacturing of functional modules, and the automatic assembly of prototypes.
From the current results, the robot's autonomous processing and independent assembly capabilities have been preliminarily equipped, and it will be promoted in the direction of independent design and independent process in the future. Gan said that the goal of independent design is to be able to convert product structure descriptions into product drawings of automatic generation capabilities, and the goal of independent process is to convert product drawings into part drawings and assembly processes.
It is worth noting that at present, Zhichang Group and Fudan University are jointly building a self-replicating robot design service platform for autonomous services, and are openly soliciting partners to provide free technical support. It is envisaged that autonomous services should actively understand human needs and provide services, which will greatly expand the application space of self-replicating robots.
Industry experts believe that the development of self-replicating robots and related technologies has painted a new future full of high-tech and intelligent elements, which will profoundly affect the development of industrial production mode and industrial Internet, and reconstruct and upgrade the intelligent manufacturing system.
A new, smarter world is accelerating.
Written by: Superb
Chart: Zhang Shunian
Edit: Jingjing
Guide: New article