Curated by Sergiu Gabriel Rolnic .
It belongs to DeepMind, the famous company acquired by Google in 2014, the latest advancement in artificial intelligence . At the International Conference on Machine Learning, researchers presented a new machine learning model called the Graph Network-based Simulator .
We are faced with a framework capable of simulating complex physical phenomena and predicting the interaction between thousands of different particles . Whether it's liquids, solids, or other deformable materials, the deep learning-based program can model their behavior and make it their own.
A bit of history
Since the birth of the scientific method, experimentation and observation of events have been the fulcrum and starting point for every invention and / or discovery. The advent of computers and their evolution has led to the creation of virtual simulations , environments that represent in all respects a model of the reality that one wants to study, and can concern physical, mathematical or biological laws.
Simulations can be used to learn about the past and future of the universe, the evolution of a species, and even to find the vaccine for the coronavirus . Creating a simulator is very expensive and can take years of engineering effort, which almost never corresponds to a proportional level of accuracy and scalability.
The Graph network-based Simulator (GNS)
The Graph network-based Simulator is a simulator based on neural networks that unlike its predecessors is much more efficient , easier to implement and accurate. The present architecture is represented by the mapping of the same data structure both in input and in output. However, the final graph will be potentially altered in nodes and arcs thanks to machine learning.
The simulator was trained with datasets that represented the domains of very different materials. Each node of the graph indicates the state of a particle, while the arcs the interactions between them. In this way it is possible to create an isolated system and predict the effects that gravity, obstacles or other materials are able to have on the model under consideration.
Simulation of reality with artificial intelligence: future evolution
Despite being tested in very particular situations, the neural simulator developed by DeepMind will be able, with the next updates, to assimilate in a natural and progressive way much more complex physical knowledge , such as Hamiltonian mechanics .
The road to the future has now been undertaken. According to many scholars, the technological singularity is now upon us, and it could be thanks to simulations that we will be able to create the first true artificial intelligence . After all, we still know very little about our brain, what consciousness is and how it could have developed.
However, if we were able to create a simulator under the same conditions as the primordial world, we could perhaps use the capabilities of computers to perform millions of simulations in an infinitesimal amount of time , and capture the anomalies that may have led to an evolutionary upgrade. Or, much more likely, there will be a technology that will come out of nowhere and completely revolutionize our world.
All that remains is to wait and be dragged into the future, or choose to actively participate and create it!
The article Simulation of reality: the last frontier of Artificial Intelligence comes from TechCuE .