Normally Artificial Intelligence systems are trained for a specific task. The development of this type of system is not easy at all. They have to be programmed with thousands or millions of sources to specialize them in some area.
Today we have technology that is capable of predicting the weather, caring for the environment, or detecting disease with greater precision than a human. But apparently we are still in the first steps as China has claimed that it wants to create the largest and most powerful AI.
It is known that the Asian country is making a strong investment in the development of emerging technologies that allow them to be more competitive and achieve greater economic leadership in the world. For example, they are pushing supercomputers and deep learning.
In that sense, the Beijing Academy of Artificial Intelligence (BAAI) recently published details about its “Wu Dao” artificial intelligence system that is attracting the world’s attention as a multimodal AI system. What that means is that it has the ability to do many different things.
According to the Engadget medium, Chinese AI has the ability, for example, to generate text, even poetry; create audio and images, for example, music videos; make the legend that accompanies the images; And, the developers have stated that it even has the ability to discover patterns to “boost virtual idols”.
The reason that the Wu Dao system has all that potential for creation has to do with its size. This AI model is huge, to give you an idea, according to BAAI, it was trained using around 1.75 trillion parameters, yes trillions, much more compared to the larger OpenAI model , GPT-3, which is trained with only 175 billion.
According to Zhang Hongjiang, president of BAAI, the intention of the academy is to create the largest and most powerful AI model possible. “The road to the general artificial intelligence is and large mainframe computers models. What we are building is a power plant for the future of AI, with Megadatos, mega-power computing, and models mega, can transform data to feed AI applications of the future, “he said.
Although it sounds that through this AI system important advances can be achieved in all areas, since deep learning will allow you to continue “evolving”, the reality is that it is not clear if all the effort that is being allocated really drives new milestones in the area. What’s more, an increasing number of experts in artificial intelligence and computer science believe that deep learning is a dead-end for general artificial intelligence.
Further more, details on exactly how Wu Dao was trained, what was in his various data sets, and for what practical applications it can be used remain scarce. This is why, although the Chinese claim that their system is 10 times better across the board compared to other AI machines, there is still no reason to believe that it will bring us closer to truly intelligent machines.
The most powerful supercomputer
Another milestone that China claims to have reached is the development of the most powerful supercomputer. Last December, state researchers from the Asian country published a report in which they claimed to have developed a supercomputer that is 10 billion times faster than Google’s “Sycamore” machine, which was presented in 2018.
in this article, the Chinese specialist explained that they built a supercomputer that uses light to perform a very specific task: sampling bosons (mediating particles of the weak nuclear interaction, one of the four fundamental interactions of nature) for the sole purpose of demonstrating and measure its effectiveness.
According to their first results, they showed that supercomputers can, in theory, solve really difficult problems that physicists and computer scientists estimate that a classical machine would take thousands of years to solve. One more sample is that they are at the forefront of technology.