Can Artificial Intelligence Achieve Infinite Potential

Human beings are very good at taking the next step according to the limited information at present, especially computers.

Can Artificial Intelligence Achieve Infinite Potential

Virtual Intelligence Processing Project

A new project of DARPA aims to train AI to process and evaluate information like human beings and produce operational results on smaller data sets than currently available. DARPA ranks this very important project at the VIP level, where VIP is also an abbreviation for the name of the Virtual Intelligence Processing project.

"If the next generation of AI is to be successfully integrated into the applications of the Ministry of Defense, it must be able to deal with incomplete, sparse and noisy data, as well as possible unexpected situations in practical problems," DARPA wrote in a solicitation letter issued on June 14. Therefore, a new mathematical model is needed to generate efficient and powerful artificial intelligence algorithms, which can learn new concepts with few examples and guide the development of new hardware (hardware that can support artificial intelligence algorithms in the future). To create these mathematical models, DARPA hopes its partners will create an artificial intelligence inspired by large-scale parallel work in the human new cerebral cortex. If brain structure is the mystery that makes humans good at processing information quickly, then brain structure is worth studying. In order to reverse engineer the human brain, we need to apply new mathematical models for various computations. These models must be complete and transparent, and can provide information for the next generation of processors, making them more suitable for the third generation of artificial intelligence.

The original intention of DARPA is to inject funds into this problem area. It believes that these areas can produce future results and can not be served by the market at present. The bidding clearly requires the establishment of mathematical models that have not become the focus of AI development. It is also looking for models that can guide future hardware development, rather than programs that can run on existing machines. DARPA is interested in the way hardware works in simulation, but hopes that partners will delay the time to make hardware for the model.

Thoughts on Project Promotion

At present, the idea of DARPA to promote the project is as follows: inspired by the brain, a mathematical model is established, information is processed on a limited data set, and then the mathematical model is designed for hardware that does not yet exist.

Although this sounds simple, the campaign also requires the sponsors to talk about the limitations of these algorithms in military missions, especially in terms of accuracy, data, computing power and robustness.

Developing work based on limited information is the future direction of military machinery. In electronic warfare, harsh environments and extremely rare and unrecorded battlefield events, more airborne processing of limited data should enable greater autonomy. Even in rare cases where weapon systems transmit data to improve algorithms, the data set is several orders of magnitude smaller than the large data set used to train most commercial machine learning tools.

If the idea of the sponsor meets the plan of the first and second phases of the project and is adopted, it will receive a reward of $1 million. For anyone who can figure out how to make future computers run efficiently on sparse information like human brains, the reward will be enormous.