Which math problem is beyond AI’s capability? Why do American military officers need AI based on an insects’ brain? How accurate is a neural network at diagnosing a genetic disease? Follow our weekly digest to answer these and other questions.
Russian Research Institute of Artificial Intelligence
As reported by a representative of National Technology Initiative (NTI) AutoNet, Russia will have introduced an organization for AI development and improvement by 2035.
Particularly, the expert announced that this institution would allow graduates of specialized universities to teach artificial intelligence various practical tasks: from driverless cars to V2X communication technologies. It is also expected that AI businesses would share their products and solutions to the organization for testing and improvement.
Besides, NTI AutoNet representatives provided their forecast concerning the appearance of a new profession of an AI specialist by that time.
Neural network diagnoses genetic diseases
An AI startup FDNA developed a neural network called DeepGestalt. It is capable of diagnosing genetic diseases by a single human photo.
Within a research project, AI specialists fed 500 000 photos of human faces to a neural network. Cornelia de Lange syndrome and Angelman syndrome were chosen as diseases to be determined. They violate cognitive processes and movement functions if not treated.
The neural network divided images into small fragments and searched for general diseases symptoms. As a result, AI managed to diagnose a disease in 90% of cases.
AI in the Pentagon
As commented by the Assistant Secretary of the Army Bruce Jette, the Pentagon intends to implement AI-led systems into top-notch weapons, particularly, anti-aircraft warfare and missile defense.
According to Bruce Jette, the US military has already established a special development center at Carnegie Mellon University. He noted that the AI introduction program would include ballistic missiles, drones, cruise missiles, and shells.
To exemplify the abovementioned, he fired back a rival’s artillery salvo. Under usual conditions, one man has to follow an enemy’s missile in order to shoot it down. However, sometimes one person is simply unable fire back as quickly as AI does from physical standpoint.
Puzzle a neural network cannot solve
Scientists from Israel Institute of Technology worked out a problem unsolvable to any neural network (at least, among the existing ones).
This task was based on an incompleteness theorem of a mathematician Kurt Gödel. It was published as far as the thirties of the previous century. The scientists claimed that an arithmetical solution is not ultimate. Modern neural networks learn from mistakes. However, some mistakes occur because maths cannot reflect the right answer by its laws.
U.S. military needs primitive AI
The U.S. Department of Defense plans to proceed with elaborating solutions initiated in the 1980s: primitive AI enabled by an insects’ brain. It is handy for simple military actions.
The institution ordered to come back to the developments of the 1980s when a specialist at Massachusetts Institute of Technology Rodney Brooks created a couple of robots. One of the most effective robots was Herbert who collected empty soda cans around the office. It could move over obstacles, detect empty cans and put them into a bin. Its AI was fueled by only 15 rules.
U.S. military officers need such primitive AI in order to solve simple tactical tasks and save money. The project launch is scheduled for April 3.