By Anton Evseev
Representatives of the U.S. Department of Defense announced the launch of a project to develop computer systems with elements of artificial intelligence. But we are not talking about creating a "computer mind". The military are no longer satisfied with the speed of analysis and data quality of modern computers. No one is going to produce "terminators" at this point.
It seems that the Pentagon is seriously interested in artificial intelligence. It has nothing to do with insufficient intelligence of the employees of the U.S. Defense Department (all idle speculations about the low intellectual level of the Americans are ungrounded). The reason is quite simple - the computer systems of the U.S. military authorities cannot cope with the array of data that comes in every minute.
The U.S. military has long complained that their analysts are simply drowning in the ocean of data from drones, satellites, the Internet, exploration, etc. As a result, this affects the speed of decision making. Often such delays are due to the fact that experts cannot distinguish the most important from a large flow of information.
The situation is exacerbated by the fact that the collected data is often imperfect, incomplete and heterogeneous. For example, specialists have to deal with fragments of documents incorrectly decrypted by satellite images, photographs that were taken at a time when high quality was simply impossible, and so on. All this data must not only be understood, but also collected into a database used by analysts. This is not possible without bringing all to a common standard, which takes lots of time.
In short, the U.S. military is so tired of all this, that they launched a major project to develop a system to extract the useful experience of large and complex arrays of digital data. The name of the project and partners are kept secret, but it is known that the initial investment will amount to no less than $200 million. Also, there was information that interaction with the project XDATA of the defense research agency DARPA is planned (annual spending of $60 million). Under the project systems for automated mining of large volumes of various unstructured data will be developed.
The U.S. Department of Defense believes that the systems with the elements of artificial intelligence must have the ability to quickly analyze the data. In this regard, the use of the so-called memristors seems promising.
Memristors are chips that combine the functions of memory and analysis. Now in conventional computers the devices that perform these functions are separated and located in different parts of the electrical scheme. Each computer unit is paired only with the neighboring electronic components. This greatly slows down the computing device (and it is inefficient from an energy point of view). As a result, even the most powerful supercomputers cannot, for example, recognize a human face in their database in a second (the human nervous system copes with the task in a fraction of a second).
Recently, scientists from the HRL lab and the University of Michigan have demonstrated the first functional array of memristor-based complementary logic on a metal-oxide semiconductor (CMOS). Interestingly, this project is also overseen by DARPA. Researchers have developed vertically integrated hybrid electronic circuits that combine new memristors with CMOS chips. This hybrid has already shown impressive results, as memristors can store data with a density of 30 gigabytes per square centimeter. At the moment it is a record in microelectronics.
Perhaps, new computers for the Pentagon will be based on this technology. According to forecasts, the use of memristors will increase the processing speed by a hundred thousand times. In addition, their use opens up the possibility to create complex intelligent machines that can "comprehend" what is happening in real time.
The Military Department also announced that in addition to a new generation of supercomputers that will deal only with the analysis, they plan to develop artificial intelligence systems for various military automated devices, such as drones. These, above all, will be the systems of "visual intelligence". They will be equipped with special data processing algorithms allowing the UAV sensors to ignore irrelevant events, and transfer only important data to its "brain".
Further, the on-board computers with elements of artificial intelligence will be expected to mark the important information with relevant commentary. This will automatically detect and identify the enemy. In the future, the development of devices that will allow robots to not only observe the enemy, but, if necessary, destroy it in automatic mode, is planned.
All of this at first suggests that the U.S. military decided to implement the ideas presented in science fiction films The Terminator. However, this is not the case - the machines developed under the auspices of the Pentagon will not be able to think like humans. At this time this is impossible in principle. They plan to only increase the speed of data analysis and quality of its use. Therefore, fears of a takeover of humans by machines are ungrounded.
No comments:
Post a Comment