The Pentagon allocates money for AI estimated at the budget of the Manhattan Project while AI from MIT learns to handle objects it has never seen before. Read about these and other pieces of news in the AI universe in the weekly digest.

Pentagon to spend $2 bn on AI

Steven Walker, Director of the Defense Advanced Research Projects Agency (DARPA), announced the promotion of the new technological solutions in the sphere of military conflicts. The development of AI systems for the U.S. Army estimated at $2 bn has started.

Compared to other Pentagon’s investments, DARPA’s ones are not considerable. For example, the budget of combat planes F-35 is expected to surpass $ 1 trillion. However, it’s the first time such a great amount of money has been allocated to AI.

MIT scientists taught AI to handle objects it has never seen before

Scientists at the Massachusetts Institute of Technology (MIT) created the artificial intelligence Dense Object Nets (DON) capable of handling objects of unknown origin.

During the training course, AI investigates an unknown object using video cameras on manipulators and detects points of its edges. Then, DON analyzes the components of the object, remembers its characteristics, and structure. On average, this process takes about twenty minutes. When the algorithm faces a new object, it tries to compare it with the previously seen one.

During the research course, the developers showed DON a brown jogging shoe and taught the system to interact with it. When AI saw other sneakers, it detected an analog of the previous object and showed the similar way of interaction.

According to DON creators, this product would boost the efficiency of robotic assistants thus allowing retailers to save money and improve the manufacturing process.

Simulator demonstrated the process of AI training

The developer Kevin Donyagard presented an AI-powered app Evolution. This is a simulator that shows how AI develops. Users can design real creatures, set a task (jump, run, walk) and then track their evolution.

All the creatures in the program have bones, muscles, and joints. In the process of evolution, they develop a brain that is a neural network learning from movements.

The training course lasts until a creature performs a task with the highest precision. Test simulation is available on the website of Evolution.

AI to ban offensive memes on Facebook

Facebook developers announced the creation of an AI-powered program Rosetta aimed at reading texts in memes on Facebook or Tweeter.

“Text understanding is important. For example, it helps search for photos and include text in reading programs from the screen, which makes Facebook available to visually impaired users,” explain the developers. They also note that text reading in images is also very important to recognize improper or hazardous content for the community.

However, Facebook researchers emphasize that their product can be widely applied. For example, Rosetta may read a photograph on the menu, road signs as well as words on labels.

To find out more about AI, join us at AI Conference Kyiv scheduled for November 14, 2018.