What can the AI system installed at Russian Mi-28NE attack helicopters do? Why did the YouTube’s AI system delete videos about Pokémon GO and how well does the text generator from Elon Musk’s OpenAI fulfil its tasks? Read about these and other news of the last week in the digest.
Company of Elon Musk developed an AI system that can create fake news
OpenAI firm co-founded by billionaire Elon Musk developed an AI system that generates qualitative texts.
Text generator GPT-2 was trained by analyzing 8 million web pages. Thanks to this, the system learned to build sentences about climate changes, US civil war, and celebrities. Besides, the AI system can make fake news.
Developers of GPT-2 fear that the program can be used to do harm, for example, for writing materials that would misinform population. Therefore, only its light version will be released.
Russia plans to launch a national program to study AI
In a message to the Federal assembly, the President of Russia Vladimir Putin offered to launch a national program aimed at researching AI technologies.
The President highlighted that Russia had to become one of the world’s leaders in genetics and AI technology by 2025.
According to Vladimir Putin, to realize such projects, the country has to speed up the establishment of scientific and educational centers.
Mathematicians taught the neural network to sleep
Three Italian researchers taught the neural network to get rid of excessive information as the brain of a living organism does it in sleep.
Mathematicians took the Hopfield neural network as a basis, which is capable of self-learning and storing data. They supplemented the algorithm of the computer program with the analogue of circadian cycle, when wakefulness changes to sleep. In this model, artificial intelligence accumulates data in online mode at first, and then consolidates memory in offline mode.
During trials, it turned out that the neural network that ‘slept’ periodically, performed more efficiently.
AI system of the Mi-28NE helicopter can independently hit targets
In Russia, Mi-28NE attack helicopters are testing an onboard surveillance and attack system that supports artificial intelligence. This system can independently analyze targets in a combat zone and can define which of them belong to the enemy. Then it chooses a weapon available onboard that fits the best to defeat the enemy, and then the AI attacks.
For the system to open fire, the pilot has to choose the target that has to be defeated. To do this, the pilot has to press the corresponding mark on the display. When required, the pilot can cancel the attack.
Thanks to the use of such an onboard system, there are plans to relieve the crew that cannot control the whole battlefield during low-altitude flights at big speed.
AI found content of erotic nature in videos about Pokémon GO
AI system used by YouTube deleted video content about Pokémon GO published by a number of popular bloggers. Besides, some users even lost their channels and accounts. The reason was the abbreviation СРthat was incorrectly deciphered by AI.
In Pokémon GO, CP means ‘Combat Points’. However, the YouTube’s AI system decided that the abbreviation stood for ‘child pornography’. For this reason, the content that contained the abbreviation CP was deleted.
After the incident, bloggers contacted the team of YouTube and explained the situation. In the end, the truth was on their side and the video hosting provider restored the blocked channels and accounts.