Developers started to wean AI off racism after the scandals with Amazon, and DeepMind’s neural network once again showed superiority over the human. Read about these and other news in our weekly digest.
Russian developers built AI to eliminate blackouts at production sites
Center of the National Technological Initiative at the National Research University "Moscow Power Engineering Institute" developed software that can reduce risks of blackouts (loss of electric power). The AI-based product will restore the operation of electric-power systems without human intervention. The first stage of trials will take place in late 2019.
Next victory of DeepMind: now over StarCraft player
In a live stream, AI developer DeepMinds howed how its bot AlphaStar beat the professional player in the RTS StarCraft II. Twenty-five-year-old GrzegorzKomincz from Poland lost to artificial intelligence with 5:0.The neural network revealed strategies not known by professionals.
After the game Komincz said, “It was nothing like StarCraft that I used to play.”
AI learned to recognize concealed social connections
The Massachusetts Institute of Technology (MIT) developed artificial intelligence capable of defining concealed social connections, a regimen on which people’s relations in a group are based. They define the leader and people that have no influence in the group.
The neural network had been previously tested on a group of players. It had to define members of the alliance in the video game and predict their actions.
The algorithm’s predictions matched the evaluations of people. This shows the capacity of AI to make conclusions even basing on a small amount of data.
Artificial intelligence can translate thoughts into words
The team of professor Nim Mesgarani from Columbia University built a neural network that can translate brain signals to words.
During the experiment, scientists played records of counting from zero to nine to volunteers. The neural network analyzed brain activity and voiced words through voice synthesizer. Professor Mesgarani says that the accurateness of reproduction was 75% – significantly better than before. Now developers plan to expand functionality of AI and teach it to recognize more complex words and collocations.
Scientists figured out how to save AI from racism
MIT scientists figured out how to save artificial intelligence from racism. They are developing an algorithm that cleans training data sets from photos that can lead to mistakes in the recognition of people with dark skin.
Researchers do not promise that the method will give 100% result, but they say that the number of mistakes will decrease. Intrinsic tests confirm that AI has become 60%less biased without losses in recognition accuracy.
You can find relevant information about artificial intelligence technologies
at AI Conference held in Moscow on April 9.