Elon Musk’s non-profit organisation OpenAI has revealed its advanced robotic hand system which is able to teach itself to manipulate everyday objects using a human-like hand.
The research company, set up by the tech entrepreneur to “benefit humanity as a whole”, thinks the dexterous hand is “one step closer to bringing true artificial intelligence in the real world”.
The state-of-the-art mechanical hand, known as "Dactyl", represents a breakthrough in robotics, where experts have long struggled to mimic the nimble movements of the human hand.
Dactyl learnt its movements using computer simulations, involving repeated trial and error experiments with a virtual robot hand until it slowly perfected complicated movements. The information was then transferred to the real robot hand.
Matthias Plappert, from OpenAI’s robotic team, told The Telegraph: “It only took 50 hours for the AI and simulations to get as good as a human hand, this would have been the equivalent of hundreds of years of data.
"This was done because the simulations could run in parallel, however it takes longer to marry the data to the it to real robot hand in the real world.
"While the hand can currently only hold something just over 2 kilograms, it has great dexterity and degrees of freedom, making it as good or a bit better than a human hand.
"It can learn skills and tricks a lot quicker than a human hand because we can run hundreds of simulations in parallel, our aim is to make the AI as adaptable as possible so it can learn movements and complicated techniques quickly."
The system uses the same algorithm and training system that Open AI has developed to beat humans in one of the most popular and complex eSports games in the world, Dota 2.
Mr Plapper added: "We don't plan to teach it how to play video games but it fun and it would be possible. The AI for the hand can be applied for many tasks, however getting it to operate a keyboard and mouse at the same time would take a long time to learn, but possible."
This article originally appeared in The Telegraph