The research follows similar work published by a separate group at the University of Oxford earlier this month. Using related techniques, these scientists were able to create a lip-reading program called LipNet that achieved 93.4 percent accuracy in tests, compared to 52.3 percent human accuracy. However, LipNet was only tested on specially-recorded footage that used volunteers speaking formulaic sentences. By comparison, DeepMind's software -- known as "Watch, Listen, Attend, and Spell" -- was tested on far more challenging footage; transcribing natural, unscripted conversations from BBC politics shows. More than 5,000 hours of footage from TV shows including Newsnight, Question Time, and the World Today, was used to train DeepMind's "Watch, Listen, Attend, and Spell" program.

Late last year, famed physicist Stephen Hawking issued a warning that the continued advancement of artificial intelligence will either be "the best, or the worst thing, ever to happen to humanity". We've all seen the Terminator movies, and the apocalyptic nightmare that the self-aware AI system, Skynet, wrought upon humanity, and now results from recent behaviour tests of Google's new DeepMind AI system are making it clear just how careful we need to be when building the robots of the future. In tests late last year, Google's DeepMind AI system demonstrated an ability to learn independently from its own memory, and beat the world's best Go players at their own game. It's since been figuring out how to seamlessly mimic a human voice. Now, researchers have been testing its willingness to cooperate with others, and have revealed that when DeepMind feels like it's about to lose, it opts for "highly aggressive" strategies to ensure that it comes out on top. The Google team ran 40 million turns of a simple'fruit gathering' computer game that asks two DeepMind'agents' to compete against each other to gather as many virtual apples as they could.

This is probably sarcasm, but the point is to get baselines for algorithms which nowadays have lots of tiny tricks which aren't reported in the paper. I had to go through spragnur's DQN code to get those tricks and it took a LONG time...

By the end of this week, it's a good bet that the world's best player of the ancient Chinese board game Go will no longer be a human being. The Chinese Go champion, 19-year-old Ke Jie - ranked number one in the world - was just narrowly beaten by Google DeepMind's AlphaGo in the first of a three-match series, and if the algorithm's winning form keeps up, it'll be a watershed moment in the evolution of artificial intelligence (AI). The latest win, played in the Chinese city of Wuzhen on Tuesday, cements AlphaGo's steady rise to the peak of the professional Go-playing circuit, after celebrated victories over European Go champion Fan Hui in 2015 and South Korean grandmaster Lee Sedol last year. After those decisive tournaments, won by AlphaGo 5-0 and 4-1 respectively, it's possible Ke had even less a chance of beating the system than his human predecessors. DeepMind's developers say the tweaked and revamped AI is now more efficient than ever, using 10 times less computational power than the algorithm that trounced Sedol in 2016.

Google artificial intelligence unit DeepMind teamed up with Chinese authorities to hold a five-day festival in the country this week focused on the ancient game of Go. The centerpiece of the event is a three-game contest pitting a DeepMind computer program against China's world Go champion, Ke Jie -- all of it livestreamed on Google's YouTube. Just one problem: Chinese Go fans couldn't watch the first game on Tuesday because the YouTube livestream was blocked in China. DeepMind's AlphaGo program won by just half a point. Major Chinese news websites were prepared to livestream the game, but the plans were suddenly canceled, according to people with knowledge of the plans who declined to be identified because of the sensitivity of the matter.

AlphaGo is at it again. Google DeepMind's Go-playing AI has defeated Ke Jie, the world's number one player, in the first of three games played in Wuzhen, China. The AI won by just half a point – the smallest possible margin of victory – in a match that lasted four hours and fifteen minutes. Though the scoreline looks close, AlphaGo was in the lead from relatively early on in the game. Since the AI favours moves that are more likely to guarantee victory, it doesn't usually trounce its opponents. In March last year, AlphaGo beat Lee Sedol, one of the world's top Go players, winning four out of five matches.

In March last year, Google's DeepMind artificial intelligence completed an historic victory. Its AlphaGo system beat world Go champion Lee Sedol in a five-game contest of the famously complex board game. Not content with defeating one of the world's best players four games to one, in one of the most challenging board games to exist, the firm can now add another victory to its tally. Ke, who has been the top-ranked Go player for the past two years claimed last year he would never lose to a "cold machine." The night before the event, Ke wrote on Weibo that "the advancement of AI has far exceeded our imagination" but added he would never play it again after this week and said he "cannot feel its passion and longing for the game of Go".

Trying to understand what is *really* going on in the universe.

The Differentiable Neural Computer is an awesome model that DeepMind recently released. It's a memory augmented network that can perform meta-learning (learning to learn). We'll go over it's architecture details and implement it ourselves in Tensorflow. That's what keeps me going.

If imitation is the sincerest form of flattery, OpenAI's newest robot system should leave humanity blushing. Not only can it successfully replicate human behaviors, it can do so after just a single demonstration of the task. The research company co-founded and chaired by Elon Musk used two separate neural networks to develop its one-shot imitation learning system. The first, a vision network, analyzes an image from the robot's camera to determine the location of objects in reality (in OpenAI's video example, these objects are blocks of wood on a table). The network is able to do this despite never having seen the actual table or blocks before.