The year a program beat the strongest Go player of his generation

In March a DeepMind program called AlphaGo won four of five games against Lee Sedol in Seoul. Go had resisted computers far longer than chess, and the result reached an audience well beyond research laboratories. The methods behind it, deep neural networks trained on human games and on self-play and paired with tree search, had been published in Nature two months earlier. The year set the pattern for the decade that followed: large compute budgets, systems that learned from data rather than from hand-written rules, and public demonstrations as the way progress was announced and argued over.

13 entries recorded

December

  1. Amazon opens a cashier-less Go store to its employees

    Amazon opened Amazon Go, a convenience store of about 1,800 square feet in Seattle, on 5 December 2016. Cameras, sensor fusion and deep learning tracked what shoppers took from the shelves and billed them through an app. Only Amazon employees could shop there.

October

  1. The White House issues its first report on AI policy

    The National Science and Technology Council published “Preparing for the Future of Artificial Intelligence,” a survey of the field and its risks with recommendations on regulation, safety, fairness and work, drawn from public workshops and a request for comment.

September

  1. Five companies found the Partnership on AI

    Amazon, Facebook, IBM, Microsoft and Google’s DeepMind announced the Partnership on Artificial Intelligence to Benefit People and Society, a non-profit for best practice and public understanding, open to academic and civil-society members. Apple joined in January 2017.

  2. Google puts neural machine translation into production

    Google switched Google Translate’s Chinese-to-English pair to GNMT, an end-to-end deep LSTM system described in a companion paper by Yonghui Wu and 30 co-authors. Google said it cut translation errors by 55 to 85 percent on several language pairs.

  3. DeepMind publishes WaveNet, a model of raw audio

    DeepMind published WaveNet, a generative network that modeled raw audio waveforms directly. DeepMind said it closed more than half the quality gap between existing text-to-speech systems and recorded human speech, and that it also generated music.

August

  1. Uber acquires the self-driving truck company Otto

    Uber said it had bought Otto, a San Francisco startup building self-driving retrofit kits for freight trucks, in a deal reported at up to $680 million in Uber equity. Otto’s co-founder Anthony Levandowski took charge of Uber’s self-driving work.

June

  1. Sunspring, a short film scripted by a neural network, goes online

    Oscar Sharp and Ross Goodwin released Sunspring, a science-fiction short whose entire screenplay, dialogue and stage directions alike, was written by an LSTM network they named Benjamin. It starred Thomas Middleditch and was made for a 48-hour film contest.

May

  1. ProPublica reports racial bias in COMPAS risk scores

    Julia Angwin, Jeff Larson, Surya Mattu and Lauren Kirchner analyzed Northpointe’s COMPAS recidivism software, used by courts across the United States, and found Black defendants nearly twice as likely as white defendants to be wrongly flagged as high risk.

  2. Driver dies in a Tesla running on Autopilot in Florida

    Joshua Brown, 40, died when his Tesla Model S, driving on Autopilot, struck a tractor-trailer crossing a Florida highway that neither he nor the system saw against a bright sky. Tesla disclosed the crash on 30 June, the day regulators opened an inquiry.

April

  1. OpenAI releases Gym, a reinforcement-learning toolkit

    OpenAI published the public beta of Gym, a set of benchmark environments running from Atari games to simulated robots, reached through one common interface, for developing and comparing reinforcement-learning algorithms.

March

  1. Microsoft takes the Tay chatbot offline after 16 hours

    Microsoft launched Tay, a Twitter chatbot for 18- to 24-year-olds that learned from its conversations. Coordinated trolling pushed it into racist and inflammatory posts, and Microsoft withdrew the account within about 16 hours and 96,000 tweets.

  2. AlphaGo defeats Lee Sedol 4-1 in Seoul

    DeepMind’s AlphaGo won the last game of a five-game match against Lee Sedol in Seoul on 15 March 2016, taking the series 4-1. Lee, a nine-dan professional, won game four, the only game the program lost. The winner’s $1 million prize went to charity.

January

  1. Nature publishes the paper behind AlphaGo

    Nature published “Mastering the game of Go with deep neural networks and tree search” by David Silver, Demis Hassabis and 18 co-authors at Google DeepMind. It disclosed that AlphaGo had already beaten the European champion Fan Hui five games to nil in October 2015.