DeepMind, the Alphabet Inc-owned artificial intelligence (AI) company best known for creating software capable of beating the world’s best go players, has targeted StarCraft II as its next big research milestone, but so far, space is proving a difficult frontier for the company’s algorithms.
DeepMind’s algorithms, including those that performed with superhuman skill across a host of classic Atari titles, “cannot win a single game against the easiest built-in AI” in StarCraft II, let alone challenge skilled humans, the company said in a blog post on Wednesday.
The built-in agents, which are created by StarCraft publisher Activision Blizzard Inc, use hard-coded rules to determine their gameplay rather than the kinds of advanced machine learning techniques in which DeepMind specializes.
Photo: EPA
The company said new breakthroughs in machine learning would be required for its software agents to master the game.
The post did not reveal how close DeepMind might be to such breakthroughs.
The algorithm that mastered the Atari games was unveiled in June last year. Since then, DeepMind has published a number of research papers that hint it might be closing in on creating software capable of many of the tasks — such as prioritizing goals, long-term planning and memory — that any system would need to play StarCraft II successfully.
The company said that its algorithms performed well at learning some basic steps, such as moving around the game environment and selecting units, that would be critical to mastering the game.
StarCraft is considered an important target for machine-learning researchers because, unlike go, in which both players can see the entire board and take turns moving pieces, players in StarCraft cannot see what is happening in the entire game environment at one time and both players move their units simultaneously.
The game also requires players to carry out sub-tasks, such as building structures and mining resources, while also conducting reconnaissance, mounting attacks and defending territory.
To succeed, a player needs to have a good memory, prioritize among tasks and plan under conditions of uncertainty.
Because of these factors, StarCraft II comes much closer to approximating many real-world situations than games such as chess, go or even poker.
StarCraft II is also used in e-sports competitions, so there are highly skilled human opponents with which an AI can match wits.
To help computer scientists use StarCraft II as a testbed for AI, DeepMind has partnered with Blizzard to create an interface that allows outside software to access and play the game.
The two companies this week unveiled the interface, along with a set of tools to help other computer scientists train AI agents to play the game, at a machine learning conference in Australia.
Among the tools Blizzard is making public are a dataset of anonymized game replays — recordings of humans playing the game — that computer scientist would be able to use to help train their systems.
“One technique that we know allows our agents to learn stronger policies is imitation learning,” DeepMind said in its blog post. “This kind of training will soon be far easier thanks to Blizzard, which has committed to ongoing releases of hundreds of thousands of anonymized replays.”
DeepMind also said it was releasing a series of “mini-game” environments that would help researchers train their AI agents on basic components of the game.
It follows efforts by other companies, such as Facebook Inc and Microsoft Corp, to open game environments, including the original StarCraft and Minecraft, to the entire AI research community.
SECOND-RATE: Models distilled from US products do not perform the same as the original and undo measures that ensure the systems are neutral, the US’ cable said The US Department of State has ordered a global push to bring attention to what it said are widespread efforts by Chinese companies, including artificial intelligence (AI) start-up DeepSeek (深度求索), to steal intellectual property from US AI labs, according to a diplomatic cable. The cable, dated Friday and sent to diplomatic and consular posts around the world, instructs diplomatic staff to speak to their foreign counterparts about “concerns over adversaries’ extraction and distillation of US AI models.” Distillation is the process of training smaller AI models using output from larger, more expensive ones to lower the costs of training a powerful new
Shares of Taiwan Semiconductor Manufacturing Co (TSMC, 台積電) have repeatedly hit new highs, but an equity analyst said the stock’s valuation remains within a reasonable range and any pullback would likely be technical. The contract chipmaker’s historical price-to-earnings (P/E) ratio has ranged between 20 and 30, Cathay Futures Consultant Co (國泰證期) analyst Tsai Ming-han (蔡明翰) told Central News Agency. With market consensus projecting that TSMC would post earnings per share of about NT$100 (US$3.17) this year, supported by strong global demand for artificial intelligence (AI) applications, and the stock currently trading at a P/E ratio of below 25, Tsai said the valuation
The artificial intelligence (AI) boom has triggered a seismic reshuffling of global equity markets, with Taiwan and South Korea muscling past European nations one by one. With its stock market now valued at nearly US$4.3 trillion, Taiwan surpassed the UK, Europe’s biggest market, earlier this month, data compiled by Bloomberg showed. South Korea is about US$140 billion away from doing the same. The tech-heavy Asian markets have shot past Germany and France in the past seven months. The shift is largely down to massive gains in shares of three companies that provide essential hardware for AI: Taiwan Semiconductor Manufacturing Co (TSMC, 台積電),
The US Department of Commerce last week ordered multiple chip equipment companies to halt shipments of certain tools to China’s second-largest chipmaker, Hua Hong Semiconductor Ltd (華虹半導體), its latest action to slow the country’s development of advanced chips, two people familiar with the matter said. The department sent letters to at least a handful of companies informing them of restrictions on tools and other materials destined for two Hua Hong facilities US officials believe make China’s most sophisticated chips, the people said. Top US chip equipment companies Lam Research Corp, Applied Materials Inc and KLA Corp, each of which has significant
RSS