A Microsoft Research team challenged PhD students to craft an advanced AI that can play Minecraft, offering three $20K prizes. Here's how it could represent a new direction for AI research.
AI has achieved milestones in mastering games like chess, Go, and, recently, poker—illustrating how successful machines have become at completing specific, narrow tasks. But can AI move beyond the narrow, toward achieving more general, human-like skills?
On Tuesday, Microsoft launched a competition to address this question. Microsoft's Project Malmo, which the company calls a "sophisticated AI experimentation platform," brings researchers together to use Minecraft as a testing tool for developing AI—smart, collaborative AI that can compete in a virtual world.
The Malmo Collaborative AI Challenge asks PhD students to enter this world and create AI that can team up with randomly assigned players to compete for a high score in Minecraft. At stake? Three Microsoft Azure for Research grants, at up to $20,000 value each, as well as three placements at Microsoft's AI Summer School in Cambridge, UK.
Contestants will use Project Malmo, which provides "code that helps AI agents sense and act within the Minecraft environment," according to Microsoft. The program can run on Windows, Linux, or Mac OS, and can be adapted for any program language. The goal is to use deep reinforcement learning, cognitive science, and other AI tools to complete the challenge.
This development represents an important step forward in AI research, away from mastering tasks like voice,image recognition, and translation, and towards a more nuanced and collaborative AI that adopts a human-like method for decision-making.
Unlike games that involve a limited number of available "moves," Minecraft offers players the ability to choose from an endless number of tasks, ranging from simple actions (like walking around in search of a treasure) to more complex tasks (like constructing a building), that rely on the collaboration of the team.
The challenge would try to "solve" another game through AI. Recently, Carnegie Mellon's Libratus, a poker-playing AI, achieved a big victory against four of the world's top poker players. Many experts saw the achievement as a major milestone in AI, since poker also involves complex, intuitive skills that have been difficult for a machine to master in the past—as well as its ability to predict an opponent's moves with incomplete information.
So what does AI in games have to do with its success in the real world? As AI becomes good at these skills, it can translate to becoming adept at other decision-making skills—whether in business, politics, or at home.
It's also important to note that AI still has limitations—even the poker victory was only a two-person matchup. Also, as Toby Walsh, AI professor at the University of New South Wales notes, this is a "very, very small step down the road to general AI."
"Minecraft is playing the role of a very simplified real world," said Walsh. "It's more challenging than the completely artificial worlds that Google's DeepMind play in, with games like Go and PacMan. But it is still a long way from the complexity and openness of the real world where at any moment you might be eaten by a tiger, run over by a truck, or amused by a joke. "
What's notable here, said Walsh, is the interactive component, which "raises lots of challenges that humans meet in the real world, of cooperation and competition. Any artificial general intelligence (AGI) will need to learn to compete and to cooperate just like we do."
Marie desJardins, AI professor at the University of Maryland, Baltimore County, sees Minecraft as an "interesting and challenging problem for AI systems, because of the fundamental complexity of the game environment, the open-ended nature of the scoring system, and the opportunity to collaborate with other game players (AIs or humans)."
But desJardins also raises concerns when it comes to these competitions. "Who owns the resulting intellectual property?" she asked. "Are these kinds of contests the best way for grad students to spend their time? Do these competitions foster or decrease diversity? Who ultimately profits from the contests?"
Whoever does profit in the end, it is likely that efforts like these, aimed at achieving AI that can master human skills, represent an inching towards an AI that could, in some domains, surpass human intelligence.
- How Google's DeepMind beat the game of Go, which is even more complex than chess (TechRepublic)
- Can Google's DeepMind beat world Go champ? Watch live this week (TechRepublic)
- IBM Watson: The inside story of how the Jeopardy-winning supercomputer was born, and what it wants to do next (TechRepublic)
- Google AlphaGo AI clean sweeps European Go champion (ZDNet)
- Google AI gets better at 'seeing' the world by learning what to focus on (TechRepublic)
- How Google's AI breakthroughs are putting us on a path to narrow AI (TechRepublic)
- Smart machines are about to run the world: Here's how to prepare (TechRepublic)
- Google AI beats humans at more classic arcade games than ever before (TechRepublic)