Playing good games should be a test of intelligence. Planning, strategizing, and learning moves and counter moves are all gaming skills we associate with complex human thinking.
Although many people claim that super-intelligent general AI (artificial intelligence) will take decades to achieve, major progress has recently been achieved in the gaming area. General AI basically means machines could perform any intellectual task humans can.
A long-standing challenge in the AI research community has been to build a game-playing system that could learn to play any game at a human level after only being provided with the rules. DeepMind’s work suggests this problem has been solved.
Chess and shogi (a Japanese chess-like game) are difficult for humans to master. Although the rules aren’t complicated, becoming a good player requires training, practise and innate reasoning skills.
Game-playing programs have been built that calculate good board positions and prudent moves. But those systems have been highly optimized for specific games.
Designing a system that can play any game is much more complicated and, until now, it’s eluded researchers.
Building a game-specific and competitive AI player is expensive and time-consuming. It requires expertise in both game-building and knowledge of the game being automated. DeepMind’s recent work on Alpha Zero eliminates this requirement. ‘Zero’ here refers to the amount of knowledge and expertise required to build a viable game-playing system. It’s the first example of a general game player that can play at a human level.
As such, it’s a harbinger of general AI.
What’s significant about Alpha Zero is the way it solves the general AI problem. DeepMind used artificial intelligence to build the game-playing AI. That means we have AI building AI – the possibilities are limitless.
Alpha Zero provides significant clues to building general AI systems for areas outside gaming. And that has profound implications for all aspects of human knowledge.
The race to general AI is important because of the economic potential of thinking computers. A computer that can think and reason like a person would have an enormous impact on every aspect of our world.
General AI would have the ability to understand language, generate speech, reason about what’s being said, solve problems, construct a coherent response based on context and recipient modelling, and respond with dialogue.
In other words, general AI would be able to do what people do.
A self-modifying general game AI system, like Alpha Zero, is an important step towards general AI. It’s as important as the transistor was to the early development of the electronics age. The companies that understand this and apply it will form the essential core of the AI industry, much like the early electronics industry that grew out of transistor technology.
This new field will open up huge opportunities for entrepreneurs, investors and even governments.
General game AI won’t have any immediately recognizable impact on our daily lives. But the ideas discovered in this process will soon begin to.
These concepts will infuse all emerging software systems. And they will significantly impact the way we work and interact in increasingly automated digital environments.
Bruce Matichuk is an AI consultant working in Edmonton on Health Gauge, an AI-based hypertension management system. Bruce is also an executive in residence at GO Productivity.
The views, opinions and positions expressed by columnists and contributors are the author’s alone. They do not inherently or expressly reflect the views, opinions and/or positions of our publication.