Google is known to develop quirky little research projects on the side-burner of its larger businesses. While many of those go unnoticed, some simply aren’t avoidable, like this new Artificial Intelligence machine (via Bloomberg) which beats the hell out of you when it comes to Pinball.
This new AI developed by Google does nothing but one thing, it just keeps on playing video games to master them and hence learns so much as to no human at least can defeat it. It’ll be interesting to see a Google AI v/s Google AI match-up though, wouldn’t it ?
The company introduced the new development in machine-learning technology on Wednesday, describing it as “the first significant rung of the ladder” to building intelligent AI, that can do some actual AI work, i.e. figuring out on its own as to what it needs to do.
The project isn’t indigenous to Google though. It has in fact been built by a London-based AI startup called DeepMind Technologies that Google acquired last year. The research project was built by exposing computers running general AI software to retro Atari games. Those general AI machines were shown some 49-0dd games on the Atari 2600 (Dad’s love) and were then asked to play those games without any sort of instructions.
So what happened next ? Well look at these points :
- The new Google AI machines surpassed human performance in 29 games.
- It outperformed the best possible algorithmic methods to win in 43 games.
- And more so, it crushed human players with a system that was more than 20 times better than a professional human game tester
But the goal of this project isn’t obviously to beat you at video games. Google’s projects are always aimed to implement something in the real-life scenarios. If an AI as intelligent as this one, could be implemented in systems which require a 100 percent accuracy, we could possibly come up with no-error mechanism.
Looking from Google’s own perspective, its ambitious self-driving car project could learn on itself, based on its traffic handling experiences as to what speed limit or what cautions it need to take care of while driving on even the most busiest of roads. Not only Google’s car, a system like this, if implemented in other heavy vehicles, could easily save Millions of lives we loose on road accidents.
Demis Hassabis, a co-founder of DeepMind and VP of engineering at Google says
[This research marks the] first time anyone has built a single learning system that can learn directly from experience and manage a wide range of challenging tasks
Though those above mentioned prospects of using AI in scenarios sounds lucrative, Google itself is far from adapting them to such scenarios. Hassabis says.
It’s mastering and understanding the structure of these games, but we wouldn’t say yet it’s building conceptual knowledge or abstract knowledge. The ultimate goal here is to build smart, general-purpose machines, but we’re many decades off from doing that.
Google is currently taking baby steps to achieve a better AI with these machines. It is now looking to upgrade this current AI to help it understand and learn three-dimensional worlds contained within games. That sounds exciting ! We’ll be waiting to see how that turns out !
