Featured Image for Artificial Intelligence has mastered Ms. Pac-Man, will come for humanity next
Gear

Artificial Intelligence has mastered Ms. Pac-Man, will come for humanity next

First, the machines beat us at chess, checkers and poker.

More recently, they destroyed us in Go, an ancient game once thought too complex for AI to get its head around.

Now, the latest game to feature in the emerging battle between AI vs humans is Ms. Pac-Man.

Artificial intelligence developed by Maluuba, a Microsoft-acquired Canadian company, has beaten all 256 levels of Ms. Pac-Man and achieved a perfect score of 999,990.

For perspective, the best score that mere mortals have achieved on a 2600 is just 266,330. Humanity just got rekt.

Ms. Pac-Man was released in 1982, two years after the original Pac-Man. Although the games are very similar, Ms. Pac-Man is more difficult. This is largely because the creators added random movements to the ghosts to prevent players from memorising patterns, as they could do in the first game.

Unlike in Ms. Pac-Man, a handful of amazing humans have been skilled enough to capture the perfect score of 3,333,360 at level 256 of Pac-Man.

This feat was first achieved by gaming legend Billy Mitchell, who is one of the protagonists in King of Kong: A Fistful of Quarters, a brilliant documentary about the high stakes and nerdy world of arcade gaming.

Although Pac-Man was meant to have no ending, level 256 is what’s known as a “kill screen”. Errors in the game’s code make it literally unplayable, so everyone has to stop right there. Ms Pac-Man has similar problems, but through the magic of emulation, players have been able to fix it.

In order to get a perfect score in Ms. Pac-Man, the team at Maluuba used something called “Hybrid Reward Architecture.”

Instead of having a single AI tackle the tricky task, researchers created more than 150 “agents” that could each focus on a single component of the game.

For example, some agents worked on the fruit bonuses, another on the ghosts and another on the little pellets Ms. Pac-Man eats. The agents then reported their findings to a “master AI”, which aggregated the suggestions and decided on the best course of action.

Doina Precup, an associate professor of computer science at McGill University in Montreal told the Microsoft Blog that the perfect Ms. Pac-Man score was a significant achievement among AI researchers who have found the game a tough to crack.

“This idea of having them work on different pieces to achieve a common goal is very interesting,” she said. Precup added that it’s similar to some theories of how the brain works and could have broad implications for teaching AIs to perform complex tasks with limited information.

Harm Van Sijen, a research manager at Maluuba and lead author of a paper about the achievement, told the blog that the best results were achieved when each agent acted egotistically and really honed in on its task.

“There’s this nice interplay between how they have to, on the one hand, cooperate based on the preferences of all the agents, but at the same time each agent cares only about one particular problem,” he said. “It benefits the whole.”

So does this all mean that the robots are gonna take over?

Not anytime soon.

At the moment, we are surrounded by machines that possess what computer scientists call “ANI” (Artificial Narrow Intelligence), which is the ability to master specific tasks.

Beating us at games is an example of ANI and isn’t something we should be particularly worried about. Artificial General Intelligence (AGI) on the other hand, is a different story. When the machines get to this level – which is competent in completing any human intellectual task and more – we might have some problems.

When will that happen?

According to the experts, the estimates for AGI emerging range from 2030 to never, with most believing that it will happen by either 2030 (43%) or 2050 (25%).

Things like chess and checkers may seem trivial, but in the long run, the progress made now will no doubt have an impact on getting AI to the AGI level. In case you were wondering, there is a step after AGI too, and it’s called Artificial Super Intelligence (ASI). At that point, we will be dealing with something well beyond what we can comprehend. Will it be nasty or nice? We aren’t sure.

If the AI does turn nasty on us, we might look back and curse the day that we got our asses kicked in Ms. Pac-Man.

About the author

Stefan is an Adelaide-based freelance writer. In his spare time, he plays tennis badly, collects vinyl and brushes up on his Mandarin. Follow Stefan on Twitter

Leave a comment