If you were to make a list of the greatest video games ever made, the original Super Mario Bros. made for the Nintendo System would have to be towards the top of that list. Thanks to original game mechanics and challenging level designs, Super Mario Bros. was an instant classic and remains popular today—30 years later. But retro-gaming enthusiasts aren’t the only ones still playing Super Mario Bros. As it turns out, the classic video game is a huge hit with artificial intelligence developers. That’s because they often use the video game to test the intelligence of their systems.
How the test works
Artificial intelligence researchers use Super Mario Bros. to test AI in a couple of different ways. The first is that they attempt to program an AI system to successfully beat the game from beginning to end. To do this, the AI system needs to learn the nuances of the game in order to time jumps perfectly, avoid enemies, and successfully reach the end of each level before time expires. From 2009 to 2012, AI researchers even held a Mario AI competition to where competitors tried to design AI that could complete the game in the fastest time.
AI researchers are also programming artificial intelligence to design playable levels that will provide players with a challenge without being too difficult. Researchers believe that artificial intelligence may soon be used to assist humans with level design.
Why Super Mario Bros.
You may be wondering, out of all the ways to test an AI system, why use a video game. As it turns out, video games are an ideal way to test artificial intelligence because they require the use of logic, creativity, situational awareness, and decision-making skills to successfully beat the game. All of these skills are necessary in the quest for true artificial intelligence. But of the tens of thousands of video games to choose from, why Super Mario Bros.? AI researchers cite two primary reasons: first, they just love to play the game like the rest of us and second, the classic platformer game’s perfect mix of complexity and simplicity combined with finely-tuned mechanics make it an ideal game for artificial intelligence testing.
While early AI systems used Atari games for testing, it’s too simplistic today thanks to advancements in artificial intelligence. But side-scrolling games like Super Mario Bros. present a bit more of a challenge because a good percentage of the level design can’t be observed by AI at any given time.
Facebook is a relative newcomer to the artificial intelligence race. IBM’s “Watson,” Apple’s “Siri,” and Microsoft’s “Cortana” all have a head start but Facebook is working hard to catch up. Their newly established artificial intelligence team consists of less than a hundred people for now but that number is expected to grow drastically in the coming years. Here’s a look at some of the human-like skills that Facebook’s iteration of artificial intelligence is already starting to master.
One of the challenges in creating artificial intelligence is getting a computer to comprehend spoken language like humans do. Facebook’s artificial intelligence uses natural language processing similar to Apple’s “Siri” to understand and respond to questions or requests posed by humans. So far, the questions and requests it can respond to are quite limited but it’s progress.
Facebook’s artificial intelligence is also learning how to read. So in addition to comprehending spoken language, it’s also able to get meaning out of words on a page. Not only can it “read” a book, such as Lord of the Rings, but it is able to search within its data bank and answer questions about books it has read.
In the past, Facebook has relied on Bing’s translating tool to automatically translate posts and comments in a foreign language so that users can understand what those with different languages are saying online. Now, Facebook is working on its own real-time translating prototype. These auto-translator are often very inaccurate so they have a team of testers providing input and grading it’s accuracy in translating so that it can learn and improve.
Facebooks already has auto-captioning. It can see a two-dimensional image and automatically tag people in the image using facial recognition software. But the software Facebook is working on now will go a step further. It will be able to read aloud the names of people in photos which will be helpful to blind people who are commenting on photos in their newsfeed.
Facebook can do more than identify people in images. Facebook can also recognize a wide range of objects and identify them correctly. It can event differentiate between hundreds of animal species and correctly identify them, something that most humans can’t even do.
Finally, Facebook can watch sports footage and identify the sport being played in the video. This goes beyond just identifying something isolated in the footage like a baseball glove. Facebook’s artificial intelligence had to learn the rules of various sports and be able to recognize them in action in order to be able to correctly identify the sport.