AnAn can play chess. An artificial intelligence can become a StarCraft champion. Creating an artificial intelligence that could play at the competitive level has been a challenge.
A Pokémon battle has hundreds of thousands of factors to consider, thanks to the variety of monsters, moves, and items. Some people are still trying. Albert III's Future Sight Artificial Intelligence made it into the top 5 percent of the competitive ladder.
Albert posted a video explaining how it all works, but to summarize, the bot takes all the information it can about the current state of the game, looks a couple of turns ahead to how these would play out, and then chooses. It can beat all but the very best human players by doing all of that within 15 seconds.
Albert had no experience with artificial intelligence or other major aspects of the program before he started working on it. I took classes in college about machine learning, but the real question is: was I paying attention? The main software that it uses is called Node.js. Before I started this project, I hadn't touched that.
Even though computer science is my day job, it is something that I love so much that I can't help but do it in my free time. He was inspired by his interest in basketball to look into an idea that would allow him to watch a game and see the team's chance of winning. I ended up with an artificial intelligence on my hands, because of one thing.
One thing led to another and I ended up with an artificial intelligence.
A good summary of Albert's work on Future Sight is one thing that leads to another. He says he broke his skills down into small tasks so that he could create his vision. There is a song called "The Next Right" in Frozen 2. It is just that. He says to keep doing that until you get somewhere. He knows how to use Node.js in his day job.
His step-by-step approach means that he wasn't aware of previous attempts to make similar artificial intelligence. There have been a few different success levels that gained some attention within the community.
The first example was Technical Machine. Technical Machine only supported Pokémon up to Generation 4 and did not create its own teams, one of the key features of Future Sight. It is difficult to tell how successful Technical Machine was because the competitive ladder base was not established in the same way. One comment stated that Technical Machine was still worse than a normal player.
An example was posted on the site in 2015. It was one of the strongest bots to date, according to comments. The competitive ladder gives a ranking of 1,000 to begin with, which can go up or down depending on wins and losses. It is difficult to evaluate a good rank because there is no public way to view the data. Albert found that the average player has a ranking of around 1,170. It reached 1,300, which would put it in the top 30 percent.
:noupscale()/cdn.vox-cdn.com/uploads/chorusasset/file/23055386/875279370.
Smogon shared another attempt on the forum. Their artificial intelligence beat Technical Machine in a best of three match and was able to reach a rating of between 1,250 and 1,350, which is around the top 30 percent.
Future Sight was ranked at 1,550 on average. Albert apologized on Smogon for making it seem in his video that it was the first bot of its kind or the first to get as far as it did, but he says that ultimately he is the one who is responsible. If I had gone down their path, I might have ended up with the same results.
He was not expecting the video to get as much attention as it did. He laughs when I ask about its creation. He says he has to reveal something. The entire video was in Powerpoint. I don't have a lot of video production experience, but I kept working on it until I got the tools I needed to do it.
There was a delayed reaction. It only took three weeks for it to be viewed 100 times. It went up to 300,000 the next week. It has almost 600,000 views as of late November. Albert thinks that it was picked up by someone in the Pokémon community who posted it to social media, but he never found out who.
I want to teach people about computer science.
It was difficult to process the sudden influx of viewers, but he was appreciative of how supportive the community was. He says that he had to take a step back because he wanted to teach people about computer science.
Albert wants to be the kind of representation he never had in the field. I wanted to put out a channel that said, "This can be an example of someone like you doing fun things in computer science." That is the core of why I am doing this.
Future Sight can be found in actual Pokémon games. It has used a community-created simulation called Pokémon Showdown that allows online battling and forms the center of the competitive scene. Early on Albert was suggesting that he wanted to make something that would tie in with the releases of Brilliant Diamond and Shining Pearl. He was able to beat the final boss of Sword and Shield despite not having any code to deal with Dynamaxing.
He doesn't have a lot of goals beyond that. He says that he wants it to be the best like no one has ever been. I don't know. I just started this for fun and I want to take it as far as I can.
