The quest to create an AI capable of playing competitive Pokémon

An AI can beat a chess grandmaster. An AI can become the StarCraft esport champion. But create an AI that could play Pokemon at the competitive level has been a more elusive problem.

Thanks to the variety of monsters, stats, moves and items, a Pokemon battle has hundreds of thousands of factors to consider for any player – or machine -. But that hasn’t stopped some people from trying. More recently, Future Sight AI, created by computer scientist Albert III, successfully climbed into the top 5% of the competitive scale.

Albert posted a video explaining how it all works, but to sum it up, the bot gathers all possible information about the current state of the game, extrapolates the possibilities for any turns it might take, looks a few turns ahead to find out how those- this would take place, then chooses the option that can lead to the greatest number of better results. By doing it all in 15 seconds, round after round, he can beat all but the best players.

That’s pretty impressive, especially considering that Albert had almost no experience with artificial intelligence or other major aspects of the program before he started working on it. “I took courses at university on machine learning, [but] the real question is: was I paying attention? he’s laughing. “The main software it runs on is called Node.js. I had not touched it at all before embarking on this project.

“Even though IT is my daily job, it’s something I love so much that I can’t help but do it in my spare time too,” he says. This passion, combined with the pandemic boredom, prompted him to ponder an idea that was first inspired by his interest in basketball. “[Some websites] would do this thing where you could watch a game and see the team’s current odds of winning, and I thought about doing that for Pokemon, “he says.” Then one sort of thing resulted in another, and then I ended up with an AI on my hands. “

One thing leading to another is a pretty good summary of Albert’s work on Future Sight AI. He says he wanted to learn new skills and just broke them down into small enough tasks until he was able to create his vision. “It’s such a bad reference but there’s this song in Frozen 2, called ‘Next to the right [Thing]. ‘ It’s just that. Keep doing this until you get to somewhere, ”he said. Nowadays, for example, he is so familiar with Node.js that he can also use it in projects in his daily work.

His step-by-step approach means that he was in fact unaware of previous attempts to create similar AIs. Past projects are not as well documented as Albert’s, although there have been a few varying levels of success that have caught the community’s attention.

A first example was Technical machine, first created in 2010. Although it was updated until 2019, Technical Machine was never fully supported Pokemon until Generation 4 and did not create their own teams, one of the key characteristics of Future Sight. Also, at the time of its release, the basis for the competitive ladder was not set in the same way, so it’s hard to say how well Technical Machine was successful overall. A Reddit comment, however, stated that “Technical Machine at its smartest level was even worse than a normal player.”

Another example was posted on Reddit in 2015 by a user who went through onmabd. Comments said it was “one of the most powerful robots to date.” The competitive scale gives players a ranking of 1,000 to start with, which then goes up or down based on wins and losses. There is no public way to view the data, and it changes over time, so it’s difficult to gauge what a “good” ranking is. However, during his creation process, Albert found that the average player rating was around 1170. Onmabd’s AI managed to reach 1300 which would put him in the top 30%.

Pokémon European International Championships

Photo by John Keeble / Getty Images

More recently, a user on Pokemon Smogon community forum via shared pmariglia another attempt. Their AI beat Technical Machine in a best of three and was able to score between 1250 and 1350 – again, around the top 30%.

Future Sight AI scored an average of 1,550 in testing. Although Albert apologized on Smogon “for making believe in my video that [Future Sight] is the first bot of its kind or the first to go as far as it has’ (in addition to detailing where the two projects take different approaches), he says that in the end, he’s happy to not knowing that other people had already tried his project. “I don’t know why I never thought to inquire [but] if I had followed their path, I might have gotten the same results, ”he says.

He also didn’t expect the video to get as much attention as it did. For starters, when I ask about his creation, he laughs. “I have to reveal something,” he said. “This entire video was animated in Powerpoint. I have to say I don’t have a lot of video production experience. [so] I had an idea of ​​what I wanted the video to look like and kind of kept working on it until I could get the tools I knew how to use to do it.

Then there was the late reaction. Posted in July, it was only seen a hundred times in its first three weeks. The following week, it jumped to 300,000. (At the end of November, it’s almost 600,000 views.) Albert thinks it was picked up by someone in the Pokemon community who posted him on Twitter, which blew him up, but he never knew who.

He says it was difficult to deal with the sudden influx of viewers, but appreciated the support the Pokemon the community was. “I kind of had to take a step back because the point of what I’m doing is that I want to teach people about computer science,” he says.

In particular, as a black man, Albert wants to be the kind of portrayal he’s never had on the pitch. “I thought I had experience in public speaking, I like doing projects that people might find interesting, so I really wanted to publish a channel that said, ‘This can be an example. of someone like you doing fun computer things. This is really the reason why I am doing all of this.

For now, his goal is to make Future Sight playable in reality. Pokemon Games. So far, it has used Pokémon Showdown, a community-created simulator that allows for online battles and is functionally the center of the competitive scene. But very early on, Albert hinted that he wanted to do something that could be related to the exits of Brilliant diamond and Shiny pearl. Most recently, he managed to beat the final boss of Sword and shield, despite the lack of code to manage Dynamaxing, which is prohibited in common competitive environments.

Beyond that, it does not have too many concrete objectives. “I mean it’s such a cheesy thing, but I want it to be the best like no one has ever been,” he said, echoing the old one. Pokemon anime theme melody. “But seriously, I don’t know. I just started this for fun and want to go as far as I still find joy in doing.


Please enter your comment!
Please enter your name here

Trending this Week