Razer Project Ava: would you pay an AI to help you get good at games?

Gaming hardware manufacturer Razer is introducing what might be its most controversial “project” yet — an “AI gaming copilot” designed to help you get better at video games. With your permission, it takes thousands of pictures of your screen, then tells you how to play a game, optimally in real time, as you continue to play.

“By analyzing millions of simulations within seconds, I’ll always have an answer to your toughest gaming challenges,” a Razer marketing video claims.

For example, watching a prerecorded video of a punishing Black Myth: Wukong boss fight, Razer’s AI assistant had all sorts of tips:

• “Get ready to dodge when his blade spins or glows with an orange tinge.”
• “Keep a close eye on his health. Once you shave off 10 to 20 percent get your dodge fingers ready.”
• “He’s going to vanish...”
• “If he grabs you, you will feel it!”

I felt weird about this almost immediately. While it could be more convenient than looking up a guide, it doesn’t credit or compensate the creators of the guides that Razer ingests in order to train its AI. It would be pretty disruptive if Ava spent time telling me that Verge sister site Polygon crafted that guide, after all!

And yet, Ava did sound pretty disruptive regardless, seemingly interrupting the game’s audio to tell the player what to do. I suppose you might only summon Ava when you need help, but it still feels like a bit of a weird fit for Razer, a company that’s long associated itself with elite gamers. (There’s also a long conversation about how “female” AI can perpetuate harmful stereotypes, but at least Razer global marketing director David Ng tells us it’ll offer other voices in the future.)

Next, we saw an actual live demo of Ava helping someone play League of Legends, acting as an AI chatbot that could help you figure out what to do and which spells and items to equip based on enhanced situational awareness of the game, API calls, guides, and even potentially historical data about the outcomes of matches played by esports teams.

Ava knows where the enemy’s champions are because it’s taking pictures of the mini-map, and it knows what you might want to use to counter them — though its answers were delayed by multiple seconds in the prototype we saw, which ran on a pair of local Nvidia RTX 4090 laptop GPUs using Meta’s Llama 3.2 LLM instead of in the cloud.

Following the match, Ava attempts to continue to act as a coach, creating replays, pointers, and feedback for you about your performance.

It’s clear that there’s a lot of work that would need to be done to make this useful, particularly if Razer wants to fulfill some of its other dreams. (It imagines Ava could help you auto-configure your computer, act as an autonomous gaming companion, and serve as a raid leader if no one wants to spend their time organizing the group.)

But unlike many of Razer’s concepts, which it never guarantees to turn into products, it seems the company is already invested in Ava and is thinking about an Ava service as a new business opportunity. It’s not waiting for feedback on the idea before it rolls out a beta, and Ng tells us it’s building out a whole team of AI developers to work on such ideas, with a planned bigger announcement at GDC in March.

The company says it’s got a patent-pending algorithm on the way it figures out how to suggest gaming tips and is working on some sort of “proprietary AI hardware” to help it run. It’s thinking about how it can beat the competition with the cloud service — because yes, there is competition for AI gaming apps that coach you to play League of Legends. I just heard about another one yesterday.