Op-Ed: Adaptive AI gaming learns as it plays, and so will the world

The blurry verbose world of AI news is looking both promising and dangerous for AI-based gaming at all levels. AI is now getting real traction in the fussy and ultra-skeptical world of game development, but it’s coming with very mixed messages.

Gamers are waiting for the Next Big Thing. The good news is that gaming is a truly murderous and thankless environment for AI slop. AI that delivers interest and fun will work. That doesn’t mean that “AI for the sake of AI” is any sort of metric for acceptance in the market.

The really big and inescapable issue is how much needs to change. AI will redefine gaming in any number of ways. Whether that’s good or not is another matter.

The fizzy sparkly do-nothing-much-at-great-expense corporate gaming motif is utterly loathed. Layoffs of people who’ve delivered good games isn’t adding credibility to the sector, either. Those good games are “replaced” by a vacuum of cut-and-paste versions of themselves. Gamers may have no idea who the devs are, but they can see what’s gone missing.

Lack of ideas in gaming is usually seen to be at Hollywood level. Endless expansions can get as tiresome as churned-out movie franchises. No actual movement, just more of the same. It’s worse when so many games use the same effects and the same repetitive rubbish.

Adaptive AI is the next big thing, whether anyone likes it or not

Adaptive AI learns as it plays. It learns directly from players. That is a huge tectonic shift in gaming, and it will revolutionize gaming. There’s a good intro to adaptive AI from HP, spelling out the history of adaptive AI, what it does, and how it does it.

Progressive learning also impacts players who have to keep learning. This is closer to the old “I beat the computer” idea than ever, and let’s face it, it’s a meaningful challenge. Winning a good impossible-level game is well worth doing.

A bit of “hopefully brief” backstory is required to explain the importance of adaptive AI.

One of the primary issues with conventional gaming is scripting. Games typically offer scripted alternatives in various forms. Games and game storylines are charted based on player choices as the game progresses. This is also directly related to basic coding. “If choice 1, go to list.” It’s Snakes and Ladders revised.

Players can feel, and can actually be, confined by lack of other choices. It detracts from the gaming experience. Experienced gamers may eventually blink at a very familiar game narrative path, but how fresh can it be?

Sandbox games were the first moves out of this Snakes and Ladders environment. Adaptive AI will free up games from this restrictive approach. More interactions and different behaviours are possible.

This sort of AI should also feed directly back into development. It’s not hard to visualize an AI that can detail any sort of dysfunctions and glitches, as well as bugs.

Machine learning is the natural working dynamic and the definition of how adaptive AI has to function. Hardware is another parameter. System specs will be critical for higher-end AI until the market tech plateaus into a common series of specifications.

Does any of this sound even slightly avoidable to you?

What if the adaptive AI screws up?

This is the question that all forms of AI must face with every single operation. Machine learning is often a grinding and very demanding process. Learning either meets required levels, or it doesn’t. AI agents either meet cutoffs or not. Adaptive AI is the full process in play.

There’s a big difference in learning gaming, though. Different players directly impose different parameters on the learning process. What beats Player 1 may not beat Player 2.

MMOs could become real tournaments, with thousands or millions of humans and AIs vs other previously unencountered AIs and humans. Multiple adaptive Ais could have to fight each other, based on their learning with no prior experience of play modes of other players.

This is where it gets genuinely interesting. The issue is depth of learning. If you’ve ever played Stockfish, you’ll know that depth of analysis is a key to how the AI is working or not working. Stockfish indicates its depth of analysis in levels. 3 is very low, 40+ is very high. At the lower levels, it’s losing.

That sort of metric will be an indicator of adaptive AI’s running logic. How many moves or options are available? Let’s say 100 options can be played, but the analysis level is 5. Either the AI has excluded most options, or it’s running out of options.

If the AI screws up, you can see the entire process. You can already hear the devs going back to core learning.

DIY gaming and adaptive AI

Largely because machine learning is so critical to AI as a whole, adaptive AI is pretty basic for gaming as a class of capabilities. AI agents learn by necessity. Adaptability is likely to become a fixture in future gaming simply because it’s necessary and delivers a more flexible gaming experience.

DIY gaming is taking hold, and adaptive AI has to be part of the mix. In China, some people with no gaming development experience at all have developed games with a prompt or two. That’s exactly where game development is going, and adaptive AI is coming along for the ride.

Don’t be too surprised to find yourself in a few years doodling around with an adaptive AI making the game you’ve always thought was a good idea. Picture this scenario. How dare it win the game you designed? Your creative instincts have woken up.