An introduction to Utility AI

Utility AI is a common technique for creating effective AI across a variety of game genres. By turning data about the current game state into numbers, we can allow our AI to make informed decisions about what it should do next.
Decisions are hard: 00:00
Utility AI: 01:13
Scoring actions: 03:04
Refining the process: 06:03
Text post: shaggydev.com/2023/04/19/util...
Further reading
An Introduction to Utility Theory - Game AI Pro - www.gameaipro.com/GameAIPro/Ga...
Improving AI Decision Modeling Through Utility Theory - www.gdcvault.com/play/1012410...
Building a Better Centaur: AI at Massive Scale - www.gdcvault.com/play/1021848...

[@Alicorn_]

How do you not have more subscribers? You're like if Game Maker's Toolkit actually gave useful game development advice.

[@TheShaggyDev]

lol

[@tobario]

Dude :D

[@ultimaxkom8728]

A shade so strong that it has a 95% chance of triggering an SSS-class hive retaliation. Luckily, I'm one of those 5%.

[@gourdbox]

Final Score, Strongly weighted: Like 100% Sub 100% Bell 100%

[@bossanova3494]

Love how broad your knowledge is with game dev stuff, I always find it really helpful

[@developersteve1658]

Awesome video. Thanks for providing more resources, too! What's cool is this approach lends itself very well to using ML which many people would consider "Real AI". Those weights could even be controlled by an orchestrator as described in the video by Ai and Games on Doom 2016's AI.

[@Kuijibo]

Man, your videos are exactly what I need. You also choose the best games as examples.

[@novicedraws8868]

i completely agree and I am implementing this to my game I think.

[@ethical_researcher4754]

Just started making AI system. This man reads my mind.

[@YJoker]

Bro this is very helpful!!

[@beidero]

Never heard the term "Utility AI" but it seems to be very similar to GOAP (just without the goals, action part would be almost the same). For GOAP you also end up scoring your potential actions and picking the best one. Based on this video to me it seems like Utility AI would be faster but less flexible than GOAP. Will consider trying this sometime in the future next time I make a game that would fit this kind of AI architecture.

[@TheShaggyDev]

Yep! Same general concept. There's a lot of instances like G.O.A.P. where a utility system is used for a different technique (because at some point you have to turn game state into something more consistent and usable), so picking a cutoff point for the video was a little tricky. I think Utility AI works great on its own for plenty of uses, but the concepts of scoring and manipulating game state to better capture desired behavior is a great foundation for a more complex AI system.

[@realMenta]

Thank you for the video! It looks like similar to GOAP. Pretty interesting!

[@IADaveMark]

Wow... a lot of this looked oddly familiar. Hmmm... wonder why? ;-)

[@TheShaggyDev]

Seriously, your talks are some of my favorites!

[@kishirisu1268]

It looks good at first, but what if we have 100 possible actions with 100 possible outcome per each action? NPC will stuck calculating utility scores, but if we want to prevent that - we return to the starting point - how to select what action to do in specific case without iteretion over each action, and come to the FSM (which will effectively reduce action range), but as you said FSM is bad so we need to use Utility.

[@eerywax5631]

What game is that at 1:59? It looks neat.

[@TheShaggyDev]

Symphony of War: The Nephilim Saga! Very Fire Emblem-y

[@eerywax5631]

Thank you!

[@loudmounthartist2257]

I’m so stupid but how did you get normalize number or score at 4:25

[@TheShaggyDev]

In that particular instance, there's no explicit normalization going on. I just multiply two numbers together that are already normalized to the range 0-1, according to the arbitrary rules of my sample logic that says Heal 25% is worth 0.4 and healing at 30% health is worth 0.8. As long as the numbers actually are

[@alejmc]

2:30 what is this game? Curious.

[@TheShaggyDev]

That's Fallout 1!

[@alejmc]

@@TheShaggyDev oh wow, those gems that I completely missed, better put these few days off during the holidays to revive some of that!