[spacemarine1]

AI tech has the spotlight right now for sure amongst VC’s. But I believe AI is also a huge tailwind for video games.

Namely, small, clever teams will be able to do big and entertaining things that were not possible before.

(But yes there will also be tons of slop.)

[conartist6]

It can't be a tailwind for a whole industry. It can only take from some and give to others.

[spacemarine1]

The industry has experienced slowed growth and is restructuring which is scary for many people right now. There have been continuous layoffs and studio closures. That sucks (though is not all caused by AI).

But a few quick thoughts: Video games have always been about cutting edge technology. Because they are interactive, they are best positioned to leverage AI tech. (unlike static media)

Prototyping tends to be the most important but most neglected process for finding the fun. AI is a catalyst for rapid prototyping such that a studio can more quickly build and assess game loops and de-risk the rest of their dev cycle before staffing up or pulling team members off other projects.

AI may long-term create more leisure time macro-economically for everyone, meaning more consumer time that may be consumed playing games. (Owen Mahoney thinks the industry will soon triple in size)

[mrguyorama]

>Because they are interactive, they are best positioned to leverage AI tech.

You have not supported this argument.

On the contrary, I think LLMs are not a big help.

We've already been here. Procedural generation was the magical solution that was supposed to help a single dev make giant worlds. And it did. It helped you make giant, utterly empty and soulless worlds.

Great for minecraft. Not so great for No Mans Sky, as there were significant limitations. Useless for anything that depends on a story or immersion in that story or characters.

This idea that you can wire an in game character up to an LLM is misguided and doesn't seem to understand what players want a character to be.

In Mass Effect 2, a fan favorite character infamously had little content for most of the game. Garrus was a very loved character, including being the love interest for a lot of people who played the game, but during 2 he just sits in a part of your ship and says "Sorry, can't talk, running some calibrations" for almost the entire game.

Put the entire game's script into an LLM as context and have it pretend to be Garrus and try and talk to it. Will the LLM take a strong stance that Garrus would take? Will it correctly figure out that Garrus would rather kill a bad guy than let him get away, and then have him make convincing arguments about that in reference to the previous mission, and then in the next mission put you in a situation where you have to decide whether Garrus is right and whether you should dome that "Bad Guy" instead of letting the police fail to aprehend him? Will that Garrus make you think about this world and your place in it and whether your personal or chosen morals are right?

Probably not.

People don't want to chat about the weather with random NPCs. People want characters that have character notes and integrate into the story and make you feel.

So far LLMs can't really write that well, and certainly are not cohesive and multidisciplinary enough to be able to build that into a game in a convincing way.

There's a famous game called "Facade" which plays up this big fanfare about how it uses "AI" and Natural Language Processing to bring two characters to life for you to talk with and navigate a crisis with, but it's almost entirely lies. The actual logic of how it works is almost identical to old fashioned text adventure parsers in their heyday. It's a heavily scripted sequence, with fairly few actual paths it can take, and the script is not that big. It got so much press for what was basically done in the 80s. I think some people have tried to hook LLMs into it, but it just doesn't feel good. The problems that AI dungeon adventure game always had still exist, just under more paint.

[spacemarine1]

One EGG studio is wiring LLM’s to npc’s in a hilarious way. We’ll be curious to see everyone’s feedback when they make a playable available.

However, I’ll say that as a result of their AI integration they are also doing way more human writing in the form of prompts and other procedural elements than if they just used old fashioned dialog trees.

I think AI can only be used as an enhancement in certain specific and controlled ways.

One mistake I see a lot right now is the assumption that you can delegate design and creative direction to AI. I think that generally yields slop. In fact I think the Creative Director/design role at a game studio may be the hardest digital job for an AI to replace. I had the opportunity to express that idea to Sam Altman once and he did refute it.

[conartist6]

1. "More leisure for everyone" is utter bollocks. We know it doesn't work like that. If it did we would all be doing nothing but leisure because of how much leisure we gained by switching to email.

2. Games are just code that's fun. How does this sounds as a process for making something fun: "Start by de-risking." Hmmm OK, yes, this tracks with my experience of private equity companies being the most innovative and successful creators of games.

[zmmmmm]

> "More leisure for everyone" is utter bollocks.

it's fascinating how this delusion persists.

More leisure will only occur if it becomes true that the marginal return of doing additional work falls almost to zero. So to say people will have more leisure time actually suggests more that they will lack opportunities to do things of value than it does that they will choose to have more leisure time. Which is depressing.

[conartist6]

If everyone could live a life of leisure, the logical end result would be roving gangs engaging in street warfare for funsies. If money is no object for anyone, what else is there to do but to seek fame, even infamy?

Each gang member could be running their AI value-miners at home, but of course since they're the only kind of value in the new AI-communist society they'll be the obvious target for the other leisure gangs. So after enough rounds of violence each leisure gang will run a fortified, paramilitary "intelligence mining" operation, and oh by the way indie software dev is punishable by death in these territories.

Is this scenario, like, 9 kinds of insane? Sure! But so is the idea that we'll all be at the beach doing Idunno what, fucking? All this to say yeah, I'm with you that anyone who describes that AI will make a future defined by a lack of productive work is describing a depressing future...

[psawaya]

I wouldn't say that for sure. Here's a take that says it'll create a lot of growth for the games industry: https://www.owenmahoney.ai/owen-mahoney-blog/size-matters

[conartist6]

It bases this entire chain of assumptions on "AI is really big and everyone agrees".

But I don't see the games industry as all that vulnerable to AI at all. Game engines already drive constantly-improving dev efficiency through improved abstractions.

[Hammershaft]

None of this is necessarily zero-sum. I'm skeptical AI is going to be a meaningful tailwind for games, but if I'm wrong it could absolutely benefit customers, studios, and labs through boosting productivity.