Richard Atlas organizes a series of roundtables in Montreal, "Rich's Relaxing Chats." I went to the latest one, at Indie Asylum. There was a fascinating roundtable on AI, and then I moderated a talk about "Where Do We Go From Here." Some takeaways:
- No one is handing off creativity to genAI.
- There was a consensus that it was tempting to give AI big vague asks, but that then you're stuck with the assumptions AI made.
- Creativity comes from working through the problem. I just finished a draft of the dialogue for an indie game. I had a couple of ideas how to make us care about the main character, but those didn't work and I did something else. I had to be writing the dialogue to realize what didn't and would work.
- Okay, one programmer said he's having genAI do art for his game. The problem with this is, when you hire an artist they're not just giving you what you ask for. They're interpreting it, and trying to give you what you need. Which may not be the same thing. I'm not a visual artist. I can ask for an image in a script, and my friend Whitney (we're doing a graphic novel together) will do something else better.
- The programmers are all vibe coding. It's impossible to keep up if you're just coding by hand.
- A lot of folks are asking AIs to make tools. For example, instead of using Jira for project management, ask ChatGPT or Claude to create a tool that does exactly what you want.
- I can imagine asking Claude to take the audio from a voiceover session, select the last take of each line in the Excel, make it a separate file, and give it a unique name on the spreadsheet.
- On We Happy Few, that was a manual task for audio, and they hated it. That functionality could be expanded to generating robospeech, putting it in the game, swapping it out for temp audio and then final audio.
- AI is great for repetitive, annoying tasks. I asked Claude to go through a 16,000 word dialogue-and-bark spreadsheet and tell me how many words each character has, so we can schedule our actors accordingly. I asked Claude to take all the dialogue out of a 990-passage Twine and put it into the spreadsheet. This was after I spent an hour on the first 100 passages.
- A big concern for some was games being cloned. Some games have been cloned minutes after the original was released on Steam. You can't patent or copyright a mechanic.
- I'm not sure how much fun the clones are to play, though. You can absolutely copyright dialogue and music, after all. Someone who participated in Steam NextFest said the first day was filled with AI slop, but by the last day, all the slop had been relegated to the bottom of the pile. If someone cloned Stories: The Path of Destinies, they'd maybe get a hack-and-slash platformer, but it's not going to have Julian Casey being sly Reynardo. (Or my clever dialogue.) You can't copyright an idea, but you can absolutely copyright the expression of that idea.
- Good news for game writers, artists and composers.
- Patent law needs to catch up with the technology. You shouldn't be able to patent a mechanic, that would stifle innovation, but how about a design? There was no patent for "the sewing machine," because you can't patent an "obvious" idea, and "automate sewing" is an obvious idea. But you can patent the mechanism for handling the needle, and the treadle, and so forth. (Singer dominated the market by buying everyone's patents, making it extremely difficult to design a new sewing machine.)
- It is going to be increasingly easier for small teams to do what big teams did. That means there are going to be way more games coming out. Steam released 19,000 new games last year. That number will go up.
- That means games can be cheaper.
- We could conceivably move towards a bookstore model. Books cost very little to write and publish. Consequently you can go into Indigo and see a hundred thousand books for sale.
- A bookstore model could mean more diversity of style and substance. Lots of games about shooting people or punching them in the face, of course, but also more games like Unpacking and Dressmaker, which are not for fourteen-year-old boys.
- It will be harder for games to get attention. Games that streamers like, and which look good on TikTok, will dominate the market (even more than they already do).
- Fads may go faster. Recently there were a slew of cozy games. Now that niche may be saturated. What's the next fad? Obra Dinn-likes? Used to be every indie game was a soulslike, a roguelike or a Metroidvania. Lower entry threshold enables diversity.
- Small teams are good for working for games that work on legacy platforms. Consoles are ridiculously expensive. Someone who makes a new game for Xbox One + Kinect might find a market. A lot of people have a Kinect in their basement, in the box with their Wii.
- Vibe coding is particularly good for prototyping new features. Hey, would this be fun? Let's see. Oh, no, it's not. Well, we only spent a day on it. How about this other thing, would that be fun? You will still need to code a new feature properly so it doesn't eat memory and processing time, but you're not coding stuff that won't be fun.
- How do juniors break in? AI does a lot of the things that juniors used to do, more cheaply. In the past I'd have handed off the word count problem for a junior or an intern to rig using Excel's various COUNT() functions. I suspect breaking in will mean showing the five or ten games you made yourself. That's the movie paradigm: make short films, get noticed, get a job.
- Like movies, though, that is much easier for trust fund babies than, say, single moms. A single mom can be a junior on a game, especially if she can work remotely. But if she needs a day job for money, plus take care of the kid, PLUS make her own games, well, only the most exceptional people are going to surmount that obstacle.
- No one knows what will be feasible once the AI companies start charging what they're spending. Right now your AI tokens are heavily subsidized by venture capital money. At some point those lads will need to get paid.
- And thank goodness, because one thing killing the industry right now is that AI is sucking up all the investment money. (The other thing is that the cost of chips is through the roof.)
- AI improvements are not plateauing. AIs now index the entire internet (holy plagiarism, Batman!), but the models can get smarter. Your brain is not an undifferentiated mass of axons. It just looks like it. It has architecture developed over millions of years. AI is just at the beginning of that.
- AIs may learn how to make themselves smarter, i.e. the Singularity. It's a singularity because absolutely no one knows what that looks like.
- I wouldn't call any of those chats "relaxing," Rich! But thanks for organizing that, and bringing babka.
