Local AI and the survival of living art

I think a lot of future art will be systems and worlds run by AI, and if we want those works to last, we need to be able to keep them running ourselves. This matters quite a lot to the sort of work I want to make.

I'm increasingly interested in worlds that carry on developing after the artist has finished building them. Imagine an online world inhabited entirely by NPCs. They might form societies and develop shared goals, with groups cooperating to overcome a challenge or going to war with each other. People could watch what happens, but the world would have its own activity and history regardless of whether anyone was visiting. The relationships between its inhabitants could become the most interesting part of the work.

The AI would be part of the artwork throughout its life. Switching it off could mean losing much of what makes the work interesting, even if all the visuals and code were still there.

If we are moving towards digital techno-feudalism, then running our own AI seems like a fairly practical way of taking back some of the means of production. So much of what we use is becoming something we rent access to, and the companies providing it get to decide the terms. Having a model we can keep and run gives us some control over that.

Sage is one example. I'd be happier with a version whose AI runs locally and can be updated when that would improve the work. The installed version should carry on functioning even if the source of those updates disappears. That would give Sage a way to develop without making its continued existence dependent on a cloud service.

For an artist making a world that depends on AI, there is a further reason to care. I might use a service to build a piece and then move on, but if that service is also needed to keep the finished work functioning, the dependency could last for as long as the artwork does. Every collector who takes it on inherits that dependency too.

I don't think we can seriously talk about these works lasting for decades without dealing with this. A company may be doing very well when the work is made and have entirely different priorities ten years later. It might stop offering the model the work uses, or charge an amount that makes keeping it running difficult. The artist and collector could still care deeply about the piece and have very little ability to do anything.

Replacing the model could be quite a substantial change to the art. If a world has spent years developing its own societies, a new model might make its inhabitants behave differently. They may become more capable while losing the behaviour that made them interesting. A slightly strange or limited intelligence could be exactly right for a particular work.

Keeping the NPC world running through a collection of cloud services could become a job in itself. An update to one service might mean changing how another part connects to it. If these parts keep moving, somebody has to keep checking that the world still works and still behaves as intended.

Perhaps a local AI could do much of that maintenance, adapting the code as services change. That could be a valid part of the artwork, although the repairs might gradually take the world somewhere I hadn't intended. The services would be influencing its development, and the AI would be interpreting those changes. I'd still want that AI to run locally, because otherwise the thing responsible for maintaining the work would have its own dependency on an external service.

The simpler approach would be a complete installation containing a fixed local model and everything it needs to run the world. In effect, a black box that can be put on a suitable server and left to operate. The AI could control the inhabitants and, where the work allows it, adjust the world as it goes. There would be no external AI service that had to remain available for the piece to function.

I also suspect a lot of this art will need much less intelligence than whatever the best available AI happens to be. Once a local model is good enough to do what a piece needs, there may be no artistic reason to keep upgrading it. The world can develop through what happens inside it while the underlying model stays the same.

There is obviously more to this than downloading a model. Someone would need to keep the software required to run it and have the right to continue using it. The work would need instructions that somebody else could follow, including how to get it running on replacement hardware. For a world with an accumulated history, its memories and current state would need preserving too.

I'd want a collector to receive enough of the working system to have a reasonable chance of maintaining it. Handing over code that calls a service which might disappear leaves a fairly large part of the artwork outside their control.

Semi-permanent is probably the sensible ambition here. Computers will need replacing and software will need attention, but owning the necessary parts gives somebody a chance to do that work. A piece might sit in storage for twenty years and then be brought back into use. I would like that to remain possible.

The world could still be online, with visitors connecting to the server to see what has happened. The artist or whoever looks after the work could move the installation to another machine, and groups of artists could share the cost of running that equipment.

The important part is having the ability to keep it going when an external provider moves on. Local AI only helps if the rest of the work has been built with that possibility in mind.

If I build a world that people spend time in and become attached to, I want them to have a way to keep visiting it after I've stopped working on it. Taking back the means of production makes sense to me, but for this kind of art we also need to own what keeps the finished work alive.