Creating your own AI

Most AI constructs in Horizon are, effectively, NPCs with a single driving motivation. GAIA’s purpose is to restart life. HADES is supposed to recognize a failing system and wipe it clean so GAIA can try again. Even CYAN, arguably the most nuanced AI shown, originally had the sole purpose of maintaining the Yellowstone caldera.

Like all characters, AIs become more interesting when you take that drive and apply it to the world, giving it nuance. HADES isn’t inherently evil, but it’s programmed to ignore everything else in the service of its goal, which puts it at odds with humans. For all we know, HADES may have been correct in its assessment that the ecosystem was unstable — which will be of little comfort to the people fleeing from eradication. CYAN became more interesting when that drive broadened to include caring for humans in the Yellowstone area, not just geology.

Start with a concept that is simple to communicate, but would require vast computing resources to execute. For example:

Just like humans don’t fit together with neat, clean, puzzle-piece edges, AI goals can and will overlap, intersect, and end up at odds. Take those simple concepts and draw them out to see where they could be taken too far. For example:

Believable AIs

Be careful not to make an AI too “good” or “evil”, especially its drivers. Just like any villain, or better yet any hero who finds themselves the villain, the AI will be more interesting if it can plausibly believe it is doing the right thing. AI drivers should be reasonable enough such that you could, with well-constructed arguments, convince people to back the AI as right. You might have to result to some rhetorical trickery, and you might not be able to convince many people, but the more people who might be convinced the harder players will have to work to justify their actions for or against it.

Creating a “rogue AI” might be straightforward enough for a one-shot or short, self-contained adventure. In a larger campaign, players should be encouraged to take into account the side effects of eliminating or altering the AI. The world of Horizon, whether humans know it or like it, only exists because of AIs. Removing any AI could have catastrophic, or at least long-lasting, effects on the world:

AIs are not deities

It might be tempting to see the AIs as something like Horizon’s pantheon of gods. This is true in some aspects, such as the global impact they can have, but very different in others.

Deities in most settings will have motivations and goals the characters will find inscrutable. This is especially true in settings where the deities are pan-dimensional, omniscient, or otherwise beyond comprehension. Horizon AIs are the opposite: their goals were defined by human programmers who defined the AI’s success conditions via code. The mystery inherent in understanding an AI is therefore about figuring out what those coded goals were, and then why they don’t align with what the players want or expect. In this way, an AI is far more human than deity.

Having said that, many of the systems one might find for interacting with deities could be applied to AIs.

For example, in a fantasy setting you might have Warlocks or Clerics who have taken a deity as a patron. It’s not unreasonable to have something similar here. Aloy could have proposed a patron-like deal with CYAN: whenever she was in Banuk land, visible to CYAN’s sensors, yelling certain words aloud might indicate that CYAN should overcharge a nearby tower, sending a lightning bolt at Aloy’s target. CYAN might have a limited number of towers through this might work, needing to rebuild them overnight, just like spell slots.

Like deities and NPCs in other settings, AIs are far more interesting when given limitations.