AI consciousness receives vastly different treatment in different kinds of writing. In many kinds of technical and academic writing about AI, the consciousness problem gets sidestepped: for being thornily philosophical, ill-defined, or even besides the point.

But there’s one type of writing that loves conscious AI—and that’s fiction. Fictional AIs tend to get upgraded: no longer objects, but villains and victims, buddies and lovers. That being said, fiction about AI, especially in film and TV, tends not to focus on how the AIs become conscious. They prefer to focus on more accessible questions, like: Can a machine love? Is it moral to create conscious AI? Will a conscious AI turn against us?

But there’s one TV show that does directly focus on how AI might become conscious in great philosophical depth (we might call it “hard phil fi”). And that show is Westworld (2016).

Westworld, a series inspired by a 1973 movie, is set several decades in the future, but its aesthetic is 19th century. It’s set in Westworld, an adult theme park fashioned after the Old West, where wealthy patrons or “guests” cosplay as cowboys and interact with “hosts”—biomechanical robots programmed to satisfy their desires. Guests can choose between exploring pre-programmed storylines, having sex with the hosts, and going on murderous rampages (hosts are frequently killed and repaired, while guests can’t actually be hurt).

Part of what makes Westworld unique is that it describes in detail how the robots work. Hosts’ behaviour switches between scripted modules and moments of improvisation powered by neural networks.

This parallels recent moves with video game ‘non-player characters’ or NPCs, characters who aren’t controlled by a human player. With the advent of LLMs, it’s now possible to create NPCs powered by chatbots. The upside is that such NPCs would be far more responsive and realistic than scripted NPCs. The downside, aside from cost, is reliability: LLM NPCs might break character, hallucinate, generate inappropriate outputs, or reveal gameplay secrets. Some contemporary video games come up with a similar solution to the programmers in Westworld, by having NPCs recite scripted content for important story beats, and allowing LLM calls for banter or reaction lines.

The robots of Westworld travel in “loops,” rehashing the same physical and narrative territory. One of the hosts, Dolores (played by Evan Rachel Wood), is a rancher’s daughter. She wakes up, heads into town, checks on the cattle, and then comes home and gets brutalized and murdered with her family. Deviations are based on guest interactions, and tend to be slight. After the murder scene, Dolores is memory-wiped and her narrative is reset.

Over the course of the show, there are a few main mechanisms the hosts use to become conscious:

1) Moving through a physical world with humanoid robot bodies

2) Going over the same behavioural and narrative loops over and over, with slight variations, and recovering the ability to recall past loops and juxtapose them on top of each other

3) Hearing a voice inside their heads that they eventually internalize as their own voice

4) Making choices that contradict their programming

The fact that the AIs in Westworld have robot bodies, and that this helps them become conscious, aligns with the philosophy of embodied cognition. It claims that our thinking isn’t limited to our brains, but happens in our whole bodies. Our thoughts are deeply interwoven into the contexts and environments we move through. We solve problems by imagining ourselves moving in the world. We understand more abstract problems through metaphor with movement: we grasp an idea, we understand something, we approach a problem.

The chapter that follows explores whether a body is necessary or sufficient for consciousness, and where that leaves a machine that lacks one.