Latency
Written by and directed by James Croke
Starring Sasha Luss, Alexis Ren, Ava Caryofyllis
Rated PG-13
Runtime: 94 minutes
Available digitally July 9

by Tessa Swehla, Staff Writer

In this current AI craze that doesn’t seem to be slowing down anytime soon, it can be refreshing to see a film that critiques the current status of AI and questions the rush to integrate it into every part of our lives. In his directorial debut, James Croke gives us one such critique in the tightly-focused 94-minute horror film Latency. Hana (Sasha Luss) is a professional gamer who suffers from acute agoraphobia due to a traumatic childhood. She stays day after day in her small apartment, testing new games and equipment, but she can barely pay the bills. Salvation comes in the form of a new technology that she is asked to test: Omnia, a sleek headset that uses electro encephalopathy (EEG) and neuroimaging, as well as a Learning Language Model AI, to “read, decode, and measure” patterns in Hana’s brain. The cheerful voice of the tutorial promises that it will eventually help reduce Hana’s reaction time and almost seem to “read your mind” after a brief calibration period. Hana sees the Omnia as a technological advantage to win tournament prize money before anyone else has access to it. But as the calibration continues, she begins to have disturbing dreams and hallucinations and starts to wonder if the Omnia is reading her mind or manipulating it.

Most of this film occurs in one location, Hana’s dim and claustrophobic apartment that makes up her whole universe. Croke comes from a production design and writing background, and it shows in the realistic and yet highly personalized set design. It is clearly the only apartment that Hana can afford, a little dingy, but filled with games and gaming equipment. Sometimes, when a character has a hyperfixation or a true vocation, the design of their space can feel a bit performative, especially when it comes to “nerd culture.” Hana’s space doesn’t feel like that at all: since the film spends so much time in her space, it feels like the production designers took the reflection of her interests and personality seriously, including games and equipment that looks used and well-loved instead of perfunctory. 

This is especially true of her Game Boy–complete with Tetris cartridge–that she plays with several times in the film and appears in flashbacks to her childhood. She tells a little girl living in her apartment complex (an unseen Ava Caryofyllis, who speaks to her through the front door) that Tetris kept her grounded and provided a retreat from her mother. The contrast between the older technology versus the new, shiny AI tech that begins to terrorize her is, perhaps, one of the better observations of the film: technology used to bring joy and comfort to us, right? When did it start making life worse?

It is important to note that what is now popularly called AI is not, in fact, what is typically meant by that term. ChatGPT and its brethren are actually what are called Learning Language Models (LLMs). LLMs use machine learning to complete tasks, which requires scraping a massive amount of data from the internet and other databases. It isn’t conscious, but the hope is that the more data it intakes and the more sophisticated its algorithms become, the more accurate its predictions will become. The Omnia is one such so-called AI: its dataset is Hana’s neural network and her reactions to sensory input. It also fully integrates her into her various devices: Hanna can type, play her games, turn on music, and text all without touching her keyboard or phone. Hanna is assured that the Omnia “is a tool: it will only do what you ask of it” which is how ChatGPT and other LLMs are are advertised.

However, there have been many noted issues with these models. For one thing, they are based on datasets that are often biased in terms of the information on race, gender, ethnicity, and sexuality, which can skew the results the AI gives. For another, the LLMs are made to give humans the answers that they think that the humans want. For example, if a human wants help with a coding problem, they write a prompt for the AI, asking it for help. This means that the machine is focused on answering the prompt that the human asked, not necessarily the problem being posed, creating patterns in the data where none exist (like a human seeing a shape in the clouds) or just giving an incorrect response–a phenomenon known as AI hallucination. This phenomenon is at the heart of Latency, which poses an unholy alliance between the human subconscious and AI hallucinating. For fans of literary cyberpunk, Latency is Synners (written by Pat Cadigan, 1991) lite.

None of this is explained in the film, Croke choosing to show the problem rather than explain it. The human subconscious alone is a scary and confusing place, uncontrollable and unpredictable, but what if a mental illness is thrown into the mix as well? What is the dataset that the LLM is drawing on then? Since Hana is the character we spend almost 95% of the film alone with, Luss’s performance is key, and she does a great job of communicating Hana’s desires as well as her fears. In the very first scene of the film, she effortlessly takes down huge fantasy creatures in a VR video game but is frozen into paralysis by the idea of opening her front door to sign for a package. She also is able to communicate the ambiguity of her illness: she wants to get better, she wants to go out into the world with her friend, and she doesn’t want to be so alone, but those desires belong to the conscious mind, not the unconscious one. By the end of the film, she can’t trust her senses or even her perception of reality anymore, a truly terrifying prospect made even more terrifying by the fact that she can no longer remove the headset (which I also see as a darkly funny nod to the ways in which I have to keep opting out of AI in every single program or browser I use nowadays).

From a horror perspective, the scares should be more creative or inventive. There are several jumpscares and a fun scene in the bathroom that plays on that anxiety dream that many people have about their teeth falling out, but ultimately, the imagery didn’t really stick with me. If I’m seeing a film about AI hallucinations of human ids, I want to see some weird stuff, you know? I also wasn’t entirely convinced by the ending, which seemed so obvious that I actually dismissed it as an option half-way through the film. It just seemed a bit straightforward for a problem which is anything but straightforward for those of us facing it in the real world (Cadigan does a better job of this for anyone who wants to read Synners). The term latency refers to the lag between a request sent from a network and when it arrives at a server, and while I can see how that applies to this film, I found myself wishing that they would delve more into the idea of Hana as a server that the AI is making a request of, rather than the other way around.

For a debut film, however, Latency is a great swing, and, as much as I love a film about AI overlords, I hope to see more sci-fi filmmakers actually take on the problem of AI as it exists now in our ever shittier technological landscape.