# AI Is Still a Tool

Published: 2026-10-22

By [Ron Edgecomb](/about/index.md)

Tags: AI, Creativity

I recently [posted on X](https://x.com/ronedgecomb/status/2113047448608538829) about Travis Knight's comments at the _Wildwood_ premiere at the Academy Museum of Motion Pictures. Speaking about the technologies involved in filmmaking, [Knight said](https://popculturenews.com/travis-knight-says-ai-is-not-a-tool-ai-can-go-fuck-itself/):

> A pencil is a tool. A paintbrush is a tool. A 3D printer is a tool. A digital capture is a tool. AI is not a tool. AI can go fuck itself.

That sounds very different from what he told [TechRadar on June 4th](https://www.techradar.com/streaming/entertainment/masters-of-the-universe-director-travis-knight-doesnt-think-ai-is-a-technology-we-should-fear-but-says-itll-never-replace-the-people-who-give-movies-their-spirit-and-soul), when he said that "artificial intelligence can be a tool, or it could be a crutch." In that interview, he opposed replacing the people who make movies, but also said the technology called for care and thoughtfulness rather than fear.

I respect Knight as a director, and I admire the craftsmanship behind LAIKA's films. That craftsmanship is part of why his earlier position resonated with me. It left room to ask what a technology actually does, who it serves, and whether it helps people tell the stories they want to tell. The categorical rejection at the premiere leaves much less room for that conversation.

It's possible he was using AI as shorthand for generative systems intended to replace artists. I don't know whether his underlying position has changed as much as his language suggests. But the distinction matters, because a rejection of replacing artists is something I can understand. A rejection of an entire field of technology is much harder to defend.

LAIKA itself offers a useful example. In 2020, the studio and Intel presented [Rotomation, a collaboration on AI-powered rotoscoping](https://github.com/swkar/Rotomation#publications). LAIKA's replacement faces can leave visible seams between their printed parts. Removing those seams involves painstaking visual effects work, including drawing and adjusting the shapes that define where a correction should apply.

The collaboration used machine learning to predict facial landmarks and rotoscoping shapes from examples of artists' existing work. Artists could then adjust those results in their usual software. It assisted the preparation for cleanup; it didn't independently create a finished performance or eliminate the need for an artist to review the shot.

In an [interview about the project](https://beforesandafters.com/2020/08/26/machine-learning-rotoscoping-and-laika/), LAIKA's visual effects team described wanting to spend less time struggling with repetitive rotoscoping and more time refining the artistic result. That strikes me as a sensible use of technology: give artists more room to concentrate on the decisions that shape what an audience experiences.

Of course, that isn't the same as using a generative model to produce a scene from a text prompt. Training a system on a studio's own production assets raises different questions from training on other people's work without their permission. LAIKA's collaboration doesn't settle the ethics of modern generative AI. It does demonstrate why the word AI, on its own, tells us very little about whether a particular application serves artists well.

I kept coming back to that distinction while watching Corridor Crew's recent video, ["I Solved Green Screen, and I'm Giving it to YOU!"](https://www.youtube.com/watch?v=svrdRAoIlxM). Their team has been developing CorridorKey, a machine-learning system for separating filmed subjects from green and blue screens. Its job is to preserve difficult details such as hair, motion blur, and transparency so artists can composite the footage into another environment. The [project documentation](https://help.corridorkey.com/docs/getting-started/#how-it-works-briefly) describes reconstructing foreground colors and transparency, using information across frames.

What I appreciated about the video was its willingness to show the remaining work. A professional VFX supervisor reviews the results and identifies flickering, lost hair detail, and edges that still need attention. Niko also acknowledges spending about two hours removing tracking-marker posts from one shot after the initial key. The model gets them further, faster, but the demonstration still depends on someone knowing what a convincing result looks like and recognizing where it falls short.

There is creative judgment in deciding what to preserve, what to correct, and when a shot is ready. Automating part of that process doesn't make those decisions disappear. It changes where the artist spends their time.

I also don't think the case for AI has to stop at cleanup. Generative systems can give people ways to attempt work they otherwise couldn't afford to make. That possibility deserves consideration alongside the concerns about how those systems are built.

Take Zack London, the filmmaker behind Gossip Goblin, and his upcoming feature, _Gods Don't Give Gifts_. According to [London's own account](https://x.com/Gossip_Goblin/status/2107945038139531744), the film was self-financed and employed 30 people. His [production breakdown](https://x.com/Gossip_Goblin/status/2107741745358508486) describes more than nine months of work involving writers, voice actors, composers, core creatives, and a post-production team that included a foley artist. He has also [stated that the film contains no AI music](https://x.com/Gossip_Goblin/status/2107510554076316065).

That describes something considerably more involved than entering a prompt and accepting whatever comes back. Writing, performance, music, editing, and the decisions that connect them remain substantial parts of making a film.

For me, the interesting possibility is that a small team can pursue an elaborate science-fiction project without first securing the resources of a major studio. That doesn't guarantee a good film, but it does give more people an opportunity to find out whether their ideas can become one. I don't think access to that opportunity should count for less simply because generative tools were used.

Consent, compensation, training data, and control over a person's work or likeness offer a more useful starting point for discussion than dismissing the technology outright. I think there's room to work through those concerns while still being excited about what these tools make possible for artists.

I also don't think increasing autonomy disqualifies a system from being a tool. Software can carry out a complicated sequence of tasks without requiring a person to approve every intermediate step. The more discretion we give it, the more carefully we need to consider its limits, the decisions we reserve for people, and who is responsible when it makes a mistake. Its ability to act independently makes those questions more pressing.

This extends well beyond filmmaking. In October, OpenAI [reported work with PerturbAI and Scripps Research](https://openai.com/index/scaling-ai-research-for-complex-biological-data/) using thousands of AI agents to analyze data from millions of mouse brain cells and identify promising directions for further experiments. Anthropic [reported using Claude to design proteins that bind scorpion neurotoxins](https://www.anthropic.com/research/claude-scorpion-toxin-binder-design), with 17 of 211 tested designs meeting its binding criteria. Further testing is needed to establish whether those proteins can neutralize venom, but the work offers a concrete example of AI contributing to early drug discovery.

OpenAI has also [shared new mathematical results produced by an AI model](https://openai.com/index/sharing-ai-progress-in-mathematics/), including many proofs formalized in Lean so they can be checked by computer. These systems are helping people investigate questions in biology and mathematics, alongside their applications in creative work. For me, that potential is another reason to judge AI by what people do with it, and to leave room for uses that expand what we can accomplish.

Nobody owes AI a place in their creative process. If Knight wants LAIKA to work within a particular set of artistic constraints, I respect that choice. Artists should be able to choose the methods that give their work meaning, including methods that take more time and effort.

I want that same freedom to extend to artists who choose to work with AI responsibly. I want them to have more control over their work, more ways to realize an idea, and a fair share of the value they create. Knight's earlier comments left room for that possibility. I still think they offered the better starting point.