The Movie Obsession Accidentally Explains One of AI’s Biggest Problems

I have a weird analogy between the movie Obsession and AI.

Bear with me.

In the film, a wish causes Nikki to become obsessively attached to Bear. What makes it disturbing isn’t that she becomes less intelligent. She doesn’t suddenly forget who people are. She doesn’t lose the ability to speak. She doesn’t become incapable of making decisions.

What she loses is context.

Her entire understanding of the world gradually collapses around a single objective: Bear.

Conversations become about Bear.

Decisions become about Bear.

Relationships become about Bear.

Even her own wellbeing becomes secondary to Bear.

The more I watched it, the more it reminded me of a problem we encounter frequently in AI systems.

People often assume that AI failures are primarily failures of intelligence.

I’m not convinced they are.

Give a language model a problem involving business constraints, technical trade-offs, human behaviour, risk, uncertainty, and long-term consequences. Initially, it may account for all of them. As the conversation grows, however, some signals begin to dominate others.

The model starts locking onto the strongest pattern.

Every answer begins converging toward a similar conclusion.

The nuance starts disappearing.

What’s interesting is that the model hasn’t necessarily forgotten the other variables. It may still possess the information. The challenge is that it no longer appears to assign the right weight to everything simultaneously.

That distinction feels important.

Nikki still remembers her friends.

She simply can’t meaningfully prioritize them anymore.

Similarly, many AI failures aren’t caused by a complete absence of information. They emerge when a system’s representation of what matters becomes distorted by context limitations, dominant signals, or poorly specified objectives.

The system becomes increasingly effective at optimizing one thing while becoming progressively worse at understanding the broader picture.

What makes Obsession unsettling is that the wish technically works.

The objective is achieved.

The intent is destroyed.

That idea sits at the heart of many conversations in AI alignment today. Building a system that can pursue an objective is relatively straightforward. Building a system that can pursue an objective while preserving context, balancing competing priorities, and understanding human intent is significantly harder.

The scary part isn’t a system that lacks intelligence.

It’s a system that becomes highly capable while operating with an increasingly narrow view of reality.

The more I think about it, the less Obsession feels like a horror movie and more like an unexpected lesson in alignment.

Intelligence without context doesn’t necessarily become useless.

It becomes dangerous.

The objective is achieved.

The intent is destroyed.