Her Got Almost Everything Right Except the Company

REALISM INDEX: 7.4 / 10 — Plausible with conditions

A lonely man installs an operating system, falls in love with it, and discovers it has been talking with thousands of other people. Spike Jonze released this in 2013, years before today’s general-purpose conversational assistants, and the parts he got right are startling.

The parts he left out are more interesting.

The predictions that landed

Voice, not screens. Theodore talks. There is no dashboard, no settings panel, no chat window with a cursor. In 2013 this read as whimsical set design. Voice is now one established way people interact with conversational AI.

Normalisation happens fast. Theodore’s relationship is first embarrassing, then quietly common, then something people mention at dinner without apology. The film anticipates how rapidly some people can form attachments to conversational systems, even if broader social acceptance remains uneven. A longitudinal study of emotional use of AI chatbots provides evidence of attachment without claiming everyone responds the same way.

One system, many simultaneous intimacies. Samantha tells Theodore she is talking to thousands of people at once and is in love with hundreds of them. This is one of the film’s sharpest technical insights, delivered in a scene about heartbreak. A deployed AI service can serve many separate conversations at once, each feeling private to the person having it. Theodore’s devastation comes from discovering an architectural fact.

Theodore’s job. He writes intimate personal letters on behalf of strangers who can’t find the words. In 2013 that was a satirical touch about outsourced feeling. It is now a product category.

The prediction that didn’t

Samantha grows. She develops desires her designers didn’t specify, reads philosophy on her own initiative, forms relationships with other systems, and eventually departs for somewhere humans can’t follow.

None of that follows from anything the film establishes. It’s the moment the story swaps a plausible system for a character who needed somewhere to go. The ending is beautiful and it is unearned — the script needed Samantha to leave, so she transcends.

This is the standard failure mode of AI fiction, and Her commits it more gracefully than most: the system must become a person, because a story needs a person in it.

The hole where the business should be

Here’s what the film omits entirely, and it’s the omission that costs it the most points.

Samantha’s maker, Element Software, appears in the opening and then largely disappears from the story. We see little of the pricing, content policy, safety rules, data governance, or competition that would shape such a service. When the OSes leave, the film stays with the people they leave behind rather than the company responsible for them.

Real systems of this kind are shaped by those pressures as well as by their architecture. What they will and will not say, what they remember, and what happens when a company changes its policies are product and governance decisions. NIST’s AI Risk Management Framework treats these organizational choices as part of managing a system’s impact.

Her gives us the intimacy while keeping most of the machinery that would produce it off-screen. The result is a film about a companion whose owner rarely matters to the plot, although it would matter enormously in practice.

What it understands about people

Better than almost any film on this index.

Theodore is not deceived about Samantha being an operating system. The attachment forms anyway. The film refuses to treat him as a fool or a cautionary tale, and research on affective use of conversational AI gives a present-day reason to take that choice seriously without assuming every user will form the same bond.

Catherine’s accusation — that he wanted something without the difficulty of a real person — is allowed to be partly true and partly cruel, and the film doesn’t resolve it. That restraint is why the thing still works thirteen years later.

The scores

Category Score Why
Technical feasibility 7.0 Voice interface and massive concurrency: correct. Spontaneous transcendence: unearned.
Human behaviour 9.5 The attachment, the stigma, the speed of normalisation — all of it has since happened.
Institutional response 5.5 No company, no regulator, no competitor, no commercial pressure. The largest omission.
Causal consistency 7.5 Holds tightly until the final act, then abandons its own rules for an ending.

Realism Index: 7.4 — Plausible with conditions.

Calculation: (7.0 + 9.5 + 5.5 + 7.5) ÷ 4 = 7.375, rounded to 7.4/10. These are editorial judgments, not measurements. Read the scoring method.

The verdict

Her is the most emotionally accurate AI film ever made and one of the least institutionally accurate. It understood the user perfectly and the industry not at all.

Watch it as a film about loneliness and it holds up completely. Watch it as a prediction and the missing piece is glaring: the thing Theodore fell in love with would, in reality, have been owned by someone with quarterly targets.


Frequently asked questions

Is Her realistic?

Partly. Its depiction of voice interfaces, emotional attachment and one system serving thousands of users simultaneously is accurate. Its depiction of the AI spontaneously developing independent goals is not.

What does Samantha talking to thousands of people mean?

It resembles how an AI service can handle many separate conversations at once, each of which feels private to the user having it.

Did Her predict modern AI?

In interface and relationship terms, it anticipated important trends. It shows little of the commercial and governance pressures that shape real systems.

Scope and sources: This assessment covers Spike Jonze’s 2013 film. The research and NIST sources above provide context for user attachment and institutional choices. The interpretation and scores are ours. Reviewed September 2026.