Authenticity On Trial: Darren Aronofsky, "Primordial Soup," and the rise of AI slop
Darren Aronofsky's "Primordial Soup" isn't just a science series. It's the newest flashpoint in a bigger fight over what counts as real work in an era of auto-generated everything.
Questions on social feeds about whether parts of the show's promos were built with generative tools lit up a deeper concern. Creators are calling the flood of low-effort, AI-produced filler "AI slop" - and it's starting to test the standards audiences expect from serious content.
Why streaming invites AI shortcuts
Netflix, Amazon, and Apple TV+ spend billions a year and still need more to keep subscribers engaged. It's not just the main show; it's the entire orbit - behind-the-scenes clips, educational explainers, social teasers, classroom worksheets.
That pressure creates a wide opening for automation. If you can crank out supplementary content faster and cheaper, the temptation is obvious.
The math: savings vs. accuracy
Science-forward documentaries are expense-heavy: deep research, expert interviews, rigorous fact-checking, and careful editing. One episode can take months and cost hundreds of thousands.
Generative tools promise 40-60% cost reductions on ancillary pieces by auto-writing scripts, whipping up visuals, or summarizing papers. But the tradeoff is clear: plausible output that can be wrong, overconfident, or stripped of vital nuance - and a hit to brand trust when viewers feel misled.
Tool or replacement?
There's a line many filmmakers draw: AI for technical assists (color, mix, cleanup) is different from AI drafting narratives, doing research, or making editorial calls. The first speeds workflow; the second rewires authorship.
The WGA deal put guardrails around scripted work, but documentaries, educational supplements, and promo content still sit in a gray zone. That's where misuse happens.
No clear rules, messy disclosures
There's no industry-wide requirement to label AI-generated elements in documentaries or educational media. Advertising has clearer expectations under the FTC, but that standard doesn't cleanly apply here, leaving most policies self-enforced and inconsistent. Federal Trade Commission
Some propose "human-created" or "AI-assisted" labels with third-party verification. Sounds great on paper - until you define thresholds, verify claims, and keep pace with shifting tools.
Science content has higher stakes
Explainers on evolutionary biology, cosmology, or climate aren't flavor text; they shape what people take as fact. Unchecked AI tends to present guesses with certainty, flatten debate, and skip context.
For topics like life's origins - where hypotheses compete and new evidence moves fast - mislabeling provisional ideas as settled fact is a disservice. Several scientific orgs now push for expert review of any AI-assisted material to protect rigor and credibility. American Association for the Advancement of Science
Audience trust is the real currency
Viewers tolerate AI in effects and music more than in educational or documentary work. The latter demands accuracy and clear authorship.
Premium platforms have more to lose. If subscribers sense padding with auto-generated filler, the premium label frays - and churn follows.
A practical playbook for studios, labs, and creators
- Label clearly: Disclose "human-created," "AI-assisted," or "AI-generated" at the asset level. Make the label visible where the content lives, not buried in credits.
- Draw bright red lines: No AI-generated interviews, citations, or factual claims. Use AI for search, transcription, and rough summaries - then hand it to experts for verification.
- Human-in-the-loop by default: Require subject-matter review for any science-facing script, graphic, or explainer. Include notes on what changed after review.
- Stand up an AI review board: Mix creatives, technologists, and ethicists. Approve use cases one by one. Maintain an audit trail of prompts, models, and edits.
- Procurement with teeth: Vendor contracts must disclose AI use, training data risks, and indemnify for errors or IP contamination.
- Budget for truth: Treat expert review as non-negotiable. Model the cost of losing trust against the "savings" from automation - then decide.
- Source of truth: Require citations for scientific claims, versioned research packets, and a change log for every update.
- Measure trust, not just output: Track complaints, completions, rewatch rates, and educator feedback. A/B test disclosure labels and adjust.
For researchers and educators partnering with media teams
- Secure veto power: If your name or institution appears, you get final say on scientific claims and framing.
- Citations or it doesn't ship: Every claim needs a source. Mark hypotheses and uncertainties clearly.
- Quarantine contested topics: No AI-generated prose for sections under active debate. Human-written, expert-edited only.
- Publish a methods note: Briefly explain how content was produced and reviewed. Transparency builds goodwill.
For individual creators balancing speed and standards
- Use AI for grunt work: Research pointers, transcripts, rough outlines - yes. Final facts, expert framing, and conclusions - your job.
- Show your receipts: Keep a living reference doc. Link sources. Invite corrections.
- Disclose assistance: A simple "AI-assisted for research; human-authored and expert-reviewed" line sets the tone.
What the Aronofsky moment signals
Whether or not AI touched "Primordial Soup" materials, the response exposed a weak point: audiences are now checking for authenticity, and they're quick to call out shortcuts. That scrutiny won't fade.
Set standards now, or the market will set them for you. The teams that win will be the ones who move fast without breaking trust.
Resources
- FTC resources on disclosures
- AAAS guidance and policy updates
- Complete AI Training: courses by job for creative and research teams
Your membership also unlocks:
