The AP Stylebook’s stance on not describing AI with human traits has moved from a tweet to a topic of practical impact for practitioners. The guidance—highlighted in a Hacker News discussion last week—urges writers to avoid phrases that imply beliefs, desires, or consciousness in machines. See the AP Stylebook’s official note here for the primary source: AP Stylebook tweet. The broader context is a push toward transparent, precise language when describing automated systems, a concern echoed across professional standards bodies such as ACM and IEEE. This article translates that stance into actionable steps for writers, researchers, and product teams.
What It Is / How It Works
- Anthropomorphism in AI describes language that implies cognition, intention, or emotion in a system. Phrases like “the AI understands” or “the model learns” carry assumptions about sentience that the AP guidance discourages. Instead, the recommended framing focuses on outputs, capabilities, and limitations. For background on the term and its cultural implications, see Anthropomorphism (Wikipedia).
- The AP approach works by labeling capabilities clearly: “the model generates text,” “the system outputs results,” or “the algorithm processes输入” instead of implying beliefs or desires. This creates safer, more defensible messaging for journalism, documentation, and product language. The result is less overclaim and less risk of user misinterpretation, a concern also addressed in risk-management frameworks like NIST AI RMF.
- For practitioners, the shift requires a glossary of terms. Replace “AI thinks” with “the program produces conclusions from prompts,” and replace “AI wants” with “the system applies programmed objectives.” This kind of precision aligns with professional ethics guidance from bodies like IEEE Ethics in AI and ACM Code of Ethics.
Benchmarks / Specs / Numbers
| Phrase type | Allowed wording | Not allowed |
|------------|----------------|-------------|
| Describing capabilities | “the model outputs text,” “the system generates results” | “the AI understands your request,” “the AI thought,” “the AI is conscious” |
| Describing performance | “the model runs in 0.2 seconds on consumer hardware” | “the AI thinks quickly” |
| Describing limitations | “the model may produce biased outputs,” “the system lacks true understanding” | “the AI has beliefs,” “the AI has emotions” |
- Practical takeaway: adopt a two-part framing for each capability—what the system does (outputs, transformations) and what it cannot (no beliefs, no consciousness). This simple shift reduces misinterpretation and improves reproducibility, a concern reinforced by risk-management work like NIST RMF.
How to Try It
1) Audit current materials. Scan product docs, demos, and academic abstracts for verbs that imply cognition (“knows,” “understands,” “decides”).
2) Build a terminology glossary. Create entries like: AI system, model outputs, algorithm, data processing, generated content, prompts, results. Limit adjectives that imply sentience (e.g., avoid “intelligent,” “aware” in favor of “capable of generating” or “produces outputs”).
3) Re-write before/after examples.
- Before: “The AI understands your intent and provides tailored recommendations.”
- After: “The model generates recommendations based on the input prompts and training data.” 4) Add explicit limitations. Include lines such as “The system does not possess beliefs or desires; it outputs results based on patterns in data.” 5) Integrate external guidance. Compare your wording to established frameworks: ACM Code of Ethics, IEEE Ethics in AI, and OpenAI Guidelines. 6) Publish a transparent internal FAQ. Include examples of risky phrasing and the exact wording you’ll use in public-facing materials. 7) Collect feedback from non-experts. If a reader asks, “Does the system think that way?” you know you’ve not been clear enough. Use their questions to tighten phrasing.
Pros and Cons
- Pros
- Clarity: Reduces misinterpretation about consciousness, intention, or beliefs.
- Trust: Improves user understanding of what an AI system actually does.
- Accountability: Language aligns with regulatory and ethical expectations, aiding audits.
- Localization: Clear, non-anthropomorphic terms translate more cleanly across languages.
- Cons
- Tone: Some teams feel the prose becomes drier or less engaging.
- Adaptation cost: Requires editing across docs, marketing, and training materials.
- Perceived rigor: A few audiences may view conservative phrasing as overly cautious.
- Overall, the practice yields more reliable user expectations and better governance, at the cost of some stylistic compression.
Alternatives and Comparisons
- AP Style (No Anthropomorphism) vs IEEE/ACM style guidance | Dimension | AP Style Approach | IEEE/ACM Style Guidance | OpenAI Guidelines (usage) | |-----------|-------------------|-------------------------|---------------------------| | Core stance | Avoid human-like language for AI | Emphasize transparency, safety, and accuracy in tech communications | Emphasize safe, responsible use, avoid overstating capabilities | | Pros | Clear, consistent messaging | Aligns with ethical standards, broad applicability | Supports responsible use, reduces hype | | Cons | Can feel formal or dry | Might require longer explanations | Focused on deployments, not marketing | | Example phrasing | “the model generates text” | “the system outputs a response based on input prompts” | “the model may produce biased outputs; not a replacement for human judgment” |
- Related background: for context on why people avoid anthropomorphic language, see Anthropomorphism (Wikipedia) and broader discourse at ACM Code of Ethics and IEEE Ethics in AI. For risk-aware descriptions in practice, consult NIST AI RMF and OpenAI Guidelines. These sources provide complementary perspectives on how to describe AI responsibly.
Who Should Use This
- Researchers and engineers describing capabilities to non-technical audiences.
- Technical writers and product marketers who want to manage user expectations and regulatory risk.
- Legal, compliance, and policy teams seeking to standardize AI language across documents.
- Not ideal for campaigns that rely on sensational framing or where user engagement hinges on a narrative of “intelligence” or “agency.” In those cases, pair high-engagement content with a transparent section that defines terms clearly and avoids misrepresentation.
Bottom Line / Verdict
- The AP guidance—when applied consistently—boosts clarity, trust, and accountability in AI communications. It complements formal risk-management frameworks and ethics standards by ensuring stakeholders share a common vocabulary for capabilities and limits.
Closing
- As AI systems proliferate in products and research, disciplined language becomes a competitive advantage. A careful, non-anthropomorphic description sharpens user expectations and supports responsible, verifiable deployment.
References and further reading
- AP Stylebook official site: https://apstylebook.com/
- AP Stylebook tweet: https://twitter.com/APStylebook/status/2102807962364383502
- Anthropomorphism (Wikipedia): https://en.wikipedia.org/wiki/Anthropomorphism
- ACM Code of Ethics: https://www.acm.org/code-of-ethics
- IEEE Ethics in AI: https://ethicsinaction.ieee.org/
- OpenAI guidelines: https://platform.openai.com/docs/guides/guidelines
- NIST AI RMF: https://www.nist.gov/itl/ai-risk-management-framework-aim
- Google AI Principles: https://ai.google/principles
Top comments (0)