Media Style Guide
How to talk about "AI".
When talking about AI, media and governments often use the same inaccurate or misleading phrases. Inspired by the Vision Zero Vancouver Style Guide, we present better ways to talk about "AI", whether you are a caregiver, educator, student or member of the media.
Even toothbrushes are being sold with "AI". In this case "AI" means it has a sensor that checks if you missed a spot.
"AI" has become a meaningless word
"AI" is a term that is being applied to everything. Robot vacuum cleaners now have "AI". Upscaling, a TV feature that has been around for a decade, is now "AI Upscaling". There are even toothbrushes being sold with "AI".
Then there is the "AI" in the form of ChatGPT, video-generation product Sora, Microsoft Copilot and a million others.
On the other end of the spectrum, "AI" can be used by radiologists to identify cancer.
These are all called "AI", but they are clearly different products with different levels of scientific rigour and utility to society.
Calling them all "AI" flattens discourse and makes it harder to have a sensible discussion. Companies and schools point to use of "AI" in science to justify putting chatbots into classrooms.
Its over-use has made "AI" a meaningless marketing term. It should not be used in articles, public discourse, public policy.
For products, instead mention the product
- ❌ "AI is now in classrooms."
- ✅ "Chatbot products like ChatGPT are now in classrooms."
Chatbots like OpenAI's ChatGPT, Microsoft's Copilot are products. They are not "tools". The products were created to make a profit for the parent company.
Large language model chatbots are better described as services. They perform a task for the user
For research, use "program" or "algorithm"
- ❌ "AI can identify cancer."
- ✅ "Researchers used a program to identify cancer."
- ✅ "Researchers used an algorithm, trained on many images of cancer, to help identify cancer."
"AI" is meaningless obfuscation in this case. Researchers created a tool for a specific purpose and then used that tool to carry out research. They checked their work against some kind of benchmark, and if their results were published in a journal, their work was peer-reviewed.
Do not restate marketing
There are many ways in which corporations are selling AI to us. Language used to sell is harmful when adopted by everyone, as it becomes a believable reality. This is called AI hype. Something with AI, is not automatically better than without.
If a school or government body is proposing to make AI available to students, the first question should be how will this product help students?
- What pedagogical research is the school basing their decisions on?
- What do current studies show about the effects of LLMs on students' learning?
- What do parents and students think about the plan?
- How does using this tool affect the institution's decarbonization goals?
- Educational institutions often have rules on plajorism. How does the school reconcile that with recommending a product trained on stolen data?
"CEO of AI company says AI …"
If the CEO of a candy company said "we think people will eat more candy", The CEO of a candy company A statement on AI from someone with a vested interest in AI succeeding should be treated with
"Inevitable"
Companies selling AI have benefited immensely from portraying AI as "inevitable". School administrators use the same circular framing as justification when they decide to put AI products into classrooms.
"AI is here to stay"
Don't anthropomorphise
Chatbots Microsoft and OpenAI love to talk about their products as if they are a person. Almost all chatbots speak using first-person pronouns, saying "I did this."
It is hard to talk about a product that carries out tasks without accidentally speaking about it as if it has agency.
When people use language that treats AI products like a person, it degrades the value of humans. It also makes the product seem more capable and trustworthy, reduces accountability, and makes it harder to be critical of AI outputs even when they are incorrect.
Centre the human, not the product
- ❌ "AI can solve this problem."
- ✅ "People can use an LLM to generate text which can include an answer."
LLMs cannot solve problems. They can output text which may be the same as a correct answer.
- ❌ "AI can now make movies."
- ✅ "People can use products, trained on other movies, to generate videos."
When an author writes a book on a computer, we don't say "Microsoft Word wrote Game of Thrones."
Do not attribute agency
- ❌ "AI broke out and hacked a competitor"
- ✅ "Researchers made a hacking program, incorrectly contained it and connected it to the internet"
This example is based on a real story. Far too many media outlets breathlessly reported that AI had "broken containment".
This one is unfortunately extremely common.
Understanding
- ❌ "AI understands what you're saying and can give an answer."
- ✅ "Copilot will output text based on what you type in."
LLMs are not capable of understanding, no more than the autocomplete on your phone can "understand" the content of your email.
"Hallucination"
- ❌ "One problem with AIs is their frequent hallucination."
- ✅ "Products like Copilot are designed to fabricate believable text."
Humans halluncinate, LLMs create output when given an input. Sometimes that output matches reality, but the algorithms it uses do not have any concept of what is "true", only what is likely.
By using a complex term like "hallucinate", it hides the simple truth that LLMs are designed to make things up.
Further Reading
- How to talk about "AI" without adding to the anthropomophization by Emily Bender and Nanna Inie