At a glance
- The claim: generative AI is a separate, uniquely bad substance rather than a type of AI.
- The fact: recognition, prediction, recommendation, classification, and generation describe different jobs AI systems can perform.
- The practical distinction: ordinary applications can contain traditional tools, AI-assisted features, and generative features at the same time.
- The missing evidence: the label “AI” alone does not establish that anything was stolen, a rule was broken, or a person surrendered authorship.
Artificial intelligence has a public-relations problem. Not because nobody has heard of it, but because the term has become a substitute for several entirely different accusations.
Something contains AI, therefore it must be stolen. Someone used AI, therefore they cheated. A book came out quickly, therefore AI wrote it. An image looks polished, therefore nobody made it.
That is not evidence. It is barely even an argument.
AI is a broad category of technology. It is not a finding of guilt.
Generative AI sits inside that category. NIST calls it a “class of AI models.” The Government Accountability Office places it beneath deep learning, machine learning, and artificial intelligence in its own explanation of the technology.
“Generative” tells us what the system does. It generates content.
That extra word does not transform AI into a separate, uniquely sinister substance.
AI has job titles
There are different kinds of AI because there are different jobs we want computers to perform.
Recognition systems identify things. They may recognize a face in a photograph, spoken language, handwriting, an object, or an unusual transaction.
Predictive systems examine patterns and estimate what may happen next. That can mean predicting the next word in a sentence, tomorrow’s demand for a product, or whether a transaction resembles previous fraud.
Recommendation systems decide what to place in front of us. Netflix suggests a movie. Amazon suggests a product. Spotify queues another song. Social platforms choose which posts they think will keep us scrolling well past the point when we meant to go to bed.
Classification systems sort information. Your email provider uses them to decide whether a message belongs in your inbox, promotions tab, or spam folder.
Generative systems produce content. Depending on the tool, that may be text, images, music, speech, video, code, or a mixture of them.
These names are useful because they describe the work being done. They are not moral rankings.
The categories also overlap. A generative language model relies heavily on prediction while producing text. An application may classify an image before recommending an edit. Your email program can recognize spam, predict the end of your sentence, recommend a reply, and generate a longer message—all without leaving the same screen.
That does not make the categories meaningless. It means software is capable of doing more than one thing.
AI was already in the software
Long before every company began attaching sparkles to its AI buttons, people were using artificial intelligence without holding a town meeting about it.
Phones recognized faces. Search engines ranked results. Cameras identified subjects and adjusted photographs. Email services caught spam. Stores recommended products. Streaming services predicted what we might watch next.
Much of it was sold as “smart” technology. Sometimes it was not labeled at all. The feature simply appeared in an update, made something easier, and became part of ordinary life.
Then generative tools became widely available, and the conversation changed. People began discussing “AI” and “generative AI” as if they were unrelated technologies with completely different family trees.
They are not.
A system recognizing the subject of a photograph and a system creating a new background are performing different tasks. Both can be AI. One identifies existing information; the other produces new content.
We can discuss those uses separately without pretending one of them somehow stopped being artificial intelligence.
What is generative inside everyday apps?
Microsoft Word still contains ordinary formatting tools. Changing a font is not AI. Find and Replace is not AI. A basic spelling dictionary is not sitting there contemplating the emotional arc of your paragraph.
Word also offers Copilot features that can draft, rewrite, summarize, and add material to a document. Those features generate language.
Apple’s Writing Tools can proofread, rewrite, summarize, and compose text across supported applications. Gmail can suggest a few words through Smart Compose or use Gemini to draft and summarize more substantial material.
The amount generated can vary wildly. Accepting a suggested phrase is not the same production process as asking a tool to draft an entire article. That distinction matters when discussing authorship, effort, disclosure, or platform rules.
It does not change the technology’s family name.
Website builders offer another familiar example. A traditional template gives someone a prepared layout to customize. That is not generative AI. When the builder creates page copy, images, sections, or an entire preliminary site in response to instructions, it is generating content.
The application may contain both kinds of tools. Most modern software does.
Asking whether the entire application “is AI” will usually tell us less than asking which feature was used and what it contributed.
Dictation does not automatically mean generation
Dictation deserves separate attention because it is especially relevant to authors.
Traditional dictation converts speech into text. The author chooses the words, speaks them aloud, and the software transcribes them. Modern speech recognition can involve extremely sophisticated AI, but transcription does not automatically become generative merely because artificial intelligence helped recognize the words.
If the author says, “The door opened, and she knew immediately that something was wrong,” and those words appear on the screen, the software transcribed the author’s language.
If another feature takes the recording and summarizes it, rewrites a passage, drafts a scene from the author’s notes, or cleans up rough dictation by removing false starts and turning spoken fragments into polished prose, that part may be generative. The author still supplied the ideas and source material; the software helped shape how they appeared on the page. That is AI assistance, not proof that the tool invented the book.
Both processes may use AI. They are not the same process.
This matters when someone sees an author producing work quickly and announces that the books must have been generated. The author may dictate. They may write shorter books, work from detailed outlines, use an editor, or simply write faster than the person making the accusation.
A release schedule does not reveal a production method. Weekly releases do not prove AI use any more than slow releases prove that every word was composed with a fountain pen beside a rain-streaked window.
No, your installed Photoshop brush is not secretly generating art
Photoshop makes the distinctions easy to see because it contains ordinary digital tools, AI-assisted tools, and explicitly generative features.
Adobe identifies Generative Fill, Generative Expand, Generate Background, Generate Similar, and Generate Image as generative features. They create new visual material.
Object selection can use AI to recognize the subject of an existing photograph. Background removal identifies which pixels belong to the subject and which belong to the background. Those tools may be AI-powered without performing the same job as Generative Fill.
Then there are brushes.
Photoshop supports downloadable .abr files containing individual brushes or entire brush packs. A brush preset can include a brush-tip shape and settings controlling its size, spacing, angle, roundness, texture, scattering, opacity, and response to pressure.
The program applies those settings while the artist paints.
That is why one stroke can produce grass, smoke, fur, fabric, freckles, clouds, leaves, or a scatter of stars. It may save an enormous amount of time. It may also require considerable skill to use well.
It is still a brush preset.
A person could create the original brush tip using generative AI. If so, the asset would have AI in its history. That does not mean a generative model creates fresh content every time another artist uses the brush.
The same reasoning applies to textures, stock images, templates, fonts, filters, actions, and other digital assets. Where an asset came from and what the current software is doing with it are separate questions.
Complicated results are not proof of generation. Neither is speed.
AI is not a moral diagnosis
Here is the misconception underneath much of this argument: AI is being treated as though the term itself means bad.
It does not.
AI can be used badly. A company can behave irresponsibly while developing it. A person can use it to mislead people, imitate someone’s identity, flood a marketplace with junk, violate a platform’s rules, or present generated work as something it is not.
Those are actions. Criticize them.
There are legitimate questions about licensing, consent, compensation, privacy, disclosure, employment, environmental costs, and the material used to train particular systems. Those questions deserve evidence and specific answers.
“AI is bad” does not provide either.
It also lets the actual responsible party disappear. If a company obtained material improperly, that conduct belongs to the company. A current author using a commercially available tool did not personally crawl the internet, acquire the training material, or make the company’s legal decisions.
Responsibility should follow the act.
That does not prevent anyone from choosing not to use a particular product. It does not require approval of every model or every business practice. It simply means we should identify who did what before assigning blame.
Otherwise, we are not holding anybody accountable. We are transferring anger to the nearest convenient target.
Using a tool is not the same as surrendering the work
The presence of AI also tells us very little about how much human work went into the result.
An author might use dictation, ask a tool for alternative wording, run a grammar check, generate a rough description, or have a system summarize research. Another person might ask for a complete draft and publish it with almost no revision.
Both can truthfully say they used AI. Their processes are plainly not equivalent.
The same range exists in visual art. An artist may use AI-assisted selection to isolate a subject, generative fill to remove an unwanted object, or an image model to create nearly everything visible on the canvas.
If we care about the difference, we need to ask about the process.
Simply spotting an AI feature somewhere in the workflow does not tell us who made the creative decisions, how much was generated, how much was revised, or whether any rule was broken.
It certainly does not prove theft.
Get to the actual issue
If the concern is training data, discuss the training data.
If a platform requires disclosure, check what its policy actually says.
If someone claims a work infringes copyright, identify the protected work and the allegedly infringing material. Copyright infringement has elements that must be shown; it is not established by saying the word “AI” with sufficient disgust.
If the complaint is about quality, review the finished work. Bad books existed before generative AI. So did bad art, lazy marketing, copied ideas, misleading advertising, and people selling things they did not make.
AI did not invent human behavior.
It gave people some new tools and gave the rest of us several new questions to answer. We will not answer them by pretending every kind of AI is identical. We also will not answer them by pretending generative AI arrived from an entirely separate technological universe.
Recognition AI recognizes. Predictive AI predicts. Recommendation systems recommend. Generative AI generates.
Those titles describe their jobs.
AI itself is not evidence that something is bad, stolen, fraudulent, or worthless. If someone wants to make one of those claims, they need facts connecting the technology to the alleged harm.
Yes, that requires more work than pointing at something and yelling “AI.”
That is generally how evidence works.
Sources and documents
These links support the factual claims in this explainer. Interpretive conclusions are identified in the article and should be read with the underlying record.
- NIST: Generative Artificial Intelligence
- U.S. Government Accountability Office: Generative Artificial Intelligence
- Adobe: Generative AI Features in Photoshop
- Adobe: Create Custom Brushes in Photoshop
- Adobe: Import Brushes and Brush Packs
- Microsoft: Draft and Add Content with Copilot in Word
- Microsoft: Dictate Documents in Word
- Apple: Use Writing Tools with Apple Intelligence
- Google: Use Gemini in Gmail
- Google: Type and Edit with Your Voice in Google Docs
- Wix: Creating a Site with AI
- Squarespace: Using Squarespace AI
