AI assistants and generative AI tools can draft text, summarize documents, answer questions, generate images, and write code, but they are statistical pattern-matching systems, not knowledge authorities. Their outputs can be fluent, confident, and completely wrong — a phenomenon called hallucination. Using AI tools effectively means writing clear prompts, critically evaluating every output against reliable sources, understanding that AI reflects the biases in its training data, and recognizing the ethical boundaries around attribution, academic honesty, and creative ownership.
Ask an AI assistant a factual question about a topic you know well and verify its answer against a trusted source. Notice where it gets details right and where it fabricates plausible-sounding errors. Then try rephrasing your prompt to be more specific and observe how the output quality changes. Discuss with someone when it is appropriate versus inappropriate to use AI-generated content.
From your study of online information evaluation, you've learned to look for authorship, evidence, and corroboration before trusting a source. AI tools require those same skills — plus a new layer of caution. An AI assistant does not retrieve information from a database or look things up: it generates text by predicting, word by word, what a plausible response would look like given everything it was trained on. This is why the outputs are often grammatically polished, contextually relevant, and completely wrong: the model is optimizing for text that patterns like a good answer, not for accuracy. The term hallucination describes responses where the AI states false information confidently — invented citations, incorrect dates, nonexistent laws, fabricated statistics. These failures look identical to correct responses; confidence of tone is not a signal of accuracy.
The practical response to this is not to avoid AI tools, but to treat every AI output as a first draft from a knowledgeable but unreliable assistant. For factual claims, especially specific numbers, names, dates, or citations, verify independently using the source-evaluation skills you already have. For tasks where factual accuracy is less critical — brainstorming, drafting structure, reformatting text, generating initial code that you'll test and debug — AI tools can provide genuine value without the same verification burden. The key mental model is: use AI to reduce blank-page friction, then apply your own judgment and knowledge to refine the result.
Prompt quality dramatically affects output quality. A vague prompt produces a generic response; a specific, contextualized prompt produces a more useful one. Providing context ("I'm a high school student writing a persuasive essay for English class"), specifying constraints ("keep it under 300 words"), and asking for a particular format ("give me three bullet points") all help the model produce something more relevant. You can also ask the model to explain its reasoning, request alternatives, or tell it that a previous answer was wrong and ask it to reconsider — AI assistants are designed to respond to follow-up and correction within a conversation.
The ethical dimensions of AI use are increasingly consequential. AI-generated text, images, and code can be submitted as one's own work in academic or professional contexts — a form of misrepresentation regardless of whether a technical policy forbids it. AI models are trained on vast amounts of existing creative and written work, raising unresolved questions about attribution and compensation for the original creators whose work shaped the model's outputs. AI systems also inherit the biases of their training data: if a model was trained on text that overrepresents certain perspectives or underrepresents certain communities, those biases appear in its outputs in subtle ways. Literacy in AI tools means engaging with these questions, not just knowing which button to click.
No topics depend on this one yet.