AI Has Risks. Francine Marie Bautista Says the Answer Is to Learn How to Use It Properly

There is an easy way to defend artificial intelligence: pretend the risks are exaggerated. Francine Marie Bautista does not take that route. When our conversation moved from AI-shaming to the actual dangers associated with the technology, she did not argue it was harmless. Instead she made a different point: “If you know how to use AI and use it properly, then you can minimize its negative effect.”

That sentence carries an important distinction. Minimize does not mean eliminate. Understanding AI will not make every problem disappear. It will not eliminate misinformation, privacy problems, bias, manipulation or overdependence. But it can change how vulnerable a person is to those problems, which makes AI literacy not merely a productivity skill but part of risk management.

The stereotype of the enthusiastic AI user is someone who believes everything the machine produces. But knowledgeable AI use should actually produce the opposite behavior. The better you understand generative AI, the more aware you become that confidence is not proof of correctness. NIST’s Generative AI Profile was created specifically to help organizations identify and manage risks associated with generative AI. That means one of the most useful AI skills is surprisingly old-fashioned: doubt. Check the answer. Verify the source. Ask where a number came from. Do not publish a citation you have never opened. Distinguish between what is known and what is inferred. If the matter is legal, medical, financial or otherwise consequential, do not treat a chatbot as the final authority. Bautista’s idea of a competent AI user is not someone who trusts AI more. It is someone who knows exactly why they should sometimes trust it less.

This is where the difference between access and literacy becomes obvious. Anyone can ask an AI system a question. The more demanding skill is evaluating the answer. Does it make sense? Can it be independently verified? Is the information current? Did the AI misunderstand the question? Did it invent a source? Is the conclusion stronger than the available evidence? Those questions turn AI from an authority into an instrument, and the human stays responsible. No prompt, however sophisticated, guarantees truth. Good prompting can improve an answer. It cannot remove the need for judgment.

AI literacy also means understanding that not everything belongs in an AI system in the first place: confidential company material, private client information, passwords, sensitive contracts, medical or financial records, identifying information about other people. A responsible user asks whether that information needs to be shared at all. Can names be removed? Can the relevant section be summarized instead? Can sensitive details be replaced with placeholders? What are the platform’s actual data controls? UNESCO’s guidance on generative AI has similarly emphasized data privacy and the need for human-centered, safe and meaningful use. The technology carries privacy risks, but user behavior can either increase or reduce unnecessary exposure to them.

One of the strongest criticisms of AI is that it can make people intellectually lazy, and Bautista does not dismiss that concern either. If a student asks AI to complete every assignment without understanding the lesson, something real is lost. If a professional automatically accepts every recommendation AI generates, judgment weakens. If a writer cannot write without asking a machine what to think first, dependence becomes a reasonable worry. But that is not the only way to use the tool.

AI can also be used to interrogate your own thinking. Ask it for the strongest argument against your position. Ask what assumptions you might have missed. Ask it to explain something at increasing levels of complexity. Compare your reasoning against an alternative approach. Use it to organize research you then go verify yourself. Use it as a tutor rather than a substitute student. UNESCO has emphasized critical thinking and digital literacy as important for evaluating AI-generated content and interacting with generative AI safely and effectively. The same technology can produce two very different habits: dependence, or deeper inquiry. The difference is largely in the user.

There is another misconception worth addressing here: that if AI becomes sufficiently powerful, expertise eventually becomes unnecessary. Bautista’s argument points the opposite direction. Expertise is what helps you recognize when AI has failed. A photographer understands when generated lighting is implausible. A designer recognizes when a concept is visually polished but strategically meaningless. A researcher notices when the evidence does not support the conclusion. A programmer recognizes dangerous code. A lawyer can identify a fabricated case. Someone with no understanding of the field may not even know which mistakes to look for. AI does not necessarily eliminate the value of knowledge. In many contexts, knowledge becomes the supervision layer sitting on top of it.

A person who truly understands a tool does not use it for everything, and that applies to AI as much as anything else. Not every conversation should be automated. Not every message should be generated. Not every photograph needs generative alteration. Not every difficult emotion requires a chatbot. Not every research question should begin and end with a model. Not every creative imperfection needs correcting. Sometimes the human process itself is the valuable part. Sometimes learning requires struggling through the problem. Sometimes authenticity matters more than efficiency. Sometimes professional expertise is simply required. Sometimes the correct use of AI is deciding not to use it at all. That is not technological rejection. It is technological maturity.

AI also creates a problem that gets less attention than it deserves: it makes production extremely easy. Someone can generate dozens of captions, hundreds of images, multiple articles and endless variations at extraordinary speed. That sounds like pure productivity until everyone is doing it, and the result becomes an ocean of generic material competing for shrinking human attention. Bautista’s argument about responsible use extends into restraint. AI can help you produce more. Human judgment should still decide whether more actually needs to be produced. The ability to generate something does not create an obligation to publish it. Increasingly, taste means knowing what to delete.

Companies have responsibilities. Developers have responsibilities. Governments, schools and institutions have responsibilities. Standards and safeguards matter. NIST’s AI Risk Management Framework itself is built around the idea that AI risk has to be identified, assessed and managed throughout how systems are designed, deployed and used. But users are part of that system too. A population that understands AI is harder to fool with AI. People who know synthetic media exists scrutinize suspicious material differently. People who know models can fabricate information verify claims before repeating them. People who understand privacy hesitate before uploading sensitive documents. People who understand AI’s limitations are less likely to outsource every important decision to it. Education does not solve every risk. But ignorance certainly does not solve them either.

This is perhaps the artificial binary Bautista most wants to break: the idea that society only has two options, embrace AI without question or reject it outright. There is a third possibility: learn it, understand it, regulate it where appropriate, question it, develop norms around it, teach people how it fails, teach people when it helps and teach people what has to remain human. That approach is less dramatic than either technological worship or technological panic. It is also probably more useful.

By the end of our discussion, Bautista’s three positions fit together into one coherent stance. She rejects the idea that using AI automatically makes a creative professional less talented. She refuses to let her enthusiasm for AI be interpreted as automatic support for Pax Silica. And she does not respond to AI’s negative effects by pretending they do not exist. The common thread running through all three is judgment. Do not condemn a tool without understanding how it is being used. Do not endorse a policy simply because it carries the language of technological progress. Do not trust artificial intelligence simply because it produces an impressive-sounding answer. Understand first. Then decide.

That may be the most useful principle in the entire AI debate, because the goal was never to make human beings obedient to artificial intelligence. The goal is to make human beings knowledgeable enough to remain in control of it.