Last week, I attended the Minnesota Evaluation Association unconference, a wonderful learning experience that included one session I found frustrating. It was called “Is AI Worth It?”. The presenter arrived with a resounding no rather than a null hypothesis. The presentation three hugely complicated subjects: bias, the environmental costs of data centers, and security, moving through them rapidly with examples that felt too selective to support the broad conclusions. I write about these concerns myself, and I welcome skepticism about both the technology and the companies selling it. What bothered me was the sense that the answer had already been settled and everyone in the room was expected to agree. I wanted to explore how the answer might change depending on the task, the people affected, and the alternatives available. Useful applications deserved a place in that conversation, along with the reasons to avoid them.
Someone called another AI session afterward. The format was more of a conversation than the previous one-way presentation had been and we shared tools and methods. People described what they were trying, where AI had helped, and what they were learning. In my own evaluation practice, I’ve used it to develop accessible materials and participatory tools that help people shape an evaluation. It can also help us work through qualitative responses and larger bodies of information, provided we check the analysis and preserve context. Tools like Gemini Notebook can make dense material more accessible for some people with dyslexia by offering a way to engage with it through audio. For organizations and individuals facing constraints on time, money, or access, these benefits deserve serious consideration, even when they come with limitations.
I understand the weariness around AI, and I encounter it at work too. People are tired of slop, worried about their jobs, concerned about privacy, and uncomfortable with power accumulating in a handful of companies. Some of these problems existed before AI, but AI can make them worse. That gives us more reason to examine what is actually happening. Data centers support much of our digital life, and AI adds demand; possible environmental benefits need evidence and don’t erase present costs. Privacy requires attention to the information we enter, the systems we use, and the protections in place. Bias requires us to examine whose experience is represented and what happens when a system gets something wrong. A useful application can still carry unacceptable costs, but reaching that judgment takes more than a blanket statement about AI as a whole.
At work, people encounter frustrating tools without much support for defining a task, supplying context, or judging the result. It’s easy to request a complete deliverable in one prompt and receive a 26-page document. But that doesn’t make everyone more productive. Now someone else has 26 pages to read. Producing more material is hardly an accomplishment if it creates more burden. In my last piece, I described AI as a paintbrush, and that metaphor still feels useful: the person holding it needs purpose, practice, and judgment. We need opportunities to learn how to work in manageable steps, check claims, protect information, and decide whether the result serves anyone. That education should also help us recognize when to put the tool down.
I want schools, libraries, workplaces, and professional associations to take this need seriously, with independent leadership shaping what gets taught. In my earlier piece on big tech’s role in AI education, I explored the promise of companies helping teachers learn these tools, alongside the risk that training becomes product onboarding and schools grow dependent on particular platforms. That tension matters beyond the classroom. Companies benefit when we adopt their products and keep using them, while education needs room to question their claims, compare alternatives, and conclude that a particular use should be avoided. Educators, researchers, workers, and affected communities should help set those priorities. We also need to collectively figure out how to teach children within this changing landscape, giving them opportunities to use AI to ask questions, practice, and develop understanding alongside work that requires them to think without it. They need to recognize when assistance supports learning and when it bypasses the thinking they need to do. Adults need that discernment just as much. I want people to have enough knowledge to make choices they can explain and defend, including the choice to work without AI. That capacity takes thoughtful teaching, and we have plenty of reasons to begin.
Anthralytic builds practical software for mission-driven organizations. Our tools make evaluation, strategy, and learning more accessible, helping teams explore evidence, involve the people affected by their work, and make informed decisions.
