You can't have a complete, honest conversation about AI without talking about its ethics. This module exists because leaving that out would be dishonest — so here it is, in my own words.
I am not here to convince you to use AI.
This whole site explains how these tools work and how I use them — but explaining is not the same as selling. If your ethical compass points away from AI, I respect that completely. Nothing here is meant to talk you out of where you've landed, and I respect any colleague who reads all of this and decides it's not for them.
What I won't do is pretend the concerns aren't real. So before anything else, here they are — named plainly, not explained away.
Each of these is a genuine, unresolved question. I'm not going to settle them for you — I just want them on the table, honestly.
Training and running these models consumes real electricity and water. It's invisible at the prompt, but it isn't zero.
Data workers and content moderators — often underpaid, often in difficult conditions — labeled and filtered what these models learned.
A model reflects the data it was trained on — including the imbalances and prejudices baked into that data — and can repeat them confidently.
Much training data was gathered from the open web without the explicit consent of the people who made it. Creators are right to ask hard questions.
What you type may be retained, reviewed, or used to improve a model. Treat sensitive patron and personal data with real care.
Serious people disagree about how risky advanced AI could become. I don't claim to know — but I won't wave the question away.
The law is unsettled. Regulation varies by country and state, and much of it is still being written — or isn't written at all.
A handful of large companies control the most capable models, the compute, and the data. That concentration deserves scrutiny.
Not everyone has the devices, connectivity, or digital literacy to benefit — and AI can widen those gaps as easily as it narrows them. For libraries, whose mission is equitable access, this one cuts especially close.
The environmental cost is the one that's easiest to forget, precisely because you never see it. Pick the kind of request you might make and get a rough sense of its footprint.
These are rough, illustrative estimates — real figures vary widely by model, data center, and region, and a single request is small. The point isn't the exact number; it's that the number isn't zero, and it adds up across millions of requests. For grounded measurements, the EcoLogits calculator does the real math.
I'd rather point you to people thinking carefully about these questions than hand you my conclusions. These are worth your time.
The trajectory of my career has given me a front-row seat to the birth of the internet and the World Wide Web. There were ethical considerations then too — different from today's, but real and seriously debated.
Back then, I made the decision to use the internet and the web to provide valuable service to our library customers. I'm making the same decision now about AI. I've weighed the concerns on this page honestly, and I've chosen to use this technology to give something genuinely useful back to our profession.
That's my decision, not a prescription for yours. Wherever you land, I'm glad you read this far.