Using AI responsibly
AI is one of those subjects where it can feel like everyone expects you to pick a side. Either it is going to solve everything, or it is going to ruin everything. I do not really buy either extreme. I think AI can be useful, but how we use it matters a lot.
For me, responsible use comes down to a simple idea: AI should help with your work, not take responsibility for it.
Madhav-Malhotra-003, CC0, via Wikimedia Commons
You should audit its work
If an AI gives you an answer, you are still responsible for checking it before you rely on it. AI tools can sound confident while being wrong, miss important context, or produce something that looks polished but does not actually work.
That means reviewing the result, checking claims against trustworthy sources, and testing anything that can be tested. If AI suggests a code change, for example, you should understand what it changed and run the relevant tests. Passing tests are useful evidence, but they are not a substitute for understanding the change or considering cases the tests do not cover.
The same applies outside programming. A convenient answer is not automatically a correct one, and handing the result to someone else does not transfer your responsibility for it.
Debian made a similar point in its General Resolution on LLM usage. The winning option, “Responsible Use of Generative AI,” does not ban or endorse the tools, but expects contributors to meet the same standards either way. They remain responsible for the quality and legality of what they submit, and should understand, review, and test AI-assisted work before including it. Blindly accepting generated material does not meet that bar. The vote applies to contributions to Debian, not everyone’s personal work, but I think the core idea holds more broadly: use the tools if they help, and make sure you can stand behind what you put your name on.
Use it to revise human-made work
I am more comfortable using AI to improve work that a person has already created. Grammar and spelling checks are a straightforward example: you write what you mean, then use a tool to catch mistakes or suggest clearer wording. You still decide which suggestions preserve your meaning and which do not.
There are similar uses in technical work. AI can help generate or review tests for code written by a person, point out a possible edge case, or suggest a small improvement. Those suggestions still need to be checked, but they can help make the revision process quicker and more thorough.
The important distinction for me is that the person remains the source of the work and the judge of the changes. Assistance is not the same thing as handing over authorship.
Keep your own ideas in the work
AI should be another tool in your belt, not a replacement for your input, ideas, or creativity. A tool can make parts of a task easier, but it cannot decide what you care about or what you are trying to say. Those choices should still come from you.
If you ask AI to produce everything and accept the result without engaging with it, you may end up with something technically complete but strangely detached from your own perspective. That might be fine for some low-stakes, routine tasks, but it is not how I want to approach writing, making things, or solving problems that matter to me.
I don’t think using AI automatically makes work less meaningful. If you’re still thinking things through and making the decisions, it can be a pretty handy tool. Just don’t hand over the whole job and call it a day. Keep enough of yourself in the work that you can look at it and say, “Yep, that’s mine.”
Have thoughts or feedback?
Join the discussion on Mastodon
