You most likely used artificial intelligence this morning. Maybe you used it before you got out of bed.
When your phone autocorrected a typo, that was AI. When Spotify queued up something you actually wanted to hear? AI. When your email sorted the newsletter into Promotions before you ever saw it, that was AI too. None of it felt remarkable, because none of it was supposed to. It just worked.
So why does the word “AI” feel like it landed from another planet roughly two years ago?
Because something different arrived and we kept using the same word for it.
Language rarely keeps pace with technology, thus it creates a problem when one word carries five or six distinct technologies on its back and the conversation about that word becomes generic. People aren’t wrong to have strong feelings about AI. They’re just often reacting to all of it at once, which means they’re not quite reacting to any of it specifically.
The most useful thing we can do right now, for ourselves, for our organizations, for our teams, is get more specific. Specific about which kinds of AI deserve more of our attention and scrutiny, and which might deserve a little less.
Assistive AI is where most of us started, even if we didn’t know. Spellcheck. Autocorrect. Grammar tools. Voice-to-text. These systems are trained to catch errors, complete patterns, and smooth out friction. A bad outcome here is small and visible: autocorrect mangles a word, you see it, you fix it. Nothing downstream is affected. The trust we extend to these tools is largely appropriate, because mistakes and bias (like autocorrect changing Minal to Milan) is easy to catch and correct..
Predictive AI is more consequential, but considerably less visible. These are the systems analyzing patterns in behavior — yours and millions of other people’s — to anticipate what will happen next. Your Netflix queue is a low-stakes version, and a bad outcome is a bad recommendation, which costs you twenty minutes. But the same logic powers credit scoring algorithms, automated hiring screeners, and insurance risk models and in those domains, a bad outcome leads to someone who doesn’t get the loan, doesn’t get the interview, and never finds out why. This form of AI requires regulation and oversight because of the systemic downstream impacts it can have in creating an unjust and unfair society.
Generative AI: text and conversation. This is the category that cracked the AI conversation open. Tools like ChatGPT and Claude take a prompt and produce original language in response— drafting emails, summarizing documents, answering questions, and working through problems. The problem is these models don’t say “I don’t know”. They are embedded with unflappable confidence and produce something that sounds right, even if they end up hallucinating data and information. Using these tools asks us to think through questions of authorship, accuracy, and voice ahead of time. Smart users of AI can embed instructions like “Do not make anything up and cite all sources for data” to avoid this negative impact, and ensure all created content is reviewed by humans with subject matter expertise before publishing. After all, deciding how you’ll handle a problem before it surfaces is considerably simpler than explaining it to a funder after it does.
Generative AI: image, video, and audio. Same underlying idea, but a bad outcome has a more pernicious impact in an important way. With text, we’ve always understood that words can mislead. But humans are wired to treat visual evidence differently. “Seeing is believing” reflects something real about how we process images and video, and synthetic media can exploit that directly. The bad outcome here isn’t an image that looks fake. It’s an image that looks real while depicting something that never happened. And unfortunately, there are no reliable detectors of AI usage. Some of us have developed enough media literacy to maybe be skeptical of certain images, but this outcome will have a very real and detrimental impact on our society as it becomes harder and harder for people to know where to go for quality, true information, whether it’s the news or how to file a tax extension. s.
Agentic AI is the newest and least settled category, and the one most people haven’t encountered directly yet. While generative AI responds to you, agentic AI acts for you. These are systems that can take autonomous, multi-step actions on your behalf (researching, scheduling, executing tasks) without requiring your input at each step. A bad outcome would result in actions taken on your behalf against your interest, and by the time you notice something went wrong, irrevocable damage may have occurred to your finances, relationships, or professional standing. This form of AI use is not widespread yet (and you can currently opt out) but it’s worth knowing what’s coming and organizing with other folks for consumer protections.
The concerns people bring to AI conversations are real — the environmental costs, the labor questions, the risks to privacy and trust. And it makes sense that generative AI is where most of those concerns are focused. Generative AI companies raised $56 billion from investors in 2024 alone— more than double 2023. That’s where the money is, which is where the attention is, which is where the legitimate scrutiny belongs. But it’s worth remembering that generative AI is one part of a much larger map. And the concerns that apply to image generation don’t automatically apply to your spam filter.
Whether you realize it or not, you’ve already been making distinctions. Autocorrect feels different from deepfakes because it is different. The negative impacts aren’t even in the same neighborhood and neither are the data concerns or environmental cost.
When a donor raises concerns about AI in your communications, are they worried about ChatGPT drafting your newsletter, or AI-generated scenes that didn’t happen? Those are different concerns and they deserve different answers. The map you now have makes it possible to ask — and answer — the right question.
If you’re working through what AI adoption and use means for your organization- what to adopt, what to question, or what to communicate to stakeholders— that’s the kind of thinking we do every day at Brevity & Wit. We’d be glad to do it alongside you. Start with our Conscious AI Primer and then reach out!
This article was written by Sage Eubanks with the help of claude.ai. As a small business, we thoughtfully engage artificial intelligence to improve our efficiency, while also using human experts and reviewers to ensure quality and mitigate bias. Photo by Yaroslav Shuraev.