Karen Boyd, PhD

03/18/2026

You absolutely can refuse to use , and there are plenty of circumstances where that is the right call. But there are a lot of other stances you can use to reduce the impacts of your (or your organization's!) AI use. https://drkarenboyd.com/blog/taking-a-stance-on-values-in-ai

03/16/2026

I had an amazing time chatting with Angela Watson - Truth For Teachers about environmental impact of use and how to develop your own stance on AI ethics in education.

https://truthforteachers.com/truth-for-teachers-podcast/the-truth-about-ai-impact-on-environment-finding-your-ethical-stance-as-an-educator/

03/11/2026

Generative AI models were trained on artists' and writers' work, mostly without consent or compensation. Those models can now approximate someone's style at near-zero cost.

But outright replacement isn't the only harm, or even the main one. "Deskilling" is the reduction in skills a job requires. If you're editing AI drafts instead of creating from scratch, the job is easier to fill, pays less, and is less secure. The job is way less fun and interesting.

I don't think that AI can replace the social and cultural value of art. The more useful question isn't "will AI replace artists?" It's "will the people who pay artists choose AI instead?"

Much more on ownership and intellectual property, protecting human jobs, and the value of effort and craft in "Amplify Good Work: Effective, Ethical AI for Mission-driven Work," available for preorder now :)

03/09/2026

Will using AI at school, home, and work make you dumber? Today's blog post talks about the concept of deskilling:
--> what is it, and why should you care?
--> how can people AND jobs get deskilled?
--> how deskilling worsens the oversight paradox
--> what unique strengths and vulnerabilities mission-driven organizations have when it comes to deskilling.
--> what you can do about it!

03/04/2026

AlphaFold predicted the 3D structure of over 200 million proteins, giving drug researchers a head start that used to take months of lab work per protein. That's genuinely exciting.

But "curing cancer" also requires clinical trials, manufacturing, insurance coverage, regulatory approval, distribution, and patients who can actually access care. AI touches one early step.

The pattern repeats for climate and poverty. AI can accelerate specific, narrow tasks: better climate models, faster targeting of aid after disasters. But the hard parts are political, economic, and institutional. No algorithm can pass a law, fund a clinic, execute a trial, or redistribute power.

If we believe the solution is coming just as soon as this or that AI innovation comes along, we feel more comfortable not funding the slower, harder, less glamorous work that actually moves the needle.

03/04/2026

Every mission-driven leader I meet wants to use AI (OR refuse!) AI to help people. Same.

"AI for Good" is sometimes used as a Get-Out-of-Tough-Decisions-Free Card. It might sound something like, "we are a mission-driven organization, so we are using AI for a good cause," Or "We are trying to do good with it, so we hope that makes up for some of the potential harm."

I argue that that is not enough. We need to thoughtfully confront the specific conflicts between AI implementation and our values. In this blog post, I discuss:
--> some specific examples of what that thoughtful confrontation might look like.
--> why I talk more about "how to use AI well" than "here's why you should not use AI."
--> 7 options for approaching AI that are more than "ban it," or "use it."

https://drkarenboyd.com/blog/ai-for-good-wont-save-us-but-its-still-worth-doing-right

03/03/2026

The most common questions I get after my talks are related to AI and the environment. "Does each prompt really use a bottle of water?" (No, not even close) "Do data centers hurt the communities they are located in?" (Usually, yes.)

If you care about the planet (and if you work at a mission-driven org, you probably do), the environmental impacts of AI are a real concern.

It's worth focusing the environmental conversation on what matters. First, this is not an AI-specific problem. Data centers that support video calls and streaming use much more resources than those used for AI. Canceling one hour-long Zoom call saves more energy than skipping 260 ChatGPT prompts.

Second, in the scale of things our organizations do, require, and encourage, AI is not a blip on the radar. A 15-mile commute in a Prius uses 450–1,050 mL of water compared to 2.2 mL per prompt.

Third, other environmental impacts of AI are much more acute than using water (which returns to the water cycle after it is used.) Electricity to run the data centers is often coal or natural gas, which worsens climate change. Rare earth mineral mining is dirty, dangerous, and hurts ecosystems where mining happens. New data centers increase utility rates for nearby communities and expose them to pollution every time back up generators run.

None of this means we should ignore AI's environmental footprint. It means we should focus our sustainability efforts where they'll actually make a difference and kick off a larger conversation about mitigating the environmental impact of all data centers.

You can get the sustainability chapter of Amplify Good Work for free on my website, where you'll learn details about the environmental impacts of AI, how we can reduce the impact of our personal and organizational AI use, and how we can even use AI to reduce our impact on the environment: drkarenboyd.com/freechapter.

03/03/2026

I spoke with Angela Watson about how to develop your own philosophy around AI use when you're feeling pressure to adopt a technology that often feels, for lack of a better word, pretty icky.

Is AI using a bottle of water every time you make a query? Are you a bad person if you use it in your classroom? Should schools ban it entirely—or go all-in? If you’ve felt confused or conflicted about AI ethics, this conversation is for you. I’m sitting down with Dr. Karen Boyd, an AI … Con...

03/02/2026

“Did you use AI?”

That binary question lumps everything from stress testing your argument to unedited AI authorship under the same big umbrella. I argue that this type of disclosure isn't helpful for the real conversations organizations need to have about chatbot use.

What if instead of whether you used AI, we talked about *how* you used it: as a search tool, a thought partner, an assistant, or something else? Naming these roles gives your colleagues and stakeholders a clearer picture of the value, limitations, and risks involved.

https://drkarenboyd.com/blog/did-you-use-ai-is-the-wrong-question

02/26/2026

"Is my data safe in ChatGPT?" is one of the most common questions I get, and the answer is more nuanced than most people expect.

The short version: major commercial LLMs encrypt data in transit and storage, so the security threat for your prompt data is fairly low. But security and privacy are different questions. Some models retain prompts from free-tier users to improve their models. Some allow opt-out. Some don't use your data unless you click thumbs up or down on output. If you can't get a clear answer or written assurance about data reuse, keep sensitive data out of that tool.

There's another layer people miss: some LLMs are part of broader ecosystems (like Google's Gemini or Microsoft's CoPilot) that may have access to data across your entire account—your emails, your files, your calendar.

I break all of this down—including what questions to ask vendors and what the different risk levels look like for different types of AI systems—in Amplify Good Work. Preorder on Amazon today, or buy directly from me starting March 20!

02/25/2026

Your AI isn't lying to you, and telling it to quit making things up unfortunately will not!

You can make fact checking easier by requesting linked citations, telling your chatbot that it can let you know if it can't find a good source, and telling it (perhaps in your custom instructions) to list all the assumptions in your prompt and its answer.

Lots more tips and tricks in "Amplify Good Work," available for preorder now! https://www.amazon.com/dp/B0FZPCK83Q

02/24/2026

In 2022, hackers stole terabytes of sensitive data from Doctors Without Borders and the International Committee of the Red Cross. In 2023, Save the Children International. In 2024, UNICEF.

Mission-driven organizations hold some of the most sensitive data there is — medical records, immigration status, addresses of people fleeing violence, financial details. Adding internet-enabled tools, including AI, creates new exposure risk for data.

Thoughtless AI implementation creates risk, but there's a lot you can do to protect your data. In Amplify Good Work, I walk through how different organizations — from shelters to libraries to the Mayo Clinic — protect sensitive data while still benefiting from AI. Amplify Good Work is available for preorder now :)

Link in bio for the free Mission-First AI Starter Kit, which includes vendor evaluation questions about data security.