AI is not a thing. It never was. Key distinctions exist: logic-based systems, machine learning models, and generative AI. Treating AI as a monolith—whether as a miraculous gift or an existential threat—distorts our understanding of what’s happening and why it matters. The benefits and risks depend on the type of AI, who uses it, and the context of use. Let’s slow down and examine this carefully.
There Are Many Different Types of AI, and That Difference Matters
When most people say “AI” right now, they mean generative AI—tools like large language models, image generators, and video synthesizers. The kind that produces text, art, music, and code on demand. The kind that has caused, and continues to cause, real and serious harm to real people. We’ll get there.
There’s also AI used in medical diagnostics—tools that can detect certain cancers and diseases earlier than human observation alone, that help researchers model protein folding, and that support drug discovery in ways that are genuinely saving lives. (You et al., 2022) There’s AI used in accessibility technology, in climate modeling, and in pattern recognition for scientific research.
These are not the same thing. They don’t have the same impacts, ethical profiles, or implications for how we live and work. Grouping them together because they’re both called “AI” is a little like grouping a scalpel and a machete because they’re both sharp. The category is technically accurate. It is also almost completely useless for understanding what we’re actually dealing with.
This distinction matters because it shapes the conversation we need to have. Concerns about generative AI are real and worth taking seriously: copyright theft, job displacement, environmental cost, and misinformation. These concerns do not automatically extend to every application of machine learning. When we conflate them, we either end up dismissing legitimate concerns or demonizing tools that are doing genuine good. Neither outcome helps anyone.
The Harms Are Real, and They Deserve to Be Named
I don’t want to soft-pedal this. Some of what has been done with generative AI specifically is genuinely bad.
Artists, writers, and other creators have had their work used without their consent to train models that then produce work in their style, effectively competing with them by using their own creative labor as raw material. This is not hypothetical damage. It’s happening, it’s documented, and the people it is happening to are largely people who were already economically vulnerable: freelancers, independent artists, creators who built their livelihoods piece by piece over years. If you’re a writer traversing this landscape, understanding how AI intersects with human originality is an important part of making informed decisions about your own practice.
Deepfakes have been used to put real people’s faces and voices into content they never consented to, including non-consensual sexual content. Real people have had their likenesses weaponized against them. This is not a fringe use case.
AI-generated misinformation and disinformation have complicated an already fragile information environment. Fabricated images, synthetic audio clips, entirely invented news stories; these spread fast and stick. The cognitive strain of handling it all is real, and it falls hardest on people who already deal with a lot. (Apuke et al., 2024, pp. 849-860)
The environmental impact is significant and largely underreported. Training large AI models requires enormous amounts of energy and water for server cooling. This is not speculation; the data centers exist, resource consumption is measurable, and the costs are borne disproportionately by communities near those facilities. (Danelski, 2023) When companies downplay this or decline to disclose it, that’s worth paying attention to. The International Energy Agency’s research on data centers and energy consumption provides a useful grounding in what the numbers actually look like.
Job displacement is complicated and ongoing. Some of this reflects genuine economic anxiety about what happens to people whose work can now be replicated cheaply at scale. Some of it has already moved past anxiety into lived reality for people in certain industries. The “new opportunities will emerge” argument is not wrong exactly, but it tends to be made by people who won’t personally bear the costs of the transition, and the workers being displaced right now do not have the luxury of waiting for those opportunities to materialize.
None of these harms is an inevitable feature of all AI. They are specific outcomes of specific choices made by specific people with significant power and not enough accountability. That distinction also matters.
The Benefits Are Also Real, and Deserve Honest Acknowledgment
I said I wasn’t going to be one-sided, and I meant it.
AI-assisted medical diagnostics now detect diseases earlier than traditional methods. In some cases, early detection can mean effective treatment rather than a much harder road. Machine learning-powered research tools are accelerating scientific discovery at speeds once impossible. Accessibility tools — screen readers, real-time captioning, predictive text—use AI to expand what is possible for people with disabilities. (Office, n.d.)
For somFor some neurodivergent people, certain AI tools are genuinely useful. They’re not a replacement for human skill or connection, but act as scaffolding. They help organize thoughts, get unstuck on a piece of writing, or process information in a different format. I am careful about how I use these tools and thoughtful about the tradeoffs, but I won’t pretend they’ve offered nothing. The point is not that AI is good. The point is that the question “Is AI good or bad?” is unanswerable because AI is not a single thing, and goodness is not binary. What we can do is look at specific applications, specific impacts, and specific trade-offs —and make more honest assessments from there.
Politics, Misinformation, and Why This Got So Messy
Part of why it’s hard to think. Some of the loudest boosterism comes from people with significant financial stakes in AI’s rapid, unregulated growth.ple sides.
Some of the loudest boosterism comes from people with significant financial stakes in AI’s rapid, unregulated growth. When executives at AI companies dismiss concerns about copyright, environmental impact, or labor displacement as overblown or technophobic, ask what they stand to gain from that. The loudest alarmism is similarly shaped by interest, whether that’s political positioning, grift dressed up as advocacy, or genuine but poorly-researched fear that has outpaced the actual evidence.
Political polarization has shifted the conversation to serve narratives over understanding. AI is now a culture-war topic, which makes it harder to think about clearly. Much of what circulates about AI—good and bad—is more emotionally compelling than accurate. (Lühring et al., 2024)
This is where critical thinking becomes both important and hard. Not as a buzzword, but the real, sometimes tedious work: checking sources, following citations, asking who funded a study, considering what someone or a company stands to gain, and noticing when something seems too perfect to be true. Evaluating and citing sources is a practical skill. It matters even more now, when it’s easy to generate plausible but false claims.
When you read a claim about AI—any claim—it’s worth asking: Who is saying this? What do they know? What are they leaving out? What do they stand to gain from my believing it?
That’s not cynicism. That’s the homework. And right now, it’s homework that matters.
How to Actually Think About This
I’m not going to give you a tidy framework. I don’t think tidy frameworks serve this topic well. I can offer how I try to approach it. to be specific. AI is destroying creativity” is a less useful thought than “generative AI trained on scraped creative work without consent is creating economic and ethical harm for working artists.” The second version is more accurate, more actionable, and more honest about where the real problem lies.
I look for the people most affected. On AI and labor, I pay more attention to what displaced workers say than what CEOs do. When it comes to AI and creativity, I care more about what artists and writers are experiencing than what a tech company’s press release claims. People with the least power and the most skin in the game are usually the most clear-eyed. to hold complexity without needing to resolve it into a simple verdict. Something can be genuinely useful in one context and genuinely harmful in another. Both things can be true. Living with that tension is uncomfortable, but collapsing it into false simplicity doesn’t actually make it go away.
I also try to stay honest about my own stakes. I’m a writer and content creator. Generative AI has direct implications for my work and livelihood. That’s real. I don’t pretend otherwise. But I also try not to let that make me unfair to all of machine learning. That wouldn’t be accurate.
Where That Leaves Us
We are living through a significant technological shift. We are also doing so in an information environment that actively undermines clear thinking. The incentives to oversimplify, sensationalize, and polarize are enormous. (Jungherr & Rauchfleisch, 2024) Those best positioned to navigate this are willing to do the slower, less exciting work of truly understanding what’s in front of them.
The harms of certain AI applications are real and worth taking seriously, and we should fight for regulation around them. The benefits of other AI applications are also real and worth protecting. The conversation we need to be having is specific, honest, and grounded in the actual impacts on actual people, especially the people with the least power.
That conversation is harder than picking a side. It’s also the only one that has any hope of getting us somewhere worth going.
References
You, Y., Lai, X., Pan, Y., Zheng, H., Vera, J., Liu, S., Deng, S. & Zhang, L. (2022). Artificial intelligence in cancer target identification and drug discovery. Signal Transduction and Targeted Therapy 7. https://doi.org/10.1038/s41392-022-00994-0
Apuke, O. D., Omar, B., Tunca, E. A. & Gever, C. V. (2024). Information overload and misinformation sharing behaviour of social media users: Testing the moderating role of cognitive ability. Journal of Information Science 50(6), pp. 849-860. https://doi.org/10.1177/01655515221121942
Danelski, D. (April 27, 2023). AI programs consume large volumes of scarce water. UCR News. https://news.ucr.edu/articles/2023/04/28/ai-programs-consume-large-volumes-scarce-water
Office, U. G. (n.d.). Artificial Intelligence in Health Care: Benefits and Challenges of Machine Learning Technologies for Medical Diagnostics. https://www.gao.gov/products/gao-22-104629
Lühring, J., Shetty, A., Koschmieder, C., Garcia, D., Waldherr, A. & Metzler, H. (2024). Emotions in misinformation studies: distinguishing affective state from emotional response and misinformation recognition from acceptance. Cognitive Research: Principles and Implications 9. https://doi.org/10.1186/s41235-024-00607-0
Jungherr, A. & Rauchfleisch, A. (2024). Negative Downstream Effects of Alarmist Disinformation Discourse: Evidence from the United States. Political Behavior 46. https://doi.org/10.1007/s11109-024-09911-3
Leave a Reply