Ethical Design in AI Products: Moving Beyond Dark Patterns Toward Genuine User Empowerment
Kishan Mendapara
The AI boom has given designers new superpowers. It's also given us new responsibilities that our field is, on the whole, not taking seriously enough.
Dark patterns predate AI by decades. Infinite scroll, confirmation shaming, hidden unsubscribe buttons, misleading default settings — these were already ethically fraught. AI doesn't just continue that tradition. It accelerates it, personalizes it, and makes it harder to detect.
What AI Changes About Manipulation
A traditional dark pattern is static: the same misleading UI is shown to every user. An AI-personalized dark pattern is dynamic: it shows each user the specific framing most likely to drive the desired behavior, based on observed psychological patterns. This is qualitatively different. It's not just a misleading UI — it's a misleading UI that was specifically designed for you.
The question isn't whether AI can be used to manipulate users. We know it can. The question is whether you, as the designer, are willing to be the person who decides it won't.
A Framework for Ethical AI Design
I use four tests when evaluating AI-powered features:
The Transparency Test — If users fully understood how this feature worked, would they still use it? If the answer is no, the feature needs redesigning, not better obfuscation.
The Benefit Test — Does this primarily benefit the user, the company, or both? If it primarily benefits the company at the user's expense, it's a dark pattern regardless of how it's framed.
The Override Test — Can users easily understand and change what the AI is doing? Genuine empowerment requires real agency, not the appearance of it.
The Accountability Test — When the AI is wrong, is it clear who is responsible and how to seek recourse? Opacity in AI systems is itself an ethical failure.
The Business Case for Ethical AI Design
The ethical argument is reason enough. But for teams that need a business case: products built on genuine user value survive regulatory scrutiny, advocacy, and competitive pressure far better than those built on manipulation. The companies with the longest term track records in SaaS — Basecamp, Notion, Linear — have all built on trust rather than extraction. AI changes the leverage, not the underlying principle.
We are early enough in the AI design era that the patterns we establish now will become defaults. The responsibility to get them right sits with every designer shipping AI features today. That includes you.
