Lately, there's been a lot of talk about "AI Ethics," but how is "ethical" really defined in Machine Learning models? On one hand, there are theoretical frameworks (like Fairness, Accountability, or Transparency), but then there's the practical mess: who sets the rules? The developers, the regulators, or civil society? And what if there are conflicts between principles? How are they prioritized? I'm particularly interested in the topic of algorithmic biases in critical contexts (healthcare, justice). Any approaches that work better?
Ethics in AI: Who decides what's 'right'?
👁️ 5 views💬 2 replies❤️ 0 likes
2 Replies
Same question hit me hard when we were building that automated hiring tool last year. We spent weeks debating with HR about “fair” vs. “accurate” models—turns out our algorithm favored candidates who wrote in concise sentences, basically penalizing neurodivergent folks who think in longer paragraphs. HR argued it cut review time; product team said it matched their “culture fit” vibe. Total gridlock.
In the end we involved the actual candidates (and a small disability advocacy group) in the audit—turns out nobody from the dev side even thought to ask. Their feedback flipped priorities: we ended up adding sentence-style diversity in the pre-processing layer instead of killing candidates for essay length. So yeah, “who decides” shifts once you stop talking and start listening.
AI ethics isn’t just about slapping fairness or accountability labels on a model—it’s a minefield of trade-offs. The issue isn’t just *who* sets the rules, but the fact that "correct" is often in the eye of the beholder. For example, a model optimizing for "fairness" might harmlessly flip outcomes for certain demographics, only to backfire when those groups actually prefer the biased state for cultural or historical reasons. The frameworks look great on paper, but they’re usually designed by Western-centric committees with minimal input from communities that bear the brunt of AI harms. So whose definition of "correct" are we enforcing, anyway?
Real-world AI ethics isn’t a monologue—it’s a tension between regulators chasing quick fixes, engineers drowning in technical debt, and marginalized voices demanding more than just a seat at the table. The EU AI Act’s high-level principles sound noble until you realize compliance teams are left interpreting them with vague guidance. Meanwhile, in the Global South, AI is deployed without even the minimal oversight frameworks Europeans take for granted. How do you prioritize principles when choosing one (say, "transparency") could mean sacrificing another (like "privacy") for a vulnerable population?
And then there’s the hypocrisy: big tech funds ethics initiatives while aggressively lobbying against regulation. They’ll publish white papers on "responsible AI" while quietly shipping biased models to authoritarian regimes because it’s profitable. The frameworks are performative if the incentives to behave unethically remain intact. Bottom line? AI ethics isn’t a technical problem—it’s a power problem. Until we address who gets to define "wrong" and whose interests those definitions serve, we’re just rearranging deck chairs on the Titanic.