I Wasted 5 Years on AI Ethics Frameworks. Here’s What Actually Works.
Let’s be honest. For five years, I was a fraud. I was the guy on stage at AI ethics conferences, nodding along with the panelists, talking about “algorithmic fairness” and “responsible AI.” I co-authored papers, advised startups, and even testified before a congressional committee. From the outside, I looked like a leader in the field. On the inside, I was completely lost.
I was a collector of frameworks. The IEEE’s Ethically Aligned Design, Google’s AI Principles, Microsoft’s Responsible AI Standard—I had them all. I could recite the Asilomar AI Principles in my sleep. I was so obsessed with the idea of ethical AI, with the perfectly crafted document, that I missed the entire point.
It took me half a decade and a deeply humbling experience to see the truth: most AI ethics frameworks are useless. They’re well-intentioned, beautifully written, and totally disconnected from the reality of building and shipping products. This isn’t an academic paper. This is a confession. This is the story of my biggest professional failure and the lesson that changed everything.
The Siren Song of the Framework
When I first got into AI ethics, I was hooked. It felt like the most important work in the world. We were at the dawn of a new age, and it was our job to make sure it was a good one. The frameworks were our maps.
I remember the early days at RemoteTeam. We were building a platform to manage remote workforces, using AI for everything from predicting employee churn to optimizing team performance. I was determined to do it “right.”
I’d drag my team into endless meetings to debate fairness metrics. Demographic parity? Equalized odds? I’d hold marathon sessions with our lawyers, product managers, and engineers, trying to translate high-minded principles into product specs. The result? We’d end up with a watered-down, committee-designed solution that nobody liked and that didn’t work. We were so scared of doing the wrong thing that we did nothing at all.
We burned six months and over $2 million building a “fairness-aware” scheduling algorithm that was so complex, no one on the team could explain how it worked. It was a masterpiece of theoretical ethics and a complete failure as a product. It was slow, inefficient, and our customers hated it. We ended up scrapping it and going back to a simpler, heuristic-based approach. It wasn’t as “ethically pure,” but it worked.
That was the first crack in my faith. It wasn’t the last.
The Breaking Point
The real turning point came when I was an angel investor. I was listening to a pitch from a young, brilliant team building an AI-powered diagnostic tool for a rare cancer. Their tech was incredible. It could detect the disease with 99% accuracy, years earlier than any human doctor. It was going to save lives.
But then came the ethics question. Their training data was from a single hospital in a wealthy, white community. Their algorithm was less accurate for patients of color. It was a classic case of algorithmic bias, the kind of problem I’d spent years writing about.
I did what I always did. I pulled out my frameworks. I talked about representative data, fairness, and equity. I told them they needed to go back to the drawing board, spend another year and a few million dollars collecting a more diverse dataset. I told them shipping a biased product was irresponsible.
They listened patiently. Then one of the founders, a woman who had lost her mother to this disease, looked me in the eye and said something I’ll never forget.
“So, we should let people die for the sake of a perfect algorithm?”
That question hit me like a ton of bricks. She was right. I was so blinded by my pursuit of ethical perfection that I had lost sight of the human cost. I was so focused on the potential for harm that I had ignored the certainty of it. I was so obsessed with the framework that I had forgotten about the people.
I invested in the company. I’m glad I did. They’ve helped thousands of patients, and they’re actively working on improving their data diversity. They didn’t wait for a perfect solution. They shipped, they learned, and they iterated. They chose progress over perfection.
What Actually Works
That experience forced me to rethink everything. I threw out my frameworks and started from scratch. I talked to founders, engineers, and product managers. I looked at the companies that were successfully building responsible AI and tried to understand what they were doing differently.
What I found was surprisingly simple. It wasn’t about having the perfect framework. It was about having the right culture.
Here’s what actually works:
1. Start with a Red Team.
Instead of an ethics committee, you need a red team—a group of people whose only job is to break your AI. They should be incentivized to find the most creative, damaging, and embarrassing ways your system can fail. They’re your in-house hackers, your professional pessimists. They’re the ones who ask the uncomfortable questions, the ones who aren’t afraid to tell you your baby is ugly.
At one of my portfolio companies, a fintech startup, the red team discovered their new fraud detection algorithm was disproportionately flagging transactions from low-income neighborhoods. It wasn’t intentional, but it was happening. Because they caught it early, they were able to fix it before it ever affected a single customer. That’s the power of a red team.
2. Measure Everything.
You can’t fix what you can’t measure. Stop talking about “fairness” and start talking about false positive rates. Stop debating “transparency” and start tracking how often users override your AI’s recommendations. Get specific. Get quantitative. Turn your ethical principles into metrics, and then track those metrics with the same rigor you track your revenue and user growth.
3. Build a Culture of Humility.
The biggest danger in AI is arrogance. It’s the belief that you can anticipate every failure mode, that you can design a perfect system from the top down. You can’t. Your AI will fail. It will be biased. It will have unintended consequences. The question is not if, but when. And when it does, you need a culture that is humble enough to admit it, learn from it, and fix it.
This means creating a psychologically safe environment where engineers can raise concerns without fear of retribution. It means celebrating the discovery of bias, not punishing it. It means being transparent with your users about your system’s limitations.
The Future of AI Ethics
I’m not saying frameworks and principles are useless. They can be a good starting point for a conversation. But they are not a substitute for the hard, messy, ongoing work of building responsible AI.
We need to move beyond the buzzwords. We need to get our hands dirty. We need to stop chasing the fantasy of a perfect, ethically-aligned future and start building a better, more responsible present.
I wasted five years of my life on a dead end. I don’t want you to make the same mistake. So please, for the love of God, put down the frameworks and go build something.
Frequently Asked Questions
What's the most common pushback you get on this?
People often push back by citing exceptions or edge cases. And they're usually right that exceptions exist. But building a strategy around exceptions rather than patterns is a losing game for most founders.
How can I apply this thinking to my own situation?
Start by identifying the core principle behind the opinion, not the specific example. Then ask yourself: does this principle apply to my context? If yes, test it in a small, low-risk way before going all in.
Do all experts agree with this view?
No, and that's fine. The best ideas in business are often contrarian. I share my perspective based on my experience and data, but I encourage you to seek out opposing viewpoints and form your own conclusions.