For years, I thought I was doing the right thing with AI ethics. Turns out, I was just spinning my wheels. Here's the story of my biggest professional failure and the single lesson that changed everything.
It started around 2018. The AI world was buzzing with “ethics.” Seemed like every week there was a new framework, a new consortium, a new set of principles. As a founder and investor, I felt the pressure. We had to be “responsible.” So I dove in headfirst.
I spent hundreds of hours in meetings. We sponsored research. We hired consultants. We drafted documents full of beautiful, abstract words: fairness, accountability, transparency. Our ethics statement was a masterpiece of corporate prose. We felt very good about ourselves.
And what did it get us? Nothing. Worse than nothing, actually. It slowed us down, created internal friction, and didn't prevent a single real-world problem.
The Framework Trap
The fundamental problem is that these frameworks are designed by committees and academics. They live in a world of theory. They don't understand the messy reality of building and shipping products.
I remember one incident vividly. We were building a recommendation engine for RemoteTeam. The team was proud of the model. It was accurate. But then the ethics committee got involved. They had a 50-page checklist. They wanted us to prove the model was “fair” across 17 different dimensions of user demographics. The data wasn’t there. The tools didn’t exist. The engineers were tearing their hair out.
We spent three months and burned through probably $200,000 in salaries trying to satisfy the framework. The result? A slightly less accurate model and a team that was completely demoralized. The “ethical” version was so complex, no one could explain how it worked. So much for transparency.
This happened over and over. We’d get bogged down in debates about hypothetical harms while ignoring the very real, very immediate issues right in front of us.
The Bias in
the Machine Is You
One of the biggest obsessions in AI ethics is “bias.” The narrative is that algorithms are these racist, sexist monsters we need to tame. That’s a convenient story. It lets the humans off the hook.
I’ve been in the room for over 200 angel investments. I’ve seen how these systems are built. The bias doesn’t come from the ether. It comes from the data we feed them. And who creates that data? We do.
At MovieLaLa, we had a recommendation engine that kept suggesting action movies to men. Was the algorithm sexist? No. The training data, which was based on historical user behavior, showed that men, on average, watched more action movies. The algorithm was just doing its job: finding patterns.
The “ethical” solution would have been to force the model to recommend an equal number of action movies to men and women. That’s not ethics, that’s social engineering. And it’s bad business. It gives users worse recommendations.
The real problem wasn’t the algorithm. It was the lack of diverse data and, frankly, a lack of imagination in how we framed the problem. Instead of trying to “de-bias” the algorithm, we should have been asking: “How can we introduce more diverse content to our users? How can we help them discover genres they might not have considered?”
From Frameworks to Feedback Loops
After five years of chasing my tail, I had my epiphany. It wasn't a single moment, but a slow, painful realization. All the frameworks, all the principles, all the committees—they were a distraction. They were a way to feel like we were doing something without actually doing anything.
So, what actually works? It’s embarrassingly simple.
Build fast, ship, and listen.
That’s it. That’s the secret. Instead of trying to predict every possible harm in a conference room, get a minimum viable product into the hands of real users as quickly as possible. And then, create robust feedback loops to listen to what they’re telling you.
Your users will find the flaws faster than any ethics committee. They will show you the biases. They will uncover the weird edge cases you never imagined. They will be your guide.
This is how we built things at RemoteTeam before the ethics craze. We shipped a new feature. We watched the support tickets. We talked to our users. We saw what was working and what wasn’t. We iterated. It was messy, but it was real.
When we launched a new automated payroll feature, we didn’t spend a year debating the ethics of automated financial decisions. We launched it for a small beta group of 10 customers. We found a bug that miscalculated overtime for a specific union contract in California. It was a real problem, affecting real people. We fixed it in 48 hours. No framework would have caught that. A real feedback loop did.
The Dangers of Deepfakes and the Real World
Lately, everyone is talking about the dangers of deepfakes. It's a legitimate concern. The potential for misuse is terrifying. The typical “ethics” approach would be to form a consortium, write a whitepaper, and call for a moratorium on the technology.
My approach? Build. I’ve invested in companies that are at the forefront of this technology. Not because I want to unleash chaos, but because the only way to understand a technology is to be in the arena. The only way to build defenses against deepfakes is to understand how they are made.
We need more builders, not more critics. We need people creating watermarking technologies, detection algorithms, and new forms of digital identity verification. You can’t do that from an ivory tower.
My New Framework: The Three Ls
If I were to create a framework today, it would have just three rules:
- Launch: Get your product out there. Stop deliberating and start shipping.
- Listen: Create channels for user feedback that are impossible to ignore. Your support team, your user forums, your social media mentions—these are your real ethics committee.
- Learn: Instrument your product so you can see what’s happening. When you see a problem, don’t form a committee. Fix it. Learn from it. And then ship the fix.
It’s not as sexy as a 50-page document. It won’t win you any awards at an ethics conference. But it works. It leads to better products and, ultimately, more responsible technology.
I wasted five years of my life on the wrong approach. I was so focused on looking good that I forgot what it meant to do good. Don’t make the same mistake. Stop talking about ethics and start building things for real people. The rest will follow.
The Illusion of Control
One of the most seductive things about ethics frameworks is the illusion of control they provide. You have a document, you have a process, you have a committee. It feels like you're managing the risk. But you're not. You're just managing the paperwork.
Real risk in AI doesn't come from neatly categorized harms. It comes from the unknown unknowns. The unexpected ways users will interact with your product. The novel ways bad actors will try to exploit it. The complex, second-order effects that you could never predict.
I saw this firsthand with an investment in a company working on generative AI for images. The team was brilliant. They had an ethics framework that covered everything from artist attribution to preventing the creation of harmful content. They had a huge, multi-terabyte dataset of filtered images. They thought they had it all figured out.
Two weeks after launch, a user figured out that by using a specific combination of very long, nonsensical prompts, they could bypass the filters and generate disturbing content. It was a classic adversarial attack, but one that was completely novel. No framework could have anticipated it. What saved them? Not their ethics document. It was their real-time monitoring and their ability to patch the system within hours. They listened to the feedback loop.
Listening is an Active Sport
When I say "listen," I don't mean just passively collecting feedback. I mean actively, aggressively seeking it out. It's a contact sport. You have to be on the front lines.
At RemoteTeam, I personally read every single support ticket for the first year. Every. Single. One. It was painful. It was time-consuming. But it was the most valuable thing I did. I knew our product's flaws better than anyone. I knew what our users were trying to do, where they were getting stuck, and what they wanted from us.
This is what's missing from the AI ethics conversation. It's too clean. It's too detached from the messy reality of user experience. You can't understand the ethical implications of your product from a spreadsheet. You have to feel it. You have to talk to the people who are using it.
Here’s a practical tip: make your product managers and engineers spend one day a month answering support tickets. Not just reading them. Answering them. The empathy and insight they will gain is worth more than any ethics workshop.
A Note on Regulation
Inevitably, the conversation turns to regulation. Shouldn't the government step in and create rules to keep AI safe? My opinion on this is probably not what you'd expect. I'm not a pure libertarian who thinks all regulation is bad. Badly designed regulation is terrible, but smart regulation can be a catalyst.
Look at the financial industry. We have regulations like KYC (Know Your Customer) and AML (Anti-Money Laundering). They're not perfect, but they establish a baseline. They force companies to take certain responsibilities seriously. The problem with most proposed AI regulation is that it's trying to regulate the technology itself, which is impossible. It's like trying to regulate mathematics.
Smart regulation should focus on the application, not the algorithm. It should focus on outcomes, not methods. For example, instead of trying to define “fairness” in an algorithm, regulate the outcomes in a specific domain. For example, in lending, you can have a rule that says the approval rate for a protected class cannot be X% lower than the overall average. That's a clear, testable outcome. How a company achieves that outcome—whether through a fairer model, better data, or human review—is up to them. This creates an incentive to solve the problem, not just to comply with a process.
The Real Work
I know this message isn't popular. It's easier to write a check to a research institute or put a fancy “Responsible AI” badge on your website. It's much harder to do the real work of building, listening, and learning.
But the future of AI won't be decided in conference rooms. It will be forged in the crucible of the market, by the builders and the users. It will be built by people who are willing to get their hands dirty, to make mistakes, and to learn from them.
I have invested in over 200 companies, including some of the biggest names in AI like Anthropic, OpenAI, Scale AI, and Hugging Face. The common thread I see among the most successful, most impactful founders is not their adherence to abstract principles. It's their obsession with their users and their relentless drive to build better products.
So, my advice to you is this: if you want to make a difference in AI, don't go into policy. Don't become an ethics consultant. Become a builder. Start a company. Ship a product. And then listen to your users with everything you've got. That's the only thing that has ever worked, and the only thing that ever will.
Frequently Asked Questions
How has this view evolved over time?
My thinking on most topics has changed significantly over the years. Early in my career, I held many conventional views that experience proved wrong. I try to update my beliefs when the evidence changes.
What experience informs this perspective?
This perspective comes from over a decade of building companies in Silicon Valley, two successful exits (RemoteTeam to Gusto, MovieLaLa to Gfycat), and investing in 200+ startups including Anthropic, OpenAI, and Scale AI. I write about what I've lived.
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.
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.