Can AI Regulators Keep Up When They Don't Understand the Tech?

Jan de Vries · 2026-09-22

Alright, let's talk about a problem that's been keeping me up at night. It's this quiet, growing gap between the people making the rules for artificial intelligence and the folks building the stuff. Vendan Ananda Kumararajah recently pointed his finger right at it: how on earth can regulators maintain any real authority when the systems they're supposed to oversee are getting more complex by the minute? It's like asking someone who's never driven a car to write the traffic laws for self-driving semis barreling down the highway at 70 miles per hour. They might get the basic idea—don't crash, be safe—but the specifics? The edge cases? The sheer technical depth? That's a whole different ball game. ### The Growing Knowledge Chasm Here's the core issue. Frontier AI—the cutting-edge stuff that can write code, generate hyper-realistic media, or make predictions we can't even fathom—is advancing faster than any regulatory framework can possibly adapt. Regulators are smart, dedicated people, but they're often coming from legal or policy backgrounds. They're not sitting in the labs training these models. They're not wrestling with the petabytes of data or the billions of parameters. Their knowledge is, by necessity, second-hand. They rely on white papers, expert testimonies, and corporate disclosures. But what happens when the creators of the technology have a vested interest in downplaying the risks or obfuscating the mechanics? Who's checking their work? It creates a dangerous dependency. The regulator's authority becomes contingent on the goodwill and transparency of the very industry it's meant to police. ### Why Independent Knowledge Matters Think about it this way. If you're going to regulate the financial sector, you need auditors who can actually read the books. If you're regulating pharmaceuticals, you need scientists who understand the chemistry. Authority without understanding isn't authority—it's just guesswork dressed up in a suit. For AI, this knowledge problem isn't academic. It has real-world consequences: - **Bias and Discrimination:** Without deep technical insight, regulators can't effectively audit for embedded societal biases that might lead to unfair loan denials or skewed hiring practices. - **Safety and Control:** How do you set standards for "safety" in a system whose decision-making process is often a "black box" even to its creators? - **Market Concentration:** If only a handful of massive companies (think budgets in the tens of billions of dollars) truly understand the tech, how do you prevent a monopoly on both innovation and governance? As one expert put it, "We're building the plane while flying it, but only a few people have been given the blueprints." ### Building a Bridge to the Future So, what's the fix? How do we close this gap before it becomes a canyon? It's not about turning every regulator into a PhD-level machine learning engineer. That's unrealistic. But it is about building new kinds of bridges. We need dedicated, well-funded regulatory bodies with the mandate and the budget to build in-house technical expertise. Think of them as the special forces of governance—small, agile teams of hybrid experts who understand both the law and the lattice of code. They need the power to conduct their own audits, to demand access to models and training data, and to run independent tests. Furthermore, collaboration is key, but it must be structured to avoid capture. Creating formal channels for ongoing dialogue with leading researchers and ethical developers—outside of corporate PR departments—can provide a vital lifeline of independent insight. The bottom line is this: we can't afford to outsource understanding. The stability of our economies, the fairness of our institutions, and the security of our societies might just depend on it. The era of polite suggestions and voluntary guidelines is over. If we want AI to serve humanity, we need watchdogs that can actually see what's happening in the kennel.