The Nuclear Reactor of Bad Ideas

Every important technology changes how ideas live or die. Which ones will be given room to grow, and which ones will be given a merciful demise.

The printing press multiplied ideas. Radio multiplied voices. Television multiplied images. The internet multiplied access.

And now AI is currently multiplying something even more fundamental: the speed at which an idea, any idea whether it’s good or bad or somewhere in between, can become “sophisticated.”

That sounds great…until you remember a simple truth.

Amplification doesn’t care what it amplifies. It never has.

We talk all day long about AI as though its greatest gift is intelligence. But what if that wasn’t necessarily true?

After all, intelligence implies that there’s some level of judgment involved. It assumes we always know the difference between a sound idea and a flawed one, between an idea that deserves further exploration and one that should be abandoned before it takes on a life of its own.

AI is more of an amplifier than a judge.

And amplifiers couldn’t care less. They have one job, and judging isn’t it.

Let’s suppose I start with an idea that contains a fatal flaw. And just to clarify, it’s not an obviously ridiculous idea. Those are easy to dismiss. I’m talking about the kind of idea that feels persuasive because it’s almost right.

It could be a business strategy built on a false assumption about customers, or a psychological theory that mistakes correlation for causation, or a worldview that quietly flatters my existing beliefs. The kinds of ideas that have always been dangerous because they don’t look dangerous at first.

Just a few years ago, an idea like that had to earn its survival.

I would wrestle with it for weeks. I’d test it, tweak it, put it through the motions. I’d get another set of eyes on it to discover gaps in my reasoning. I’d encounter objections I couldn’t answer. Reality would push back. Other people would push back. Even the effort of developing the idea would expose weaknesses that weren’t visible at the beginning.

I can’t tell you how many times that’s happened to me before AI came along. Indeed, I trashed infinitely more wild and crazy ideas than I care to admit here.

That kind of cognitive gauntlet was never the enemy.

It was part of the filtering system.

Many bad ideas didn’t fail because someone proved them wrong. They failed because they couldn’t survive enough honest thinking. They collapsed under the weight of their own contradictions before they ever reached a wider audience.

AI turns that equation upside down.

Now I can begin with the same flawed premise and, within an hour, surround it with extraordinary intellectual scaffolding. I can generate supporting arguments, historical parallels, compelling analogies, anticipated objections, persuasive rebuttals, visual frameworks, case studies, and even a year’s worth of content explaining why the idea deserves attention.

The idea itself may not have improved by a single millimeter.

Yet everything built around it has become richer, more coherent, and more convincing.

That distinction feels profoundly important, and frankly I’m amazed that no one is talking about it.

We humans have a tendency to mistake elaboration for validation. We assume that if an idea can support an intricate framework, it must rest on a solid foundation. If someone can explain something elegantly for three hours, answer every objection, and produce a hundred pages of supporting material, we instinctively grant the idea more credibility.

But elegance is not evidence.

Complexity is not truth.

Coherence is not correctness.

An idea can become beautifully articulated while being fundamentally flawed.

That realization led me to a metaphor I haven’t been able to shake.

AI reminds me less of an encyclopedia than of a nuclear reactor.

A reactor does one thing extraordinarily well. It releases enormous amounts of energy. It doesn’t ask whether that energy will power a hospital, a research laboratory, or a weapon. Those decisions exist outside the reactor itself.

Its power lies in amplification, not judgment.

The same may be true here.

Feed AI a genuinely insightful idea, and it can help that insight spread farther and faster than ever before. Feed it an original scientific hypothesis, and it can help explore implications the researcher might have missed. Feed it a creative breakthrough, and it can uncover fascinating new directions.

But the opposite is also true.

Feed it resentment, and resentment becomes articulate.

Feed it vanity, and vanity becomes systematic.

Feed it an untested assumption, and the assumption can emerge looking like rigorous thought.

The reactor hasn’t malfunctioned. On the contrary, it has done exactly what it was designed to do.

The uncomfortable question is whether we have confused the ability to expand an idea with the wisdom to decide whether it deserves expansion in the first place.

For centuries, intellectual progress depended upon resistance.

A scientist proposed a theory, and other scientists attempted to dismantle it.

A writer advanced an argument, and critics searched for weaknesses.

An entrepreneur built a company, and the marketplace exposed hidden assumptions.

Resistance was expensive, frustrating, and often humbling.

It was also invaluable.

Resistance forced ideas to earn momentum.

AI dramatically reduces the cost of development. That is one of its greatest strengths. But in reducing the cost of development, it also reduces one of the natural opportunities for weak ideas to reveal themselves before they become persuasive.

Perhaps that’s why I’ve become less interested in asking AI better questions and more interested in asking myself better ones.

Not, “Can this idea become a book?” or “Can this become a course?” or “Can I build a business around this?”

Those questions come later.

The first question should be much simpler.

Is the premise actually sound?

Have I challenged it? Tested it? Tried to disprove it?

Have I looked for evidence that would make me abandon it altogether?

Or have I just recruited the most powerful intellectual amplifier in history to help me become more confident in something I haven’t examined deeply enough?

That possibility should make all of us a little uneasy, because it quietly changes where our responsibility begins.

It’s also incredibly risky. A bad idea that’s amplified can result in lost time that could’ve been spent pursuing better ideas, lost revenue on expanded duds that never should’ve seen daylight, and even lost confidence in our own abilities if we invest all our resources into something that didn’t have any right to exist in the first place.

The machine can elaborate almost anything, but only a human can decide whether the thing being elaborated deserves another hour of thought.

I suspect this is becoming one of the defining intellectual responsibilities of our time, and I’m not sure I have any brilliant solutions, other than to strongly suggest you ask AI to poke holes in every idea you put forth. Prompt it to give you the “hard truth” and to not worry about hurting your feelings.

Or you could Pretzel it, which puts ideas, premises, content, sales copy and countless other material through the ringer using a proprietary 7-step process. I’m having a Pretzeling Masterclass next week, it would be great to see you there. Sign up here.

At the end of the day, the future will not be shaped only by those who generate the most ideas. It will also be shaped by those who know which ideas should never enter the reactor.

Maybe discernment has always been the scarce resource.

AI didn’t create that scarcity.

It simply made its absence far more consequential.