A core idea from Dangerous With AI
What is the
Amplification Thesis?
Artificial intelligence does not replace human ingenuity. It amplifies it. The catch is that it amplifies weak judgment, bias, and error just as efficiently.
The direct answer
AI makes the signal bigger. It does not decide whether the signal is good.
Think of AI as an amplifier connected to a guitar. A skilled musician becomes more powerful. A missed note becomes louder too. The amplifier has no taste, conscience, or responsibility. It scales whatever reaches it.
The promise and the danger are the same capability.
Give a thoughtful professional a fast, tireless AI collaborator and creativity, learning, analysis, and productivity can expand dramatically. Give the same system a rushed assumption or an unchecked claim and the mistake can spread with equal speed.
The old computing rule was “garbage in, garbage out.” In an AI-first world, the more accurate warning is “garbage in, garbage at scale.” The error does not become wiser. It becomes polished, persuasive, and easier to repeat.
What becomes more valuable when AI is everywhere?
Not access to the model. Access will become common. The scarce inputs are judgment, taste, lived experience, a clear point of view, and the discipline to verify what leaves your hands.
- Bring a real signal. Know what you think before asking the machine to multiply it.
- Pressure-test the idea. Ask the AI to find the strongest case against you.
- Control the blast radius. Add more human review as consequences become harder to reverse.
A practical framework
The Amplifier Check
-
01
Is the thing I am about to multiply actually good?
If the central idea is weak, more output will only make the weakness larger.
-
02
If it is wrong, how loud does it get before someone catches it?
The wider the reach and the higher the stakes, the stronger the human checkpoint should be.
What should professionals do differently?
Invest in the input, not only the tool. Use AI to draw out your thinking, challenge it, and scale it after it has earned the right to be scaled. The goal is not less capability. It is capability with judgment attached.
That is what being dangerous with AI means in the good sense: formidable with the tool, disciplined about the result, and accountable for both.