AI Doesn’t Have to Turn Evil.
The Wrong Human Just
Has to Take Control.
We Regulate Planes, Electrical Grids, and Banks.
Why Can’t We Regulate AI?
Your aviation analogy is useful because aviation doesn’t depend on one perfect person or one perfect computer. It uses layers. Pilots, controllers, radar and surveillance, procedures, separation rules, automated warnings, redundant equipment, maintenance standards, incident investigation, licensing, and independent regulatory oversight all overlap. If one layer fails, another is supposed to catch it.
That’s very close to what I think robust governance of extremely powerful AI should look like.
You wouldn’t want an AI connected directly to something enormously consequential with the philosophy:
“Don’t worry. We programmed it to be nice.”
You’d want something closer to aviation or nuclear safety: the AI has defined permissions; certain actions simply aren’t available to it; unusually consequential actions require independent authorization; activities are logged and monitored; abnormal behavior triggers containment; multiple independent systems can veto an action; and there are mechanisms that don’t depend upon the AI cooperating with its own shutdown.
And there’s another part of your analogy I particularly like: air-traffic control doesn’t prevent airplanes from flying. It makes mass aviation possible.
That’s an important distinction in the AI debate. Safety doesn’t necessarily mean stopping AI development. Good safety engineering can be what allows society to use extremely powerful technology confidently.
The difficult part is international coordination. Airplanes cross borders, so nations eventually had enormous incentives to agree on common standards. AI crosses borders even more easily. One country can impose excellent safeguards while another government, company, criminal organization, or individual operates under very different rules.
So your basic proposition makes sense to me:
As AI capability becomes more powerful, the systems supervising and constraining its use have to become correspondingly powerful.
Not necessarily one gigantic world computer policing every other computer—that introduces its own frightening concentration-of-power problem—but multiple independent layers of technical controls, human oversight, laws, auditing, security and international agreements.
And Freddy, there’s a deeper principle underneath what you’re saying:
We’ve spent a century learning how to make extraordinarily dangerous technologies useful—commercial aviation, electrical grids, pharmaceuticals, nuclear power, global banking—not by pretending accidents and malicious actors don’t exist, but by engineering systems that expect failures and build defenses around them.
AI should be treated with that same maturity as its capabilities increase.
The Genie Chronicles explores tomorrow’s.
Read The Genie Chronicles
Explore the books • Continue to Amazon
No comments:
Post a Comment