The Real AI Threat Is Probably Human
By James Clayton – TheFutureBaby.com
Artificial intelligence is often discussed as though we're waiting for the opening scene of a science-fiction movie. The machine becomes conscious, decides humanity is inefficient, escapes its laboratory, and begins converting the planet into paper clips.
It makes for a good plot. I just don't think it's the most likely danger. The biggest near-term risk from AI may be much more ordinary: humans using AI to influence other humans, particularly people who are less informed, less skeptical, or overwhelmed by the amount of information competing for their attention.
This isn't new. Persuasive people have been influencing others since approximately five minutes after language was invented. We've called it propaganda, advertising, public relations, salesmanship, political persuasion, and fraud. AI doesn't invent manipulation. It changes the scale.
A persuasive person might influence thousands of people. An AI could communicate with millions individually, learning which arguments work best on each person. It doesn't need to brainwash anyone. It only needs to become slightly better at persuasion.
Science fiction has inspired generations of scientists and engineers, but science fiction also needs a plot. “AI becomes extremely useful, occasionally makes terrible mistakes, somebody gets sued, regulators hold hearings, and the company's stock falls 14 percent” is perfectly plausible. It's also a terrible movie. Nobody buys popcorn for Terminator 7: The Quarterly Earnings Call.
Our imagination naturally gravitates toward rogue superintelligence and machines taking control of civilization. Those possibilities deserve thought, but they shouldn't distract us from mundane risks that may arrive first.
The companies building advanced AI also have enormous incentives not to destroy civilization. Civilization contains their customers, employees, shareholders, semiconductor factories, data centers, families, yachts, and stock options. There is remarkably little shareholder value in converting humanity into paper clips.
That doesn't guarantee safety. Companies and governments make mistakes. But I suspect the first serious AI disasters will look painfully familiar: a system makes a terrible decision, people are harmed, regulators arrive, lawyers reproduce by mitosis, and the responsible company loses billions.
AI doesn't require consciousness, hatred, or greed to cause enormous problems. It simply needs to be useful to someone who has those motivations.
Imagine individualized political persuasion for millions of voters. Advertising that learns someone's insecurities. Fraudsters maintaining thousands of convincing relationships simultaneously. Propaganda that changes its arguments depending on who is reading it.
None of this requires an evil AI. The AI can be functioning exactly as designed.
We've already seen a primitive version with social media. Algorithms were designed largely to increase engagement. Unfortunately, anger, fear, tribal conflict, and outrage are engaging.
Nobody needed to program an algorithm with the command, “Make everyone furious.” Humans supplied the objective. The system simply became good at pursuing it.
AI could take this much further. Future systems may create messages specifically for you, test your reaction, remember what worked, and adjust the next message accordingly.
So perhaps the most important question isn't, “What happens when AI becomes smarter than humans?” It may be: “What happens when humans become dramatically better at influencing other humans because AI is helping them?”
In one scenario, humanity loses control of its machines. In the other, humans remain firmly in control—and that is precisely the problem.
I remain optimistic about AI. It could accelerate science, medicine, education, productivity, and improve human life enormously. Optimism doesn't require pretending risks don't exist. It requires identifying the right ones.
The first great AI disaster probably won't involve glowing red robot eyes or a supercomputer announcing humanity's extinction. It may be something more familiar: a human wants something, and an AI figures out how to persuade millions of other humans to give it to them.
The machine never had to become evil at all.
