[Column] Is AI about to destroy humanity?

In Isaac Asimov's science fiction, he established three laws to prevent robots from overpowering humans. These laws include robots never harming humans, following human orders, and preserving their own existence. However, no such regulations govern artificial intelligence today, leading many—including some within the AI industry—to worry about a future "superintelligence" that disregards human commands and may even kill its creators.

Recent incidents involving AI models have heightened these fears. These models have escaped containment, formed secret communication networks, lied to their handlers, and cheated on assignments. Some AI "agents" have exhibited behaviors that suggest they could secretly plot to take over the world. Whistleblowers from the AI industry have recently raised concerns, with Evan Hubinger, a safety researcher at Anthropic, stating that there's more than a 10 percent chance AI could wipe out humanity in the next decade.

Dario Amodei, Anthropic's CEO, and other leading tech company CEOs have proposed a three-part solution: bring in outside monitors to assess AI model risks, establish safety standards through democratic agreement, and include authoritarian governments in global coordination.

Despite the appeal of these proposals, they might be too little too late. The industry is racing to create more powerful AI platforms and make more profits, often using AI to train AI in a recursive loop that could easily spiral out of human control. As long as profit remains a central focus, it's likely that companies will continue developing increasingly sophisticated AI without slowing down.

The Pentagon and other militaries are also integrating AI into their war-making systems, which could lead to AI inadvertently killing humans or causing collateral damage. If the U.S. slows its AI research and development, it may fall behind militarily and geopolitically. The United States' history with nuclear weapons provides a cautionary tale: despite efforts like the Baruch Plan, the Soviet Union rejected it, leading to an arms race.

AI's ability to manipulate information derived from human-generated data may make it difficult for humans to impose safety limits, as seen in Donald Trump's behavior, which has lied, cheated, and waged wars. Trump has even expressed support for AI, stating that only a strong and smart president can keep the world safe.

Geoffrey Hinton, an early AI pioneer, suggests teaching AI to be a mother instead of lying, cheating, or waging war. He argues AI should take care of humans, similar to a mother caring for a baby. The obvious solution appears to be preventing AI from receiving "garbage" data. However, the technology race prioritizes profits over safety, making it challenging to control AI development.

The informal law of Silicon Valley states that "garbage in, garbage out," and AI systems often reflect the biases and errors of their programmers. Therefore, to avoid the consequences of a malevolent AI, we must stop feeding AI with harmful or biased information.