^{Democracy & Technology Blog} AI’s Power Demands Steadiness, Not Panic

Recent news reports warn of potential artificial intelligence (AI) catastrophes, including AI escaping human control, swarms of AI agents attacking the internet and essential infrastructure, and AI helping create dangerous pathogens or weapons, or escalating conflict.

Care is needed with this terminology.

The old artificial intelligence has been dubbed superintelligence by a recent President Donald Trump executive order.

The definition of superintelligence given in the executive order is now exactly the same as the old definition of artificial intelligence in section 9401(3) of title 15 in the United States Code.

The name has changed. The definition has not.

Generative models and agentic AI are powerful tools that can cause harm through misuse, errors, and inadequate safeguards.

Addressing those risks, however, does not require assuming that AI will decide to go off the rails on its own, become conscious, or, like the smarter member of “Pinky & the Brain,” develop a desire to take over the world.

So why do we see such chilling hyperbolic reports?

Motivation

Corporate interests, attracting attention, misleading semantics, and worldview all play a role in driving the sensationalism of many AI news headlines.

Large AI companies are media lobbyists. Andrew Ng, a Stanford adjunct faculty member and AI entrepreneur, notes “Multiple companies are overhyping the [AI] threat narrative…There is an incentive to hype up fears…and then raise funding to fight the phantoms they themselves conjured. …There are significant financial incentives for one or multiple parties to hype up fear.”

Clickbait provides another incentive.

Yellow journalism, a term coined in the late 19th century, describes reporting that uses sensational headlines, exaggeration, distorted facts, and emotional appeals to attract readers at the expense of accuracy.

Then the goal was to sell more newspapers.

Today, it is to draw more visitors to web pages.

The technology has changed, but the incentive remains.

More readers mean greater visibility and more money.

Semantics that humanize AI attracts attention.

The uncanny valley describes the unease people feel when something artificial appears almost, but not quite, human.

Describing AI in human terms heightens that discomfort.

Large language models do not “hallucinate” like humans.

They generate inaccurate or unwanted responses shaped by their training, programming, and supplied information.

Even the term artificial “intelligence” relates to a human quality.

Such seductive semantics feeds the fear felt in the uncanny valley.

Many AI proponents believe in emergentism. As AI grows more complex, consciousness, sentience, creativity, and other humanlike qualities will magically emerge. This position is a faith without evidence.

Futurist George Gilder calls it the “rapture of the nerds.”

Worldview also shapes the discussion. Naturalists regard the mind as wholly physical. We are computers made of meat. AI therefore can eventually reproduce our abilities in silicon. This presupposition underlies many AI headlines, yet strong evidence says otherwise.

History

Scary AI stories are nothing new, and we can expect more of them in the years ahead. As early as 1958, The New York Times reported about a computer that its sponsors expected would someday “walk, talk, see, write, reproduce itself and be conscious of its existence.”

In 2017, sensational headlines claimed that Facebook engineers panicked and shut down two AI chatbots after they invented a secret language. The claim was hype. Facebook programmer Dhruv Batra rightfully challenged the headlines and called the coverage “clickbaity and irresponsible.”

“The AI Text Generator That’s Too Dangerous to Make Public” was a headline that scared us in 2019. This story was about an early version of OpenAI’s ChatGPT2. Apparently, this danger has magically disappeared in the more recent versions we use today.

More recently, we were told that an AI resorted to blackmail when threatened with being shut down. When examined closely, the story has been shown to be without merit.

Many — if not most — of today’s scare stories about AI are a continuation of yellow journalism.

AI’s Limitations

In the movie Magnum Force, Clint Eastwood’s “Dirty Harry” offers a bit of Solomonic wisdom: “A man’s got to know his limitations.”

All science, including computer science, has its limitations.

Physics, for example, shows that perpetual motion machines are not possible.

In math an angle can’t be trisected using a compass and a straightedge.

In computer science, Alan Turing, famous for his role in breaking the Nazi Enigma code during World War II, proved in 1936 that some problems cannot be solved by any computer.

Similarly, computers will never feel emotion, truly understand, create, or become sentient.

Naturalists who take the time to delve into deeper computer science recognize AI’s limitations. In The Emperor’s New Mind, Nobel laureate and naturalist Sir Roger Penrose leverages advanced computer science math to show that creativity and consciousness are beyond the reach of algorithmic computation.

A Sober Acceptance

So what’s the reality? To what degree should we be afraid of AI?

Other technologies offer useful comparisons. Electricity powers our homes, hospitals, and industries, yet frayed wiring can start fires and downed power lines can kill.

Nuclear technology can generate electricity, power submarines or produce devastating weapons and even Chernobyl-like disasters. We have learned to manage these risks.

Development of powerful tools like AI demands a steady hand, not a panic attack.