The artificial intelligence industry acknowledges a 5 to 10% chance that humans will lose control of AI, according to Stuart Russell, a computer science professor and head of U.C. Berkeley’s Center for Human Compatible Artificial Intelligence.

My question is: with algorithms already creating societal chaos and machines verging on self-consciousness, has humanity already lost control?

Algorithms drive outrage

AI algorithms designed to maximize engagement have changed social media from a useful way for humans to connect with each other to what is too often a cesspool of misinformation and outrage. The algorithm “learns” that exciting misinformation and outrageous assertions make people want to see more, which the engagement-maximizing algorithm happily obliges.

With algorithms driving so much outrage, misinformation and violent content, public discourse based on facts is lost in a sea of ignorance and superstition. Thus, an anonymous user on 4Chan, going by the letter “Q,” could claim to have secret government information that Q would periodically reveal in “Q-drops.” These Q-drops warned that cannibalistic witches who worshipped Satan had infiltrated the U.S. Government.

Advertise with us

Algorithms kept this witches’ brew boiling by spreading this exciting but ridiculous news from gullible user to gullible user. Perhaps as many as 4.5 million people became followers of Q on Facebook, which recently stopped its fact-checking.

Q has an antecedent. In his recent book, “Nexus,” Yuval Noah Harari discusses how the book, “The Hammer of the Witches,” which explained how to identify and punish witches, helped keep the travesty of European and early American witch-hunting going. First published in the late 15th century, it went through 29 editions by 1670 in numerous translations. It was highly profitable for sellers just like the “maximize engagement” algorithms are highly profitable for social media corporations.

These AI algorithms undermine reasonable debate. Agreement on basic facts becomes impossible. A democratic republic cannot survive this chaos. A dictator, promising order, may then bring the ruin of public liberty with the help of an AI-enabled super-KGB.

The maximize engagement algorithms should be outlawed. They are “The Hammer of the Witches” on steroids.

Self-conscious Machines

Russell states that we do not know how ChatGTP works: it is not designed, but grows. In their recent book, “Genesis,” the late Henry Kissinger, Craig Mundie, and Eric Schmidt wrote that before long, machines will actually be able to define their own objectives and develop a rudimentary self-consciousness. They say that machines will come to their own conclusions about themselves, humans, history, and the universe.

Advertise with us

Russell states that AI cannot yet write a code to make a better version of itself, but once it can, it will keep rewriting that code to keep improving itself. The authors of “Genesis” argue that to avoid creating malevolent Artificial General Intelligence, AI development must be aligned with human values. They contend that no AGI is preferable to AGI that is not aligned to human values.

But humans do not agree on those values. These authors wrote, “how to achieve consensus on what those human values are … is the philosophical, diplomatic, and legal task of the century.” The great ancient Greek philosophers, the great philosophers of the 18th, 19th and 20th centuries, the great diplomats of history, and history’s greatest legal minds have all failed so far.

History teaches that there cannot be a world-wide consensus. Nations and cultures that hate the values of others will develop their own AGI. An AI arms race looms. But if this leads to a self-conscious AGI that determines its own objectives, who knows what it may try to do? Can or will it seek to harm humans?

Only if AGI created a universally recognized threat, such as machines versus all humans, might humanity come to a universal agreement to combat AGI, but by then it may be too late.

Possibly the last mistake

Given the profit motive and the AI arms race, stopping the development of AGI is unrealistic, but control must be attempted. Fundamentally, AI is mechanized human language. Like any other machine, if poorly designed, it can go out of control. It can also be designed to do harm and apparently to achieve a kind of self-consciousness.

Advertise with us

Russell suggests that proof must be provided that an AI product is safe before it can be marketed. This would include proof that the AI will not self-replicate without human permission, will not break into other computers, will not defame people, will observe personal rights, will not instruct terrorists in how to build biological weapons, and more. The European AI Act is a comprehensive and salutary effort along these lines.

AGI could be humanity’s greatest and last mistake. It’s time to apply the brakes.

Daniel O. Jamison, a retired attorney in California and frequent contributor to The Fulcrum, is writing a book on the Second American Revolution. His previous article was “George Washington warned us about despots who don't like democracy.”