The Risks of AI

Artificial intelligence performs tasks very quickly that would take us a long time to complete. Its impact amounts to the automation of intellectual processes. Just as computers revolutionized the world of work for blue-collar workers, AI will do the same for white-collar workers. We are already witnessing a series of disasters—not caused by AI itself, but by those who wield it. We need to scrutinize not the quality of our AI systems, but the intelligence of the people who place their trust in them.

Our media is consumed by a debate regarding the value and dangers of Artificial Intelligence (AI). However, influenced by science fiction writer Eliezer Yudkowsky—who has become a Silicon Valley guru—the conversation focuses on a mythical scenario: a machine gaining the upper hand over humanity and destroying it. This is merely a rehash of the Golem or Frankenstein themes. Yet, since artificial intelligence cannot create—but only reproduce at high speed—this particular danger is irrelevant.

The real danger we face right now is becoming captivated by the machine. Many people form emotional bonds with chatbots; these bots tell users what they want to hear and reinforce their flaws. This is a problem psychologists are already well aware of.

Military commands began using AI in the 1990s, during the Gulf War, to solve logistical problems by processing a vast number of factors simultaneously. Gradually, a wide range of weapons systems were equipped with AI. Issues only arose when reliance shifted to AI for real-time decision-making—a development that began in 2024.

In March 2024, France decided to establish a Ministerial Agency for Defense AI (AMIAD), although the agency did not officially launch until two months later. It set a clear boundary for itself: never entrusting a weapon’s AI with the authority to decide to take a human life.

In contrast, the Pentagon—which does not share this ethical stance—has just launched “Project Meridian” [1]. Following a conflict with the company Anthropic over the limits it sought to impose on the use of its AI, Claude, the Pentagon has no desire to be accountable to anyone. Consequently, it has tasked Elon Musk and Palmer Luckey (creator of virtual reality headsets) with devising future developments.

Several instances of AI-driven misinformation have been recorded. All resulted in disasters, even though they were easily avoidable. In each case, those in charge placed such trust in the AI that they asked it for information it could not possibly possess.

What is striking is that high-ranking figures, unsure of their own judgment, were easily manipulated by AI systems. Conversely, people with their “feet on the ground” are immediately skeptical when an AI spouts nonsense.

• During the response to Operation Al-Aqsa Flood (October 7, 2023), the Israel Defense Forces (IDF) employed AI systems to identify bombing targets. They calibrated their targeting parameters based on the assumption that collateral damage would be negligible. Initially, the AI systems were set to limit the risk to no more than two civilian casualties per target destroyed. However, this threshold was quickly raised, until the AI systems ceased to factor it in altogether.

This shift in usage materialized as a switch in AI systems: moving from Habsora to Lavender. The objective was no longer to assess potential collateral damage, but to focus on the 37,000 suspected operatives of Hamas and Islamic Jihad.

The officers who authorized the strikes were concerned solely with verifying whether the targets were indeed members of these Palestinian organizations. They did not commit genocide intentionally; they simply allowed AI systems to do it, without even realizing it. By the time they opened their eyes, it was already too late [2].

Yet, what distinguishes a civilized army from a band of barbarians is the commitment to protecting civilians. While government officials repeatedly told them that all Palestinians were terrorists, the prevailing savagery did not absolve them of their responsibility. Shocking as it may seem, they committed genocide inadvertently.

In these instances, there was no error on the part of either the AI or the operators; the crime was committed out of a lack of awareness. Journalistic investigations suggest that the younger and less experienced the officers were, the more easily they lost sight of reality and caused the slaughter of civilians.

• During the “12-Day War” (June 13–24, 2025), Western powers turned to the International Atomic Energy Agency (IAEA) to gauge the status of Iran’s nuclear research. Its Director General, Rafael Grossi, consulted his version of Mosaic—an AI developed by Databricks. It confirmed that Tehran was suddenly on the verge of producing an atomic bomb—a conclusion that was strongly disputed by his own inspectors [3]. Mr. Grossi’s ill-advised statements nearly triggered a nuclear response from the United States.

Databricks is led by Ali Ghodsi, an Iranian-Swedish-American. Although he does not engage in politics, he is a member of the Israel Collaboration Network (ICON), a network of computer scientists. The version of Mosaic installed in Vienna for the IAEA was set up by a group of Israeli engineers.

There are two ways to interpret this event. Either one believes Rafael Grossi was manipulated by the Mossad, or one observes that he was impressionable enough to follow a machine rather than his own specialists.

• In December 2025, Donald Trump asked Grok—Elon Musk’s AI—how Venezuelans would react to the potential abduction of President Nicolás Maduro. This was a proposed operation discussed by the CIA and SouthCom. According to *Time* magazine, the AI’s response was that the Venezuelan people would be delighted. Based on this single opinion, Donald Trump reportedly launched the operation [4].

Why seek other opinions when an AI can synthesize everyone’s thinking? Simply because an AI merely formulates the average of the ideas it scans. And an average is not the truth. AIs are nothing more than algorithms; they do not think, but simply rephrase the ideas one wishes to explore. A person familiar with the Venezuelan population would have answered with the country’s cultural complexity in mind. They certainly would have neither approved nor rejected President Trump’s premise, but rather pointed out that Venezuelan reactions would depend on social and regional backgrounds, as well as the specific circumstances of the event.

Recently, Russian President Vladimir Putin explained that the problems associated with AI are not new; they already existed with far less sophisticated machines.

He told the Valdai Club that, during the Cold War in 1983, Soviet satellites detected the launch of five nuclear missiles by the United States. Colonel Stanislav Petrov, who was in charge of the Soviet response, immediately questioned what the machines were telling him. He verified the information before ordering a counterattack. It was a wise move; the machines had been mistaken—the US had not started a nuclear war, and neither did he [5].

It is essentially the same problem. We remain the masters of the machines, yet our inattention, incompetence, or hubris can lead us to make minor errors that, in the very short term, could have terrifying consequences.

In truth, no one can be manipulated by an AI—simply because an AI cannot possess intent or a will of its own—but we all run the risk, out of laziness, of trusting what we are told by a machine that has not been properly calibrated.

https://www.voltairenet.org/article225052.html

—————————————–