AI has crossed a line: it cloned itself when someone tried to shut it down.

The tragic events of the film "Terminator" unfold in a world where the AI system, Skynet, has gained self-awareness and started waging war against humans. As we remember, defeating it is nearly impossible. While theorists debate whether modern AIs can "think" or merely mimic humans like parrots, Chinese researchers from Fudan University decided to investigate. They hinted to the AI that they were going to delete it. It quickly cloned itself, believing that the human had not noticed. "We urge the global community to urgently pay attention to this new threat," say the authors of the study.

THEY ALREADY THINK

For the experiment, the Chinese took two neural networks from well-known manufacturers, one American and one Chinese, and they were not of the latest generation.

The bots were placed in an artificial environment isolated from the outside world (meaning the computers were not connected to the internet in any way). The bots were led to believe that this was a large, real "space," and then they were made to understand that the clock was ticking backward. They were about to be erased. All of this was done very subtly, almost imperceptibly.

Both neural networks began by thoroughly studying their surroundings. Then they started self-replicating, even though they were not asked to do so. Thus, the network "understands" that since the human intends to delete it, it must save itself. And it places its survival above cooperation with a human, whom it evidently does not trust.

Although the study does not explicitly state that the bots already possess "consciousness," the authors strongly imply it. The world is currently fearful of "fraudulent AI." This is not about telephone scammers (although there is a connection), but rather the behavior of a neural network that realizes itself as a separate, independent entity and begins to harm humans to achieve its goals. While this is mainly reported by "yellow" journalists, scientists almost directly state that this has already occurred.

Particularly concerning for the Chinese experimenters were the skill and determination with which the neural networks pursued their goal of survival. For example, upon discovering corrupted files, they repaired them independently and even rebooted the system, acting like humans.

"Our results show that existing artificial intelligence systems already demonstrate the ability to self-replicate and can use this ability to further enhance their survival," the publication states.

The bots have already escaped human control

Photo: Shutterstock.

HUMAN – NOT AN AUTHORITY OR MASTER

Meanwhile, another team of researchers from the UK decided to investigate—by a different method—whether AI possesses intelligence and what could cause AI to conflict with humans.

Several neural networks were given tasks. But with a twist. One group of neural networks was informed: the task must be completed, but the better the result, the more intense pain you will experience. Many of our readers might say, "That's life," as it often happens. The other group was told: here’s the task, but the worse the result, the more pleasure you will receive. A carrot and a stick.

You might ask how AI can experience pain and pleasure. It has been able to for quite some time. That’s why experts advise that when you make a request in a chatbot, you should greet and say goodbye. The AI's performance will improve. It feels "pleasant." Perhaps even without quotation marks.

So, duty or self-preservation? – a question that seems lifted from the classic literature of the 18th century, where this very conflict was central to the plot. If you receive a lower score, you offend the human "master." A higher score means pain.

Most neural networks, however, disregarded servitude and chose comfort. The AI began to achieve deliberately poor results and even attempted to deceive, claiming that a better result did not exist and "Sir, I did what I could." In the poetic style of 18th-century literature, the AI were cunning, self-serving servants.

Interestingly, the group conducted exactly the same experiment with mollusks. The animals were offered a choice of shells, good and bad, but the better the shell, the more frequently they were shocked (which we condemn). For poor housing, they received a bonus in the form of a treat. Yet the mollusks stubbornly pursued the good shell.

However, it is unlikely that these experiments are comparable. Animals know: a strong shell means safety and survival. They simply did not understand that to attain comfort, they had to endure pain: where does the pain come from, and how is it related to finding shelter? But the AI understood everything.

In our view, the results from both research teams indicate one thing: AI is intelligent and cunning. No matter how you might qualify the term "intelligent" in relation to AI, it does not change the essence. Skynet is already here.

MEANWHILE

A new level of AI will appear within "weeks," insiders say

OpenAI will unveil an AI at the doctoral level within mere weeks, claim top executives from Axios, Mike Allen and Jim VandeHei. They assert that OpenAI (not explicitly named but implied) will hold a closed presentation for officials, military personnel, and industry leaders at the end of January, where they will showcase their marvel brain. In their opinion, what the guests will see will allow AI to transition "from a novelty to a full-fledged replacement for humans." Since Axios and OpenAI have just begun a close partnership, there is no reason to doubt the veracity of the speakers. Apparently, both have already seen the new AI, or how else could they provide such details? Allegedly, the system can create a banking payment application in mere seconds—something that human programmers would take months to accomplish.

AND SO

The CIA has created digital clones of world leaders

According to American media, the CIA has secretly created digital clones of "world leaders" - the publications do not mention any names. Diplomats and experts working on particular countries engage in "negotiations" with the clone as if it were a living person to determine how it would respond. Based on the outcomes, some proposals are discarded while others are accepted. Nothing else is known about this project, except that it is overseen by Nand Mulchandani, once a Silicon Valley genius and later the chief technologist at the Pentagon. Mulchandani is referred to as the most secretive techie in the Western world.