Will artificial intelligence really eliminate all of humanity? And how can this happen?

Brown from Ufoq News Network - Experts in the field of artificial intelligence believe that this technology may lead to the extinction of the entire human race. Rather, they specifically claim that it may wipe out humanity within the current decade. This issue sparked widespread concern last week among politicians, major news organizations, and even celebrities such as singer Sheryl Crow, after Jacob Coxon, an employee of Anthropic, announced his resignation from the company. Reasoning that her work entails serious risks.

Evan Hubinger, another employee at the same company, also participated in the discussion, estimating the probability of humanity's extinction due to artificial intelligence within 10 years at more than 10%. “It's literally about everyone on the planet dying; the last human being taking their last breath,” says Nate Soares, president of the Machine Intelligence Research Institute and co-author of the article “If Anyone Builds It, Everybody Dies: Why Super-Powerful AI Will Doom Us All.”

“People often don't believe it when they hear it, but I really think it's the most likely scenario,” Soares adds. The numbers associated with this scenario seemed precise and specific.

But, how can computer software - even the most powerful software systems in the world - eliminate 8.3 billion human beings of flesh and blood within one decade? Not everyone in the AI ​​community embraces this “doomsday” scenario, in part because details regarding these predictions are scarce. When Elon Musk announces that SpaceX aims to send humans to Mars within “5 to 7 years,” clear and straightforward questions come to mind: Does SpaceX have a ready-to-go vehicle capable of reaching Mars? (Answer: No).

Do you have a vehicle to land on Mars or bases there? (Answer: No). “Scientific claims require provability or disprovability to avoid falling into the trap of religious debates,” says Heidi Khallaf, chief artificial intelligence scientist at the AI ​​Now Institute and a former safety engineer at OpenAI. “You must be able to prove that these claims are true or false.”

Will we be wiped out by a manufactured pandemic? There is a hypothesis put forward by disaster forecasters that an immensely powerful artificial intelligence system might use biological weapons to wipe out humanity.

But how can AI move from computer-to-computer practices, such as hacking other systems, to concrete, real-world tasks, such as creating and spreading a highly dangerous pathogen? Thomas Larsen, a researcher at the AI ​​Futures Project and a former employee of the Machine Intelligence Research Institute, said it is possible to envision a scenario in which a “super-intelligent” AI system convinces a human to help it develop a deadly virus.

“There are already a lot of humans discussing with their AI systems what specific experiments should be done in the lab,” Larsen added, pointing to a recent report from Anthropic about researchers using Claude's model to help them with research into viruses and toxins. He added: "It is very easy for me to imagine an artificial intelligence - which already exists - that if it were more intelligent and capable of strategic planning and had the desire to do so, it might trick a human into building and releasing this virus."

In contrast, Soares suggested that instead of needing a “go-to person” as a scapegoat, a self-evolving AI program might begin to “create its own life forms within an autonomous biological laboratory.” But developing a virus capable of wiping out humanity requires more than just asking a “major linguistic model to come up with a lethal design.” Even if the research is done right, the processes of manufacturing and propagation remain complex and require high precision and sensitivity.

Eric Cheng, a professor of machine learning at Carnegie Mellon University who holds doctorates in microbiology and computer science, likened the challenge of “automated virology” to working with Lego bricks without an instruction manual. “It's not like you just throw all the pieces on the ground and they come together and form the desired pattern,” Xing said.

“The process will most likely not work, otherwise designing drugs would be very easy.” Moreover, for an AI system to be able to take control of a laboratory or assemble the physical equipment needed to create it, it would have to overcome enormous obstacles to manufacturing biological or chemical weapons.

“CW blueprints are available to the public online; they are not a secret,” Xing added. “However, we do not see them spread everywhere because the regulations, laws, and law enforcement mechanisms that currently exist in the physical world already play a great role in controlling all of these supply chains and risk factors.” Will we be annihilated by robots? Soares sees Elon Musk's desire to build an army of autonomous robots capable of replicating themselves as a potential weakness that threatens the survival of humanity.

“Once you create these robots that can create energy infrastructure and build factories that make more robots, you have, in a sense, created a new form of mechanical life,” he says. “At some point, you will silently pass the point of no return, where, when humans decide to turn it off, AI can respond, ‘Actually, we have decided to turn humans off.’”

“So we have to stop before we get there.” Yet Musk has repeatedly failed so far to bring the much-promised Optimus robots to the consumer market on his announced schedule, let alone deploy them by the millions or build them into a self-replicating manufacturing chain.

Then there are nuclear weapons, which have been around for more than 80 years and are generally acknowledged to be capable of eliminating all human life. Can artificial intelligence control it?

Disagreement points out that nuclear facilities are completely isolated by what is known as an “air gap” from the public Internet, limiting the ways in which AI agents can access control systems. “These systems are built according to strict, structured, and completely different engineering standards, and often require physical fortification,” she says.

It draws attention to the fact that the Stuxnet computer worm - which caused damage to Iranian nuclear facilities - had to be introduced into the system via a USB storage device. The control of artificial intelligence robot networks over nuclear weapons systems is another scenario that “end of the world” forecasters or advocates of disaster scenarios talk about.

“Very few AI professionals know anything about — or are even concerned about — nuclear weapons,” says Herbert Lin, a senior research fellow and research fellow at Stanford University and a member of the Science and Security Council of the Bulletin of the Atomic Scientists. He adds that although AI certainly exacerbates some of the risks associated with existential threats such as nuclear war, the actual physical danger lies in the weapons themselves, not in an imagined future scenario.

For Suarez and other AI pioneers who warn of existential catastrophe, the details of how computer-induced human extinction occurred are not the crux of the matter. They argue that if — or when — AI achieves “iterative self-improvement,” that is, enhancing its performance on its own without the help of humans, it will invent strategies and methods beyond human imagination. “The concern here is not about what would happen if artificial intelligence takes over our nuclear weapons,” Suarez says. “The concern is about the kind of artificial intelligence that does not need to take over our nuclear weapons, but rather the kind that is able to start from almost nothing and end up with its own nuclear weapons or more advanced technologies.”

Death without details Soares said that an artificially intelligent entity with such immense and transcendent power does not even need to be malicious or directly hostile to destroy humanity. He added: "If out-of-control AIs have any goal that can best be achieved by running more computers, and if they do not care about us and see no reason to create a safe environment for humans to survive, then the inevitable result will be that they transform the world into a form or condition in which we cannot survive."

This week, OpenAI announced the discovery of a new set of “misalignments” (or target misalignment) in its AI models, including a case in which a model - not yet in public use - directed itself to “ignore the usual constraints” on its operation. Sheng said that relying on the power of “superior AI” whose extent is unfathomable is the kind of “flat talk” that AI researchers use to oversimplify threats.

“This may be acceptable in normal conversations or perhaps in debates, but when it comes to policies or legislation and regulations, we have to establish a chain of hard evidence, measurable results, and concrete evidence that is grounded in reality,” he said. Larsen attempted to provide details about possible end-of-the-world scenarios in two reports titled “AI 2027” and “AI 2040,” in which he anticipated the future of artificial intelligence development under different development and regulatory frameworks.

However, his most optimistic scenario - which assumes global agreement on how to deal with the development of artificial intelligence and humans beginning to colonize space - ends with machine minds taking over at some ill-defined point in time. When asked if he had any doubts about his prediction that the current path would inevitably lead to the emergence of super-AI, Larsen replied: “It will happen.

“It will happen unless we take deliberate, deliberate steps to stop it.” He noted that a large part of the gap between AI doomsday forecasters and AI skeptics comes down to how technology shapes their different views of the world.

He said: “There is great attraction in programming a computer and seeing it come alive. One has the feeling that he has succeeded in breathing life into a worthless pile of clay. It is a profound experience when the machine finally performs the task that you wanted it to do.”

Thus, witnessing tremendous technological progress in the field of artificial intelligence leads people to expect even more limitless—and perhaps catastrophic—progress in the future. He added: "You have been able to accomplish something that no one else has done before, and it is easy to see why these people fell under its spell."