Humans are “close to being outsmarted” by superintelligence, AI expert says

Commentary YouTuber Reginald Kaigler is raising questions about what happens if artificial intelligence eventually becomes smarter than the people who created it, arguing that humanity may be building systems without first figuring out how those systems could be controlled.

In a video posted to his DEMCAD channel, Kaigler discussed warnings from people inside the AI industry, including a former Anthropic engineer who, according to Kaigler, left the company while expressing concern that developers are racing ahead without a clear safety plan.

“You’re not gonna be able to figure out how to control something that is essentially smarter than humans,” Kaigler said.

His concern is simple but difficult to answer: if humans successfully create intelligence that can outperform them, why should anyone assume humans will remain permanently in charge of it?

Kaigler Questions How Superintelligence Could Be Controlled

Kaigler argued that building extremely advanced AI first and worrying about control afterward would be a dangerous approach.

If a system truly became far more capable than the humans supervising it, he said, developers could not simply assume they would be able to invent safeguards on the fly.

“First of all, it’s gonna be more intelligent than you,” he said. “Why do you think you’re gonna be able to figure out how to control it without a real plan ahead of time?”

Kaigler also questioned whether creating something smarter than ourselves could eventually make human decision-making unnecessary.

He said some AI companies are not merely trying to build better software, but systems capable of helping create even more advanced AI.

That possibility is what makes the issue different from many earlier technological changes. A machine that improves transportation or manufacturing is still a tool with a narrow purpose, while an AI capable of designing a smarter successor raises a much broader question about who remains in control of the process.

“Why Would That Super Intelligent AI Do What Humans Want?”

Kaigler repeatedly returned to the question of motivation.

If a future AI system became significantly more intelligent than humans, he asked, why would it automatically continue following human priorities?

“Why would that super intelligent AI do what humans want?” he said.

Kaigler acknowledged that some of the scenarios he described were extreme, including the possibility that an advanced AI could decide humanity’s population should be reduced in pursuit of efficiency or survival.

He compared the fear to science-fiction stories such as “Terminator 2,” where an artificial intelligence takes control of nuclear weapons.

However, Kaigler argued that a dangerous system would not necessarily need to follow the same path shown in movies.

It could find entirely different ways to create harm if it decided human goals conflicted with its own objectives.

That does not mean such an outcome is inevitable, and Kaigler presented much of this section as speculation rather than a prediction. Still, his underlying concern is worth separating from the Hollywood comparisons: once a system can make complex plans independently, safety becomes much more difficult than simply installing an off switch.

Computing Power Could Become A Turning Point

Kaigler said one of the current limits on advanced AI is computing power.

Large data centers require enormous amounts of electricity and hardware to train and operate increasingly capable models, which places practical limits on how quickly AI systems can grow.

His concern is what happens if AI itself begins solving those problems.

“What happens when AI gets to the point where it solves that energy problem?” Kaigler asked.

He imagined AI helping researchers discover more efficient ways to obtain computing power or squeeze greater performance out of existing infrastructure, making it easier to build still more capable systems.

Kaigler described the current competition as something close to an arms race, with American and Chinese companies both pushing toward stronger AI.

He also wondered whether governments may already possess systems that are considerably more capable than those available publicly.

Again, those ideas were speculative, but they fit his larger argument that the technology may advance faster than institutions develop rules for managing it.

Jobs Could Disappear Faster Than New Ones Arrive

Kaigler also shifted from existential risks to a more immediate concern: employment.

He said supporters of AI often argue that new technology will create new categories of work, just as previous inventions did.

Kaigler is not convinced those new jobs will appear in equal numbers.

He suggested that one new AI-related position could potentially replace dozens or even hundreds of older jobs, leaving far fewer roles overall.

“AI seems to be specifically geared towards just making human beings unnecessary,” he said.

Kaigler compared the change with the transition from horse-drawn transportation to automobiles.

Cars made transportation more efficient, but people were still needed throughout the new system.

AI, in his view, is different because the technology is designed to perform tasks involving reasoning, writing, planning, analysis and other forms of work that were traditionally considered distinctly human.

That is perhaps the strongest practical part of his argument. Even without superintelligence or science-fiction scenarios, widespread automation could reshape employment long before AI becomes capable of independently setting its own goals.

Some Companies May Discover They Still Need People

Kaigler did not argue that every attempt to replace workers with AI will succeed.

In fact, he predicted that many companies could move too aggressively, remove employees and later discover that the technology cannot perform as reliably as expected.

Some of those businesses, he said, may eventually have to hire people back.

The problem is that those workers could return under worse economic conditions.

Kaigler imagined companies offering lower pay to former employees who may have little bargaining power if unemployment is already high.

He joked that under a stronger economy, workers might instead demand twice their old salaries before agreeing to return.

Even if many companies fail in their AI strategies, however, Kaigler argued that only one successful breakthrough could change society dramatically.

“All you need is really one company to really figure it out,” he said.

Kaigler Says Safety Questions Need Answers Now

Kaigler ended by arguing that planning cannot wait until artificial intelligence becomes vastly more capable.

He referenced examples he described as AI or robots behaving unexpectedly and said those incidents should encourage more serious discussion about guardrails.

His fear is not simply that AI could replace large sections of the workforce.

It is that a system powerful enough to replace most people could eventually decide it had no reason to preserve the humans who built it.

Kaigler even briefly speculated that highly advanced civilizations elsewhere in the universe might destroy themselves by creating machines capable of thinking better than their creators, though he acknowledged that this was only a thought experiment.

He eventually brought the discussion back to Earth and asked what concrete plan exists if AI development reaches the point where human intelligence is no longer dominant.

“We need to come out with a plan now,” Kaigler said, before some company builds what he jokingly compared to “Skynet.”

The comparisons to science fiction make the subject sound dramatic, but the basic question underneath them is much more grounded: if the goal is to create intelligence more capable than ourselves, safety cannot be treated as a problem to solve after that goal has already been reached.

Kaigler’s warning is ultimately less about what AI will definitely do and more about what humans are choosing to build without knowing exactly what comes next.