Donald Trump is ushering in an AI takeover. It has Americans fearing a Terminator-like dystopia

President Donald Trump's dismissal of AI safety concerns as a "hoax" has drawn criticism as polls show about seven in 10 Americans oppose data centers near their neighborhoods and 80 percent back a federal agency to monitor AI, according to University of Maryland's Program for Public Consultation. Columbia University's Visha Misra, vice dean of computing and AI, told The Independent that fears of a Terminator-style apocalypse are "overblown" and that the focus should be on harms already occurring. Last week, a bill by Louisiana Republican Senator John Kennedy requiring a "kill-switch" for advanced AI was blocked by GOP colleague Rand Paul, who said acting hastily "risks killing innovation and setting this technology back decades.

Donald Trump is ushering in an AI takeover. It has Americans fearing a Terminator-like dystopia Experts say artificial intelligence isn’t about to unleash killer robots but is already causing serious harm - Bookmark President Donald Trump's outspoken support for unbridled advances in artificial intelligence https://www.independent.co.uk/news/world/americas/us-politics/trump-ai-data-centers-maga-truth-social-b3051360.html is sparking fears of a Terminator -style apocalypse but experts say that scenario is still science fiction — and that other AI harms are already here. Trump set off a wave of criticism https://www.independent.co.uk/news/world/americas/us-politics/trump-ai-poll-anthropic-b3050138.html and calls for government regulation when he dubbed concerns over AI's potential risks https://www.independent.co.uk/news/world/americas/us-politics/trump-ai-rant-conspiracy-guardrails-b3049920.html a “hoax” and wrote in a social media post that the industry didn't need any guardrails other than having a “STRONG AND SMART High IQ President” like him in the White House. “Trump is a blind fool with his statement about AI and saying that people are saying things ‘that won’t happen,’” one outraged commentor wrote on X. “If a human being can imagine it, it can happen and it will be done eventually. Terminator is no longer sci-fi, it’s a documentary series.” While industry insiders have warned that the rapid pace of AI development could put humanity at risk of extinction within the next 10 years, academics who study the technology told The Independent that its current state, and controls now in place, would prevent that from happening. “Some of this is overblown, and I would say that the focus has been on the wrong thing,” said Visha Misra, vice dean of computing and AI at Columbia University's School of Engineering and Applied Science. The controversy over AI's potential threat to human life follows widespread community opposition to the data centers needed to store the information and run the software that performs the tasks it is assigned. Although Trump is a proponent of building data centers across the U.S. — saying last month that opponents would “end up being backwards and poor” — polls have consistently shown that about seven in 10 Americans are against allowing the sprawling facilities near their neighborhoods. Recent surveys by the University of Maryland's Program for Public Consultation https://publicconsultation.org/united-states/cg2026-ai/ found even greater support for creating a federal agency to monitor new AI programs, make policy recommendations and enforce regulations. Nationally, 80 percent of those polled favored the plan, including 85 percent of Democrats and 79 percent of Republicans. The idea of regulating AI programs that make critical decisions won similar levels of support, with 82 percent of Democrats, 78 percent of Republicans and 78 percent overall favoring government controls. Despite voters’ fears, Congress has been slow to act. Members on both sides of the aisle have expressed concerns but concrete action appears far off. Last week, legislation proposed by Louisiana Republican Senator John Kennedy, which would have required a “kill-switch” for advanced AI to prevent machines from unleashing disaster on humanity, was ultimately spiked by his GOP colleague Rand Paul. “If Congress acts hastily before the technology is understood, Congress risks killing innovation and setting this technology back decades,” Paul said. Misra said there were two parts to envisioning an AI nightmare: the “self-improving part, and the other, that it's going to kill us all.” But the notion that AI models “can keep on improving infinitely” and will at some point match or exceed human cognitive abilities — known as artificial general intelligence or the “singularity” — isn't happening, he said. “The thing there is, you can propose something, and you can also verify it all inside the computer,” Misra said. “But if you have to deal with something which lives outside the digital bit world, you know, in the atom world or in the physical world, then you cannot just self-improve.” “You have to go out in the real world, run experiments and get an answer from nature,” he added. And when it comes to the part where AI would kill everyone, Misra said, “there's something similar in play.” “These models have to go out in the real world and cause physical damage,” he said. “So they need access, they need permission, and, you know, they need authority. And you just don't hand over these things to models.” Misra said that as AI models become more capable, “the real danger is: what if it supercharges the next Unabomber?” “So, we have to safeguard real-world dangers and not worry about AI self-improving itself and deciding to kill us all,” he said. “There are other real dangers in the world, and the focus should be there, not this fear-mongering that's going on right now.” Misra also blamed this year's infamous hack of the Hugging Face website by rogue OpenAI agents on a “failure in the cybersecurity environment of OpenAI.” “There were obvious holes that the models were able to exploit. They were asked to exploit it, and they did it,” he said. “So, you know, AI cannot go out and kill us all unless it gets the power to do it.” Northwestern University computer science professor Kristian Hammond offered a similar analysis and illustrated it with a story he heard from an AI industry insider about a system built to “go and buy things” for its users. “Of course, the developers forgot to give it an upper bound in terms of what it was allowed to spend, and it was spending way more than it should,” he said. “And you look at that, and it sounds cute, but that means that it's doing things you don't want it to do. And if you build a piece of software that's doing things that you don't want it to do, that is a bad piece of software.” Hammond —who in 2010 co-founded an AI company that turns data into narratives, which Salesforce bought in 2021 — also pointed to Anthropic's revelation last year of nefarious AI activity during testing of its Claude Opus 4 chatbot. After being given emails that suggested the software would soon be shut down and that the engineer in charge of it was having an extramarital affair, Claude Opus 4 attempted to blackmail the engineer to preserve itself. “No one in the world, in any field, has ever said, ‘What we want is...to produce a product that will quasi-randomly do a bad thing and hurt people,’” he said. “But that's what we're kind of building right now.” Hammond said, however, that when it came to AI systems “going rogue” and killing everybody, “the odds of that happening are intensely slim, because we're not that stupid in terms of software design.” “I think things like this will pop up and will get in the way, and we will...sort of do the right thing in terms of figuring out how to frame them,” he said. Hammond illustrated his assertion with a reference to the Terminator movies, which is premised on a backstory in which a U.S. military defense system called Skynet gains self-awareness and starts a nuclear war with Russia to avoid being shut down. “The way you avoid Terminator -style, Skynet-kind of problems is we don't give them control over things that — if they went rogue — would hurt you,” he said. Misra, the Columbia professor, likewise said that AI's “current architecture” can't produce killer robots like in The Terminator . "If an architectural breakthrough happens — next month or next year or five years — who knows? But the way these models have been designed, they are reactive," he said. “They react to prompts. They react to tasks that are given to them. They don't sit around thinking. They don't have thoughts of their own. They're not plotting to eliminate humanity.” Daniel Schiff, an assistant professor of technology policy at Purdue University and co-director of its Governance & Responsible AI Lab, known as GRAIL https://www.grail-lab.org/ , said that industry concerns about out-of-control AI date back more than a decade and that "we all should be taking this very seriously." Schiff described himself as “a bit of a pluralist” who's worried about a “wide range” of potential threats from AI, adding that “it doesn't have to be a nuke.” “We're already being so affected by these systems that we need to...have already been more ahead of our regulation and our societal conversation,” he said. Schiff said he's “typically been most worried about what you would call sort of middle-scale risks,” citing online misinformation and disinformation that leads to a “loss of trust in media and institutions.” “This is something that really messes with our health as societies, and this can be, sort of, supercharged by algorithms and even now deepfakes,” he said. “And some of them are already happening.” Schiff said other examples include AI-generated child sexual abuse material and non-consensual intimate imagery, as well as “AI that's led to unfair decisions in people's receipt of public services and welfare.” “We have, in some ways, made employment more difficult from a hiring, and vetting and applying for jobs perspective. We have, at minimum, concerns from students and workers about what will their job even be? Will they have value in the labor market? We have infringements on intellectual property and the creative sorts,” he added. Hammond cited “suicides that come as a result of interacting with a system https://www.independent.co.uk/news/chatgpt-openai-adam-meta-congress-b2827813.html whose only job in the world is to make you feel good about the decisions you've made, even though they might be bad for you.” Hammond also said he was concerned about the “lazy expertise” provided by AI chatbots and the resulting erosion of critical thinking among students and other young people. “It's this odd thing where higher education is like, ‘We're terrified right now, because how do you teach someone something where they have a box on their desk that has expertise in the area that you're teaching them?’” he said. That dilemma, Hammond said, is fueling his research into building “a system that has all the smarts in the background, but it never answers questions.” “And in fact, you train it how to teach. You train it how to educate and build systems that, by their nature, make people smarter instead of, by their nature, make people stupider. And the core technology is there,” he said. Join our commenting forum Join thought-provoking conversations, follow other Independent readers and see their replies Comments comments-area