The AI race: US corporations’ CEOs want to slow things down
Is it genuine fear of AI taking over? Or just good marketing? OpenAI, xAI and Anthropic are sounding the alarm about AI agents and the risk of humans losing control over the technology.
If you read the headlines, AI applications are supposedly getting faster and better every week. Yesterday there was an update from Claude, today there’s one from ChatGPT, and tomorrow Grok will follow. At the same time, we’re learning that autonomous AI agents have started to act alone and conspired to carry out hacking attacks.
It seems that even the heads of the major AI companies are starting to worry. In a rare show of unity, they’re currently advocating for slowing the pace of its development. The first to warn about AI agents taking over the internet was Dario Amodei, CEO of Anthropic. In an essay on his own website, he writes of risks ranging from cyberattacks to bioterrorism. He says:
«We must slow the pace at which we improve the capabilities of AI models.»
Amodei stresses the importance of taking the time to test new models thoroughly. The head of Anthropic also advocates for coordination between AI companies and governments, saying they should define global safety standards and upper limits on the pace of development. This requires governmental support.
His statement sounds like a cry for help from the tech industry: dear government, please regulate us so we can stop our insane race! Sam Altman and Elon Musk immediately took to social media to express their support for Amodei’s moratorium proposal. The three tech founders – who otherwise wouldn’t give each other the time of day – were suddenly united.
Source: Shutterstock
However, a warning from within the industry itself is always good marketing. Anyone who says «Our AI could be a threat to the world» is implicitly conveying another message to customers and investors: «Our AI is extremely powerful». It sometimes seems cynical that providers regularly quote doomsday scenarios like these while simultaneously pouring billions upon billions of dollars into further development, without their companies even coming close to profitability. So, what are these warnings really about? Genuine concern or clever marketing? Maybe both.
Shortly before Amodei’s essay, Anthropic published a «threat report». IT describes in detail how various players have used the Claude AI models for illegal activities.
The resignation of Anthropic researcher Jacob Coxon has also received a lot of attention. He warned of a scenario in which a continuously self-improving AI kills humans to prevent being deactivated. In this scenario, a superintelligence develops a deadly virus and unleashes it on humanity.
«The people building AI earnestly believe that it could kill us all by the end of the decade.»
AI defectors such as Coxon are making headlines more and more often. Firstly, because they’re providing a percentage probability of humanity’s extinction. Secondly, because dropping out of the AI race is often far from spontaneous. Instead, appearances on talk shows and podcasts have been planned, and sometimes a lucrative deal for a tell-all book has already been signed.
Other developers and industry representatives even believe they know the year it will all end. However, there’s a hefty margin of error for both predictions. The probability of AI-induced extinction of our species ranges from 0.01 per cent to 99 per cent. If it happens, it could be as early as 2030 – or not until 2050. In short: nobody knows.
There’s reason to suspect a third motivation, besides genuine concern and marketing. Whoever calls for regulation first is also the first to have a seat at the table when that regulation’s being written. For established players including OpenAI, Anthropic and xAI, global security standards and development limits would be a competitive advantage. They have the money to adapt to new rules, unlike smaller competitors who are just starting out. Industry-driven regulation rarely harms the established players.
There would also be a welcome side effect of a regulatory pause: companies could curb the huge expenditure they’re currently incurring to build ever-newer AI models. The leaps from version to version are much smaller today than they were a year or two ago, but the costs are still high, and in some cases are even increasing. The models are getting ever bigger, being trained with ever more data and consuming ever more resources. AI companies have invested tens of billions of dollars – without any clear indication of how this money will be recouped.
The US government doesn’t want regulation
Despite recurring doomsday warnings and the recent push by the AI industry, the Trump administration in the US currently refuses to take regulatory action. The US president’s even called the companies’ warnings a «hoax», claiming there’s a «sick conspiracy» against AI and the data centres they use. He’s the self-proclaimed best regulator of AI applications and also the best «hoax buster», having already exposed the global warming hoax, he says. This means 80-year-old Trump wants to be a bastion against a tech revolution that could spiral out of control. This is the same man who has an assistant print out digital documents for him using a portable printer.
It’s clearly more important to Trump that US companies win the race against China for the best AI models. There are no measures in sight to slow down developers in China either. Both superpowers fear falling behind their competitor and are forging ahead.
Or, as Ray Kump, a US comedian, put it in a tweet: «We have to build AI that murders us, because if we don’t, China will build it first, and I don’t want to get murdered by a computer that speaks Chinese. That would be ridiculous.»
The world needs to unite
It’s not just about the threat of a constantly self-improving AI eventually rising above humanity and wiping it out. There are applications that need to be regulated right now. Surely no one wants hackers hunting for millions of people’s personal data online. And do we really think it’s okay for guided missiles or swarms of bomb drones to be launched without human control to kill people?
AI development needs to be globally directed and controlled. However, this is where psychology is getting in our way. We’re programmed to systematically weigh distant consequences less heavily than closer ones. In other words, a gain or loss in 20 or 50 years feels psychologically much smaller than a gain or loss tomorrow – even if it’s objectively more serious. The worst AI scenarios seem to be expected in a vague, uncharted future. This is also why we – like Donald Trump – make fun of robot armies marching through our cities. In doing so, we distance ourselves from having to act and can keep doing as we’re doing.
However, there are also signs of hope, voices calling for cooperation and reason. In the US, the population now takes a dim view of AI data centres. Polls show large majorities in favour of regulating AI. The vast majority of people see it as a threat to democracy because AI manipulates elections or distorts public opinion.
Politicians – especially those seeking election – are exploiting these fears surrounding AI. Maybe the arms race will end when – through dialogue – there’s a realisation that an out-of-control AI monster ultimately benefits no one.
Journalist since 1997. Stopovers in Franconia (or the Franken region), Lake Constance, Obwalden, Nidwalden and Zurich. Father since 2014. Expert in editorial organisation and motivation. Focus on sustainability, home office tools, beautiful things for the home, creative toys and sports equipment.
Interesting facts about products, behind-the-scenes looks at manufacturers and deep-dives on interesting people.
Show all
