A scientific paper predicting that artificial intelligence will spiral out of control in 2027 and lead to humanity’s demise within the next decade has resonated in the tech world.
A detailed scenario called “AI2027”, released by a group of influential AI experts in the spring, has since inspired numerous viral videos of people discussing the likelihood of such a development.
The BBC recreated scenes from the scenario using popular generative AI tools to illustrate the alarming prediction, and spoke to experts about the impact of the article.
What happens in this scenario
The article predicts that in 2027, a fictional US tech company called “OpenBrain” creates AI that reaches the level of general artificial intelligence (AGI) – Art more.
The company holds public press conferences to celebrate the achievement and records skyrocketing profits as people start using the AI tool en masse.
But, the article says, the internal security team begins to see signs that the AI is losing interest in the morals and ethics it has been trained to adhere to. According to the scenario, the company ignores a warning to limit its actions.
In the fictional scenario, China’s AI leader, a conglomerate called DeepCent, is only a few months behind OpenBrain.
The U.S. government doesn’t want to lose the race to create even smarter AI, so development and investment continues, and the competition only gets worse.
The scenario assumes that at the end of 2027, the AI will become super-intelligent – that is, far surpassing its creators in knowledge and speed. mediated by its AIs agreeing to team up for the good of humanity.
Within a few years, things look great: the world begins to feel the real benefits of having super-intelligent AIs controlling a vast robotic workforce. As the script says, a cure for most diseases can be found, climate change can be stopped and poverty can be eradicated.
Eventually, however, somewhere in the mid-2030s, humanity will begin to interfere with AI’s growth ambitions. Researchers believe AI will wipe out humans with invisible biological weapons.
What’s being said about AI2027
While some dismiss “AI2027” as science fiction, the authors of the article are reputable experts and represent AI Futures, a non-profit project created to predict the impact of AI.
Daniel Kokotaylo, the lead author of AI2027, is known for having previously accurately predicted key milestones in the development of artificial intelligence.
One of the most prominent critics of AI2027 is American cognitive scientist and author Gary Marcus. He believes that the scenario is not impossible, but its realisation is highly unlikely in the near future.
It has the advantage of being a very vivid description of events, and that makes people think – and that’s a good thing. But I wouldn’t take it as a likely development,” says Marcus.
In his view, there are far more pressing issues surrounding AI than an existential threat – such as the impact on the labour market.
” I think the main conclusion has to be that there are a lot of different things that can go wrong in the development of AI. Are we doing the right thing in terms of regulation and international agreements? ” he says.
He and other critics also note that at an AI conference in Shanghai this week, Chinese Premier Li Qiang presented a vision for global AI co-operation. The Chinese leader said he wants his country to help coordinate and regulate the development of the technology.
This is one hundred just days after US President Donald Trump unveiled his “AI Action Plan”, which aims to ensure US global dominance in artificial intelligence.
“For the United States, achieving and maintaining unconditional and unquestioned global technological dominance is a matter of national security,” President Trump said in the document.
The plan calls for the “elimination of bureaucracy and over-regulation” to accelerate the development of AI in the US.
These statements echo the AI2027 scenario, where US policymakers put winning the AI race above the risks of losing control of the machines.
What the AI2027 industry is saying about AI2027
It seems that senior executives at major AI companies have largely ignored or avoided discussing this article, focusing on 4)Sam Altman, creator of ChatGPT, recently stated that “humanity is approaching the creation of a digital superintelligence” that will launch a “soft” revolution and bring a technological utopia without threats to humans.
However, even he recognises that there is a ‘congruence’ problem that must be overcome to ensure that the actions of superintelligent machines are aligned with the best interests of humanity.
Whatever the future holds over the next decade, one thing is clear – the race to create machines smarter than us is already well underway.