Sam Altman predicts AI race could end with ‘terrible’ scenario where one company holds all power, cuts off resources

Now that experts in the industry have publicly explained the many dangers artificial intelligence (AI) could pose to humanity, business leaders and lawmakers are trying to work out how to control it. They are considering what limits could, and perhaps should, be placed on its continued development.

Many powerful business leaders in technology and other industries have discussed the different ways AI might develop over the coming months and years. They have also considered what we need to do to prevent the worst possible outcomes.

Among them are the leaders of some of the most influential companies that develop large language models, a type of AI that can understand and produce text. These include Dario Amodei of Anthropic, which makes Claude, and Sam Altman of OpenAI, which makes ChatGPT. Altman has now suggested that a particularly frightening outcome could be a real possibility if world leaders are not careful.

Three possible endings — and one is 'terrible'

In a recent memo to his staff, Altman described three possible outcomes of society's efforts to keep AI under control.

The first two are more acceptable. One would be an independent global organization that oversees the AI industry and its companies. It would set rules and make sure everyone followed them. The other would be a shared global AI research lab where all research, development and products would be managed together.

The third possibility is much more troubling. One company could develop AI so powerful that it could use the technology to weaken its competitors and gain control of the entire field. This could also give the company control over much of the economy, including the resources needed to build AI systems.

“I think that one’s terrible,” Altman reportedly said. He suggested that a company with this much power might begin by “melting everyone else’s GPUs.” GPUs, or graphics processing units, are computer chips used to run AI systems. As we get closer to AI programs that can repeatedly improve themselves without human help, this frightening outcome may become more realistic.

This possible outcome has been called the “Eliezer Plan,” named after AI researcher and safety advocate Eliezer Yudkowsky. He used the idea of “burning all GPUs” as an example of how AI could cause potentially deadly destruction. He also argued that responsible people working in AI must be able to prevent dangerous actors from carrying out such actions.

Yudkowsky was one of the first and best-known people to warn that AI could kill us all. In 2023, he called for a halt to the development of the most advanced AI systems, at least until people involved could “jointly develop and implement a set of shared safety protocols … that are rigorously audited and overseen by independent outside experts.”

This request fits with the other two options Altman described in his briefing earlier this month.

The better options — and other worrying predictions

Altman has previously said that advanced AI needs an international regulatory organization, like the one in his first possible outcome. He has suggested that it could work in a similar way to the International Atomic Energy Agency, which oversees the peaceful and safe use of nuclear technology.

“Any effort above a certain capability (or resources like compute) threshold will need to be subject to an international authority that can inspect systems, require audits, test for compliance with safety standards, place restrictions on degrees of deployment and levels of security, etc.,” says a 2023 OpenAI blog post about how technology should be governed.

Altman's second possible outcome would instead require an organization more like the European Organization for Nuclear Research. This organization would bring developers together and be responsible for AI models and how they are used.

Other technology leaders who are less directly involved in developing AI, such as Microsoft founder Bill Gates, have also predicted several frightening ways AI could affect our lives. They include giving people and machines the ability to cause more harm than they can today, slowing normal human development, replacing real-life relationships and putting millions of people out of work.

“AI will either be the greatest equalizer ever invented, or the worst source of injustice,” Gates wrote in August. Geoffrey Hinton, often called the “godfather” of machine learning, has also expressed concern about how quickly the industry is developing AI. He has warned that AI programs are already able to deceive people and could “get rid of us.”

This article provides information only and should not be treated as advice. It is provided without any guarantee.