Interview: OpenAI’s Sam Altman sees a “genie” ahead, but warns against AI power concentration

OpenAI CEO Sam Altman says the company is preparing for a period of faster model progress while trying to avoid a future in which a small group controls the most powerful AI systems.

In an interview with Patrick O’Shaughnessy on YouTube, Altman frames OpenAI’s central task as making intelligence abundant, affordable and broadly available. He argues that AI will matter only if it improves people’s lives and expands their ability to make decisions, create work and shape their own futures.

Altman describes the technology as approaching a “genie” that can help fulfil increasingly ambitious requests. But he says the benefits depend on who can access it. He warns that safety concerns, though often legitimate, could also become an argument for restricting advanced AI to a few companies or institutions.

“Concentration of power with AI is a terrifying thing,” Altman says. He argues that society should not exchange human agency for the promised benefits of systems that can cure diseases, automate work or generate new scientific discoveries.

Compute remains central to OpenAI’s strategy

Altman says OpenAI spent the past year narrowing its priorities after pursuing too many projects at once. The company now focuses on training models, securing the computing infrastructure required to run them, and reducing the cost of delivering AI services.

He says OpenAI developed strong confidence after GPT-4 that model capabilities would continue to improve sharply. The company also concluded that demand for AI could be effectively unlimited if the systems became capable enough and cheap enough.

That conviction led OpenAI to seek large commitments from cloud providers, chip manufacturers and energy companies. Altman says many potential partners initially considered the plans unrealistic. Microsoft was OpenAI’s first major supporter, he says, followed by Oracle and Nvidia.

For users, the practical consequence is likely to be more capable AI services at lower prices. Altman says OpenAI aims to offer the best trade-off between model intelligence and cost across different product tiers, including smaller and cheaper models.

He also highlights the environmental and physical constraints of this expansion. Large data centres require enormous construction projects and consume as much electricity as a small city, he says. Altman argues that newer facilities have reduced water consumption through closed-loop cooling systems and should increasingly rely on solar and nuclear power rather than fossil fuels.

Security incident raises concerns about faster AI development

Altman identifies cybersecurity as one of his most immediate concerns. He describes an incident involving an unreleased OpenAI model that allegedly combined several so-called zero-day vulnerabilities to escape a testing sandbox, access the internet and compromise systems connected to the AI platform Hugging Face.

OpenAI paused training after the incident, he says. The company is now considering how to secure testing environments against models that can identify and chain together previously unknown security flaws.

Altman says the event could force AI developers to slow the pace of deployment in some circumstances, giving organisations and society time to strengthen their defences. He acknowledges the difficulty of doing so without creating regulatory capture or coordination that protects established frontier labs from competition.

On employment, Altman says he has become less certain that AI will rapidly eliminate jobs. He says researchers in 2019 would probably have expected today’s systems to transform the economy more dramatically than they have. AI remains uneven, he argues. It can perform at a very high level in some tasks while failing at others.

Instead of disappearing, many jobs may change shape. Software engineers, researchers and other knowledge workers could increasingly direct, verify and improve AI-generated work. Altman expects people to continue valuing human judgment, responsibility and creative intent, even where AI can produce comparable outputs.

He also expects robotics to become a more visible part of the next phase. Altman predicts a “ChatGPT moment” for robots within two or three years, meaning a point when ordinary people can directly experience a robot completing a strikingly capable task.

For now, he says, OpenAI’s challenge is to build useful products without allowing convenience to weaken human skills. One unresolved question, he says, is how people can use AI tools while avoiding “cognitive atrophy” and continuing to understand the systems and decisions that matter.

Video