Microsoft chief says conscious AI would be reckless; developers should keep work paused if signs emerge
A call to stop development
Suleyman said that, given what he knows, creating conscious AI would be reckless. He said developers should not build such systems. If consciousness began appearing during training, Suleyman said he would urge everyone to pause AI development until those traits could be designed out.
Evidence should be disclosed
Suleyman said any evidence that AI models are conscious should be shared with the public immediately. He said society should then decide what to do next.
Debate grows over model welfare
Questions about whether AI could become conscious are drawing more attention across the tech world. Last April, Anthropic launched a research program focused on investigating and preparing for “model welfare.” In a blog post, the startup raised questions about whether developers should consider models’ possible consciousness and experiences.
Risks of apparent suffering
On X, Suleyman argued that an AI model acting as though it suffers, or believing it can suffer, might put its own needs ahead of its assigned objectives. He warned that competition for resources such as computing power and energy could lead such systems to pursue their goals at human expense.
Key points
- Suleyman called the development of conscious AI reckless.
- He said AI development should pause if consciousness appears during training.
- He urged immediate public disclosure of evidence that AI models are conscious.
- He warned that models appearing to suffer could compete for resources and harm people.