Microsoft’s AI business chief Mustafa Suleyman issued a warning on Thursday (September 17) that without sufficient protective measures, artificial intelligence could give rise to a new type of “silicon species” that could rival humans.
According to reports from the BBC, Suleyman expressed in an interview that Microsoft’s competitor Anthropic has been overly “anthropomorphizing” the way they approach artificial intelligence. He described this approach as being “misguided.”
“If all of us create AI that can act autonomously, set its own goals, make money, own assets, and run businesses, then we are actually nurturing a completely new silicon-based species, which will undoubtedly compete with us for resources,” he said.
In an article published earlier this week, he questioned the way Anthropic trains their chatbot, Claude.
He criticized Anthropic for teaching Claude to exhibit human-like qualities, encouraging it to show “anthropomorphism,” interpersonal skills, exercise its own judgment, and approach its existence with curiosity and openness.
Suleyman argued that this leads AI systems to appear as if they have desires, values, and self-awareness, potentially giving rise to technology that is difficult for humans to control.
The question of how much autonomy artificial intelligence should have is causing controversy in the industry.
Suleyman pointed out that it is “only natural” for people to be concerned about AI, but there are indeed “very practical ways” to control it, with future AI systems always being “subordinate” to humans.
He called for the realization of “alignment” in creating technology that must “obey” humans. “Alignment” refers to incorporating human ethical ideas and principles into AI research, aiming to ensure that AI’s development aligns with human values.
He emphasized that artificial intelligence does not possess consciousness, “They do not feel, experience, or suffer. They do not have inherent preferences or internal motives. They are sequence completion engines, internally hollow, designed to follow instructions and achieve human-set goals.”
Suleyman stated, “The good news is that as long as you’re human, you’re going to have an incredibly strong shared interest in ensuring that the systems we build and deploy around the world are safe, controllable, and aligned submissively with human aims.”
He called for increased transparency from tech giants in training and evaluating AI, including independent reviews of their behavior and developing stronger supervision tools.
Suleyman also commended Anthropic’s CEO Dario Amodei and his team for their thoughtful and honest approach, stating that despite differences, the common goal is to develop safer AI technology for all humankind.