In a recent exchange, Microsoft AI chief Mustafa Suleyman and OpenAI CEO Sam Altman have expressed starkly different views on the timeline and vision for achieving artificial general intelligence (AGI). This debate highlights the evolving strategies within the tech industry regarding the future of AI.
Key Takeaways
Mustafa Suleyman estimates AGI could take up to 10 years to develop, depending on hardware advancements.
Sam Altman believes AGI is achievable with current technology and may arrive sooner than expected.
The distinction between AGI and the singularity is a focal point of Suleyman's argument.
Tensions between Microsoft and OpenAI are evident as both companies pursue their own AI models.
Diverging Timelines for AGI
During a recent Reddit AMA, Sam Altman claimed that AGI could be realised using today’s hardware. In contrast, Mustafa Suleyman expressed scepticism, suggesting that the current capabilities of hardware, such as Nvidia’s GB200 GPUs, are insufficient for AGI development. Suleyman stated that while AGI is plausible, it could take up to 10 years to achieve, depending on advancements in hardware generations.
He elaborated:
Each hardware generation takes approximately 18 to 24 months to develop.
Achieving AGI may require two to five more generations, translating to a timeline of five to ten years.
Suleyman cautioned against making categorical predictions, labelling them as “ungrounded” and “over the top” due to the high uncertainty surrounding AGI.
Understanding AGI Versus Singularity
A significant aspect of Suleyman's argument is the distinction between AGI and the concept of singularity. He defines AGI as a system capable of performing well across various human-level training environments, including both knowledge work and physical labour. In contrast, the singularity is characterised as a self-improving system that rapidly surpasses human intelligence.
Suleyman stated:
AGI does not necessarily lead to singularity.
His focus is on creating AI systems that are practical and beneficial to humans, rather than pursuing theoretical superintelligence.
He acknowledged the challenges in robotics and other areas but remains optimistic that AI systems could perform much of human knowledge work within the next decade, even if AGI or singularity is not achieved.
Altman's Adjusted Perspective
At The New York Times DealBook Summit, Altman revised his stance on AGI, suggesting it could arrive “sooner than most people think” but might have less impact than previously anticipated. He noted:
AGI can be developed, and the world will continue to function similarly.
The transition from AGI to superintelligence will take significantly longer, with safety concerns primarily arising at the superintelligence stage.
Tensions Between Microsoft and OpenAI
The differing perspectives of Suleyman and Altman reflect broader tensions between Microsoft and OpenAI, particularly as Microsoft develops its advanced AI model to compete with OpenAI’s GPT-4. Suleyman acknowledged these tensions, describing them as a natural part of any partnership:
“Every partnership has tension. It’s healthy and natural.”
He emphasised that partnerships evolve and adapt over time, indicating that the relationship between Microsoft and OpenAI will continue to change in the coming years.
As both leaders navigate their distinct paths toward AGI, the ongoing debate underscores the complexities and challenges of shaping the future of artificial intelligence.
Sources
Microsoft’s AI boss and Sam Altman disagree on what it takes to get to AGI - The Verge, The Verge.
Microsoft AI chief and Sam Altman differ on AGI timeline and vision - Tech Edition, Tech Edt.
