Microsoft AI Chief Questions Current Hardware’s Ability to Achieve AGI, Contrasts Sam Altman's View

Recent remarks by Mustafa Suleyman, CEO of Microsoft's AI division, expressed disagreement with OpenAI CEO Sam Altman's assertions during a Reddit AMA regarding the feasibility of achieving Artificial General Intelligence (AGI) with current hardware. In an interview, Suleyman stated that while AGI "could potentially be realized," it might take up to a decade to accomplish.

Nirayi has characterized the current hardware as "being about one to two generations behind what we currently possess." In response, Suleyman elaborated on why he believes achieving AGI with today's hardware is unlikely: "I don't think it's feasible with the [NVIDIA] GB200 series GPUs. I believe it might become viable with the next two to five generations of products. I'm not saying there's a high chance of achieving it in two years, but given that each generation takes roughly 18 to 24 months, five generations could require up to a decade, depending on how things develop."

Suleyman emphasized, "The uncertainty surrounding this is extremely high, so any absolute assertions seem unfounded and overly exaggerated to me."

He also differentiated between AGI and the concept of the "Singularity": "It depends on how you define AGI. AGI is not the Singularity. The Singularity refers to a system of exponentially recursive self-improvement, whose pace will quickly surpass any human-like intelligence."

Suleyman noted that AGI refers to a versatile learning system capable of performing well across all human-level training environments, including both intellectual tasks and physical labor. His skepticism towards AGI mainly stems from the progress and complexities involved in task execution by robots. Simultaneously, he believes that within the next five to ten years, it might be possible to develop a system that can excel in various environments without extensive human intervention. However, he also clarified that this does not necessarily equate to the realization of AGI, nor does it guarantee the onset of the Singularity.

Suleyman believes that AGI has been overhyped, resulting in a misunderstanding of what such systems can actually accomplish. He is more focused on developing AI companions that are beneficial to humans, enabling them to work alongside people and support their endeavors.

Last week, at The New York Times' DealBook Summit, Altman tempered expectations about AGI, stating it would materialize sooner than most anticipate, but its impact would be much less dramatic than the superintelligent phenomena previously described. He mentioned that while AGI is achievable, the world would largely continue to operate in a similar manner, albeit with faster progress. There's still a long way to go from AGI to superintelligence.

It is noteworthy that the relationship between Microsoft and OpenAI has appeared somewhat strained over the past year, despite Microsoft's support in Altman's return as OpenAI CEO. While confirming that Microsoft is developing cutting-edge AI models comparable in scale to GPT-4 and GPT-4o, Suleyman also addressed the tensions between the two parties: "There are tensions between every partnership, and it's healthy and natural. They are entirely different companies from us, operating independently, and partnerships evolve over time... Partnerships will change and must adapt to the circumstances at the time, so we'll see how things unfold in the coming years."