It will take at least 7 more years! Microsoft AI CEO refutes Ultraman: Current hardware is not strong enough, AGI won't come that quickly

In recent months, multiple reports have indicated that Artificial General Intelligence (AGI) may be achieved earlier than expected.

AGI refers to a general learning system capable of performing exceptionally across a wide range of tasks, and it is considered the ultimate goal of AI development. Several leading companies in the AI field, including OpenAI, Anthropic, and Microsoft, are competing to make strides in AGI, but there are differing opinions among these companies regarding when and how AGI will be realized.

OpenAI CEO Sam Altman is optimistic about the realization of AGI. He predicts that AGI may be achieved by 2025 and believes that the process will be "quiet" and will not bring significant disruption to society. A technician from OpenAI even stated that the latest reasoning model, o1, has already met AGI standards. Although the model's cognitive abilities have not yet surpassed those of humans, its performance in most tasks has already outperformed humans.

In contrast, Anthropic CEO Dario Amodei has a more cautious prediction, setting the time window for AGI realization between 2026 and 2027.

Microsoft is also actively positioning itself in the AGI field. Earlier this year, Microsoft hired Mustafa Suleyman as the head of AI business, responsible for managing consumer AI products such as Copilot, Bing, and Edge.

The realization of AGI relies not only on breakthroughs in algorithms and models but also on support from hardware technology. Although OpenAI CEO Sam Altman previously claimed that AGI could be achieved on existing hardware, Suleyman expressed skepticism about this.

Suleyman pointed out that current hardware technology, such as NVIDIA's GB200 series chips, is still insufficient for achieving AGI. However, he is optimistic about the potential of future generations of hardware and predicts:

"The realization of AGI is possible in the iterations of the next two to five generations of hardware technology."

Regarding the specific timeline for AGI realization, Suleyman provided a relatively cautious prediction. He stated:

"I believe the probability of AGI being realized within two years is high, but a more realistic timeframe might be within the next five to seven years."

Suleyman is optimistic about the capabilities of AI systems in certain areas, such as future AI potentially being able to learn new tasks quickly with less pre-training. However, he is relatively conservative about breakthroughs in robotics. He believes that achieving similar breakthroughs in robotics may be a challenging process. Breakthroughs in robotics technology require not only powerful software algorithms but also involve complex hardware control, which is more difficult.

Regarding the concept of AGI, Suleyman provided his own definition:

"A general learning system that performs well in all human-level training environments."

However, Suleyman is cautious about the realization of AGI and the arrival of the technological singularity. He believes that even if AI systems capable of excelling in multiple fields emerge in the future, it does not necessarily mean the realization of the technological singularity. The technological singularity typically refers to the critical point at which AI surpasses human intelligence, a concept that is contentious in both academia and industry In fact, Suleyman is more interested in developing "AI systems that are truly useful to humanity." In his view, the core goal of future AI should be to become "a partner to humans," working for humans, supporting human decision-making, and being a part of the team