Microsoft AI CEO Predicts AI Models to Achieve "Deeply Human-Like" Continuous Planning by 2025

Microsoft AI CEO Mustafa Suleyman has projected that by the end of 2025, artificial intelligence models will transition from providing "one-shot answers to continuous planning," a development he believes will lead to "deeply human-like" capabilities. Suleyman, a co-founder of DeepMind and Inflection AI, emphasized that this significant leap will be driven primarily by increased "compute and scale," rather than the invention of entirely new algorithms.

"By the end of 2025, models will move from one-shot answers to continuous planning," Suleyman stated in a recent social media post. "With persistent memory and long-horizon planning, models will become deeply human-like." He further clarified that this evolution would require "no new algorithm, mostly compute and scale."

This prediction highlights the growing focus within the AI industry on developing models that can maintain context over extended interactions and execute complex, multi-step tasks. Persistent memory allows AI systems to recall past conversations and learned information, fostering more coherent and personalized interactions. Long-horizon planning, a critical area of AI research, enables models to strategize and anticipate future actions across a broader timeline, moving beyond immediate responses to more strategic problem-solving.

Suleyman's vision underscores Microsoft's strategic direction in AI, emphasizing the practical application of advanced models in real-world scenarios. The company has been actively integrating AI capabilities across its product ecosystem, with a focus on creating AI companions that can engage in "symbiotic relationships" with users. This approach suggests a future where AI agents act as proactive assistants, managing tasks and workflows with greater autonomy and understanding.

The assertion that compute and scale are the primary drivers, rather than novel algorithms, points to the immense investment in computational resources by major tech companies. This includes advancements in specialized hardware and the development of larger, more sophisticated models trained on vast datasets. Industry analysts note that while algorithmic breakthroughs are always sought, optimizing existing architectures with greater resources can unlock previously unseen capabilities, pushing the boundaries of what current AI can achieve.