Where’s AI’s Next Stop? The Ultimate Analysis by Three International Experts

At present, generative AI stands at a critical juncture in the tech industry’s latest round of platform shifts. Its future trajectory is both brimming with imaginative possibilities and fraught with considerable cognitive anxiety. This article compiles the perspectives of three experts—Benedict Evans, Roger Penrose, and Kevin Kelly. In their recent articles, interviews, and speeches, each expert elaborates on their own thoughts about the core logic underlying AI’s future development. Their discussions revolve around several key questions: the evolving forms of AI, the potential for AI to develop consciousness, and how we should approach the uncertainties inherent in AI’s advancement.

In a moment when technological frenzies intertwine with cognitive anxieties,

Top thinkers from different fields

Each was given their own answer.

I hope to help everyone as they face this transformation.

Offer a rational and optimistic perspective.

Prominent tech analyst Benedict Evans:
AI is currently in a transition period toward platform shifts, and the implementation of the technology will proceed in three phases.

Original text: “Currently, AI is at a critical juncture of platform transition, and its development trajectory remains uncertain. The deployment of AI technologies will go through three stages—‘absorption, innovation, and disruption.’ At present, the focus is still on ‘absorption’ scenarios such as programming and marketing; in the long term, however, AI will fundamentally reshape core industry issues.”

Technology deployment is divided into three levels: absorption (automating clearly defined use cases and integrating them as features), innovation (developing new products, bundling and unbundling services), and disruption (redefining the problem itself). So far, most successful use cases have remained in the “absorption” stage—such as programming, marketing, customer support, and automation. But the real question is: What will “innovation” and “disruption” actually look like?

In Evans’s view, AI is currently at a critical stage of platform transition—much like the shift from mainframes to PCs or from the internet to smartphones—and its ultimate form of development remains uncertain. The deployment of this technology will unfold across three distinct levels: “absorption,” “innovation,” and “disruption.” At present, AI is still in the “absorption” phase, primarily used for automating tasks such as programming, marketing, and customer support; these applications are expected to persist for another two decades. However, the real highlights of the future lie in the “innovation” and “disruption” phases. At the innovation level, large language models (LLMs) hold the potential to shift from “relevance-based recommendations” to “intent understanding,” thereby reshaping the logic behind product bundling and unbundling. At the disruption level, AI could—much like the steam engine—replace certain human labor, redefine fundamental industry challenges, and give rise to entirely new industrial paradigms.

At the same time, he pointed out that the AI industry today presents a “dual reality of ice and fire”: tech giants are pouring massive capital into competing for dominance in this field, yet the product forms and business models remain unclear. Nevertheless, even if the industry experiences a bubble burst, the transformative impact brought by AI is already irreversible.

02
Roger Penrose, a British mathematician, mathematical physicist, and philosopher of science:

The essence of AI is “computable pattern recognition.”

“AI’s essence is ‘computable pattern recognition,’ yet it’s difficult to achieve true consciousness—current AI can only identify patterns and execute rules based on data, but it cannot grasp the underlying nature of those rules. Yet, from the perspectives of Gödel’s theorem and the physical level, today’s computing technology struggles to access non-computable processes and thus remains incapable of generating consciousness,” says Roger Penrose.

“First, true intelligence necessarily involves consciousness, and I’ve always firmly maintained that these devices do not possess consciousness—unless we introduce other elements, they’ll never acquire it. Right now, so-called ‘intelligence’ is merely a computable concept: it simply scans vast amounts of data, looks for patterns, or attempts to analyze data in this way. But intelligence, in reality, depends on consciousness.”

Gödel’s theorem demonstrates that there are aspects within mathematical systems that cannot be computed. To transcend the limitations imposed by established rules, one must grasp the deeper principles underlying those rules—a capacity for understanding that requires the involvement of consciousness. Although computers can efficiently perform computational tasks, they always remain at the level of merely applying rules without ever truly comprehending why those rules hold in the first place. Consequently, they are incapable of achieving this kind of breakthrough understanding. In other words, artificial intelligence does not know whether these rules are actually true; it lacks the conscious awareness to recognize them, cannot understand why they lead to truth, and fails to grasp the underlying rationale behind their validity.

Penrose refutes, from a fundamental logical standpoint, the possibility of current AI generating consciousness. He argues that today’s AI merely performs pattern recognition and executes pre-established rules based on massive amounts of data—it simply cannot grasp the essence of those rules themselves. From the perspective of Gödel’s theorem, mathematical systems inherently contain elements that are uncomputable. Breaking through these rules requires a deep level of understanding that involves conscious awareness—a capability that AI fundamentally lacks. From a physical standpoint, consciousness might involve non-computable processes in the quantum realm, whereas today’s computer technologies are entirely grounded in computable mathematical structures and thus remain utterly incapable of accessing this domain. In short, AI driven by today’s computers will never be able to evolve into “superhuman intelligence” with genuine consciousness.

03 Kevin Kelly, Founding Editor-in-Chief of Wired magazine

The future is the age of “artificial intelligence”—focusing on four major evolutionary directions.

The future won’t see just one single form of artificial intelligence. Instead, the evolution of AI technology in the future will focus on four key directions: symbolic reasoning (to address logic’s shortcomings), spatial intelligence (to understand the real world), emotional intelligence (to possess empathy), and intelligent agents. Kelly argues that the future won’t bring about a single, universal form of artificial intelligence; rather, hundreds or even thousands of “weak AIs”—each with its own unique strengths—will emerge. These will be “AIs,” and some may even develop “alien intelligences” with ways of thinking fundamentally different from humans. It’s precisely these differences that will become their distinctive advantages.

Uncertainty No. 1: Can Artificial General Intelligence be achieved? The answer is: We don’t know yet. Perhaps in the future, we won’t see true Artificial General Intelligence at all—but instead, we’ll see hundreds or even thousands of “weak AIs” that excel in specific domains.

Uncertainty No. 2: In the future, will AI computing rely primarily on centralized cloud systems, or will it increasingly depend on edge and local devices? Currently, most companies are betting on the centralized approach. We’re seeing large language models continue to grow in scale, the number of chips steadily increase, and model performance seemingly becoming ever stronger.

However, is this trend sustainable? We’re not sure. Larger models mean higher energy consumption and bigger data centers, which also bring about issues of centralized control.

In contrast, another trend is quietly gaining momentum: edge computing. Edge computing offers faster response times, stronger privacy protection, lower energy consumption, and greater independence.

Therefore, in my view, the future AI computing architecture will likely take on a “hybrid” form: centralized cloud computing and decentralized edge computing will coexist, but the dominant force will gradually shift toward the edge side. It is precisely this distributed structure that may well represent the true form of an intelligent society.

Uncertainty No. 3: Judging from the available data, artificial intelligence is bringing about not unemployment but rather an increase in efficiency... The introduction of AI has transformed the structure of work, but it hasn't made jobs disappear. Instead, it has freed people from repetitive, standardized tasks, enabling them to focus on more creative and judgment-intensive work.

Regarding the future direction of AI, he has identified four cutting-edge areas: symbolic reasoning to address logic’s shortcomings; spatial intelligence capable of understanding the real world; emotional intelligence endowed with empathetic capabilities; and autonomous agents that can act on their own. Moreover, he predicts that AI computing architectures will evolve toward a “cloud-edge hybrid” model, gradually shifting more processing power to the edge. At the same time, he believes AI won’t lead to widespread unemployment—it will instead free up human labor to focus on more creative tasks. In terms of mindset, he advocates adopting an “evolutionary utopia” perspective on the future—not striving for perfection, but aiming to improve just a little bit every day compared to yesterday.

From three perspectives—industrial implementation, the fundamental nature of underlying intelligence, and technological evolution coupled with mindset adaptation—three experts have outlined the core contours of AI’s future: The industrial transformation brought about by AI is long-term and irreversible, yet its disruptive potential can only be fully realized during the “innovation-to-disruption” phase. Under the current technological framework, AI will hardly attain true consciousness; thus, there’s no need to fall into extreme anxiety over “uncontrolled superintelligence.” In the future, AI will evolve toward diversified specialization and will transform human work and life through a gentle “restructuring” rather than outright “replacement.” Maintaining rational optimism is the best attitude for navigating this transformative era.