The illusion of thinking: Can you think without being moral?

In its recently published paper titled “The Illusion of Thinking,” Apple made a bold but necessary observation: today’s most advanced artificial intelligence (AI) models don’t truly “think.” At best, they mimic reasoning. At worst, they collapse under complexity, unable to generalize or course-correct in meaningful ways. But what Apple may have inadvertently spotlighted is not just a technical limitation — it is a moral one. Perhaps the illusion is not merely about cognition but about conscience. And perhaps, to think in a truly human sense is to think within a moral framework.

Apple’s research examined large language models that simulate reasoning, such as Claude, OpenAI’s o3-mini and DeepSeek. These models perform admirably when solving simple problems but frequently fail — sometimes spectacularly — as problem complexity increases. The paper calls this phenomenon a “complete accuracy collapse.” In layman’s terms, the machine begins confidently, stumbles quietly and finishes disastrously.

That collapse, while framed as a failure in logic or computational architecture, hints at something deeper. When humans encounter uncertainty or escalating complexity, we don’t just apply logic, we apply values. We ask: What matters here? What’s at stake? Who could be harmed? These questions cannot be answered by probability tokens alone.

This is why thinking, in the human sense, is more than deduction or syntax. It’s moral navigation. It involves weighing competing truths, understanding human needs and adapting to the emotional context of a situation. When a person wrestles with a difficult decision — whether to forgive a friend, to report wrongdoing at work or to risk something for the greater good — they are not “reasoning” in the sterile sense. They are engaging in ethical deliberation.

Machines don’t do this. And Apple’s findings suggest they may not be able to — not yet, and maybe not ever, without radically rethinking what artificial “intelligence” should even be — and what human “thinking” truly is.

Let’s imagine a scenario where an AI is tasked with allocating scarce medical supplies. It’s told to be efficient. It will likely prioritize based on data — age, survival rate, resource optimization. But a human ethicist might ask: Should we also factor in care responsibilities? Social vulnerability? Dignity in death? None of these considerations are computational necessities, but they are moral imperatives. They serve as guiding lights in situations where paths have not been previously trodden and must be freshly cut.

Apple’s paper inadvertently underscores this. When AI tries to “think,” it mimics patterns rather than values. It generates the appearance of deliberation, not its substance. It has no ethical core — no internal compass. As a result, when faced with dilemmas that demand more than logic, the illusion shatters because it has no moral framework to guide its deliberation. At some point, especially for something new, there is no “how” without a “why.”

Critics may argue that AI doesn’t need morality — it just needs better training, better algorithms and more guardrails. But those are surface fixes. The deeper problem is that true reasoning requires a why, not just a how. Without a moral framework, “thinking” becomes a dangerous illusion: precise, persuasive and hollow.

This is not a new debate. Philosophers from Immanuel Kant to Hannah Arendt have warned us about the perils of reason divorced from ethics. Arendt called it the “banality of evil” — when thinking becomes procedural and blind to harm. Today’s AI systems risk becoming that: intelligent but indifferent.

Apple, to its credit, doesn’t overpromise. The paper does not claim artificial general intelligence is near. It argues the opposite: We’re still far from building machines that can reason like us. But the question it leaves hanging is whether we even want machines to think like us, if they can’t also care like us.

To build ethical intelligence, we must move beyond performance metrics and benchmark tests. We must ask: What does it mean to think well? Not just to solve problems, but to solve the right problems, in the right ways, for the right reasons. That’s not just a software challenge — it’s a philosophical one.

The future of AI depends not only on breakthroughs in computing power but on breakthroughs in moral imagination. If we want machines that assist human flourishing — not just efficiency — we need more than neural nets. We need moral nets. We need frameworks that encode empathy, justice and respect for the messy, beautiful complexity of human life.

So yes, the illusion of thinking is real. But the greater danger is the illusion of meaning, believing that computation alone can replace the hard, ethical work of being human.

Apple’s research reminds us that thinking isn’t about having the right answer. It’s about asking the right questions, guided by values that machines do not — and cannot — yet possess.

“I think, therefore I am.” Descartes would probably be surprised to find his axiom a new rallying cry for the human in the AI world.

Jason Lim (jasonlim@msn.com) is a Washington-based expert on innovation, leadership and organizational culture. The views expressed in this article are the author’s own and do not reflect The Korea Times’ editorial stance.