A researcher from Google DeepMind has argued that large language models (LLMs) may never achieve true consciousness, despite their rapid progress and human-like responses. The claim challenges a popular belief that advancing AI systems will eventually “wake up” or become self-aware. According to the report, this view comes from a recent research paper that questions the very foundation of linking intelligence with consciousness.
The core argument is what the researcher calls the “abstraction fallacy.” LLMs operate by processing symbols, patterns, and probabilities—essentially manipulating language at a mathematical level. But consciousness, the paper argues, is not something that emerges from abstract computation alone. In other words, simulating human-like conversation is not the same as having real experiences or awareness.
Another key reason is the lack of physical embodiment and real-world interaction. Humans and animals develop consciousness through continuous interaction with the physical world—through senses, bodies, and biological processes. LLMs, by contrast, are trained on static datasets and operate in isolated digital environments. Without sensory experience, goals, or survival-driven behavior, they lack the conditions many scientists believe are essential for consciousness.
The conclusion is clear: even if AI becomes more intelligent or capable (moving from narrow AI to AGI), it may still remain fundamentally non-conscious. This means future AI systems could become extremely powerful tools—capable of reasoning, creativity, and complex decision-making—without ever having feelings, self-awareness, or subjective experience. The debate, however, is far from settled, as some researchers argue that future architectures—not current LLMs—might still open the door to machine consciousness.
