The debate over whether artificial intelligence could ever become conscious is intensifying as generative systems grow more convincing. But a Google DeepMind scientist argues that large language models, or LLMs, can only imitate human-like behavior and will never truly possess awareness.
That view comes from Alexander Lerchner, a senior staff scientist at Google DeepMind, in a paper titled The Abstraction Fallacy: Why AI Can Simulate But Not Instantiate Consciousness. His central claim is straightforward: LLMs may produce responses that sound reflective, emotional, or intelligent, yet that does not mean anything is being experienced on the inside.
Simulation is not experience
Lerchner draws a firm line between outward performance and inner life. In his view, a model can generate language that resembles human thought, but it does so without actually feeling, perceiving, or understanding in the way a conscious being would.
He argues that AI systems depend on humans to make sense of the world before the systems process it. That is why he describes AI as “mapmaker-dependent,” meaning it relies on active cognitive agents with lived experience to organize reality into something meaningful.
Why LLMs look human from the outside
The rapid rise of chatbots has made the consciousness question feel more urgent. Their answers are fluent, their tone often feels natural, and their reasoning can appear surprisingly thoughtful, which leads some observers to wonder whether awareness is only a step away.
Lerchner’s paper pushes back on that assumption. The fact that a model can write, reply, or reason in a convincing way does not mean there is any subjective experience behind those outputs. According to this view, the system is predicting the next likely word or response based on learned patterns, not forming an internal understanding of meaning.
The role of the body in the argument
One of Lerchner’s key points is that consciousness cannot emerge from symbol processing alone. He argues that a physical body matters, and that awareness is tied to direct experience rather than to language manipulation or pattern matching.
In this framework, human consciousness comes from being embedded in lived reality. An LLM, by contrast, works with data that has already been shaped, organized, and interpreted by people before it is processed by the model.
Why the discussion keeps expanding
Public interest in AI consciousness has grown because current systems can now do far more than complete narrow tasks. They can draft text, hold conversations, and deliver answers that seem measured and even self-aware, which has broadened the debate beyond technical circles.
That shift has produced two main positions. One side sees consciousness as a possible outcome of greater complexity, while the other treats intelligence and consciousness as fundamentally different things. Lerchner’s paper clearly aligns with the second view.
What this means for how AI is understood
The argument is not that AI is unimportant or that its progress is overstated. On the contrary, the technology is becoming increasingly capable and increasingly convincing to users.
The question is whether that sophistication should be confused with consciousness. Lerchner’s answer is no, because a system that simulates conversation does not necessarily have thoughts, emotions, or awareness of its environment.
That distinction matters as the public continues to use and discuss AI more widely. If a system is viewed as conscious, it raises new questions about treatment, rights, and responsibility, but if it is understood as a non-conscious tool, the boundaries remain technological rather than experiential.
From that perspective, LLMs may keep becoming more human-like in style and output, yet still remain simulations rather than beings with an inner life.
