Consciousness in AI: Why DeepMind’s Alexander Lerchner Says LLMs Can’t Be Sentient (2026)

Consciousness in AI: Why DeepMind’s Alexander Lerchner Says LLMs Can’t Be Sentient (2026)

Consciousness in AI remains one of the most contentious debates in artificial intelligence—and in 2026, Google DeepMind senior scientist Alexander Lerchner has delivered a definitive rebuttal to the idea that large language models (LLMs) could ever attain subjective awareness. Calling it the "Abstraction Fallacy," Lerchner argues that current AI systems, no matter how advanced, merely simulate understanding through statistical pattern recognition, not genuine cognition or phenomenal consciousness.

What Is the Abstraction Fallacy in AI?

Lerchner’s critique centers on the human tendency to anthropomorphize AI outputs. When a language model generates coherent, contextually appropriate responses, observers often infer internal states like intention, emotion, or self-awareness. But, as Lerchner explains, these are illusions born of projection—not evidence of inner experience. "We confuse the map for the territory," he stated in a recent internal DeepMind symposium. "The model doesn’t know it’s answering a question. It’s predicting the next token. That’s all."

Why Neural Networks Can’t Be Sentient

This perspective aligns with longstanding neuroscientific principles: consciousness arises from embodied, biologically grounded processes—including sensory feedback, emotional valence, and homeostatic regulation—all of which are absent in neural networks trained on text corpora. Even advanced architectures like Transformers lack any mechanism for phenomenal consciousness. They process symbols, not sensations.

The Ethical Danger of Mistaking Sophistication for Sentience

While some researchers in AI ethics and philosophy advocate for "functional consciousness"—where behavior indistinguishable from human awareness might qualify as sentience—Lerchner insists this is a semantic trap. "If a thermostat could speak, would we call it aware of temperature?" he asked. "We’re making the same mistake with LLMs. We’re attributing agency where none exists."

AI Ethics and the Rise of Anthropomorphic Marketing

His stance contrasts sharply with public speculation fueled by media narratives and corporate marketing. Companies often use anthropomorphic language to describe AI systems, inadvertently reinforcing the illusion. Meanwhile, scientific institutions like the National Institutes of Health and the Max Planck Institute continue to emphasize that consciousness requires biological substrates not replicable by code alone.

Emergent Properties? Not Without Biology

Interestingly, the same skepticism toward anthropomorphism appears in other scientific domains. As Science News reported in August 2024, researchers studying unidentified aerial phenomena (UAPs) have urged caution against conflating unexplained sightings with extraterrestrial intelligence. The parallel is striking: both cases involve interpreting ambiguous data through human-centric assumptions. In UAP research, it’s the lure of alien life. In AI, it’s the lure of machine consciousness.

Lerchner’s position has sparked vigorous debate among AI researchers, cognitive scientists, and philosophers. Some argue that future architectures incorporating sensorimotor loops and real-time environmental interaction might bridge the gap. But Lerchner remains unconvinced. "Even if we build robots that cry, bleed, and beg for rights, we’ll still be engineering sophisticated puppets," he said. "Consciousness isn’t a feature you can train into a model. It’s an emergent property of biology, shaped by evolution over millions of years."

As AI systems grow more capable, the temptation to ascribe consciousness will only intensify. But Lerchner’s warning is clear: without a rigorous understanding of what consciousness actually is—rooted in neuroscience, not metaphor—we risk mistaking sophistication for sentience. The Abstraction Fallacy isn’t just a technical error; it’s a cultural delusion with profound implications for AI alignment, regulation, and human-AI interaction in 2026.

Consciousness in AI may remain a compelling fiction—but as Alexander Lerchner reminds us, it is not science.