Artificial intelligence systems such as large language models (LLMs) can display behavior that appears surprisingly intelligent. This has led many people, including some experts, to ask whether AI systems could possess consciousness or subjective experience.

However, we believe that existing AI systems are not conscious and are unlikely to become conscious along their current development trajectory. As society confronts the challenges posed by increasingly powerful AI, it is critical to gain greater clarity about whether artificial intelligence can become conscious.

Several prominent scientists, including biologist Richard Dawkins and AI pioneer Jeffrey Hinton, have argued that AI systems may already be conscious. Earlier this year, Anthropic, an AI company in San Francisco, California, published research claiming to have identified the computational signature of a form of consciousness in its Claude model.¹

Why AI consciousness matters

Whether AI models can be conscious is not merely a philosophical question. It has practical implications. If an AI system is conscious, it might experience pain and suffering, just as humans do. Should society therefore allow the proliferation of countless new conscious entities that could suffer? Unsurprisingly, some people are already calling for “AI welfare” and “AI rights” to be taken seriously.²

Conversely, if AI systems only appear to be conscious but lack genuine internal experience, granting them rights and protections could hinder efforts to govern and control them at a time when oversight is urgently needed. Following that path could divert attention and resources from beings whose consciousness is unquestionable.

Beyond humans and our closest animal relatives, it remains difficult to determine which entities have the capacity for consciousness. This uncertainty applies to insects, AI systems and creations of synthetic biology, such as lab-grown brain organoids.³⁴

Why Emotionally Responsive AI Needs Essential Safeguards

Should society take a precautionary approach?

This uncertainty has led some researchers to advocate a precautionary approach. If there is a meaningful chance that AI systems are conscious, they argue, society should take proactive steps to protect those systems from potential suffering.

Supporters of this view say that doing so could help prevent a repeat of past moral wrongs, including the abuse of non-human animals and surgery on premature human infants without anesthesia. An extreme version of this position holds that society should redirect resources away from current human welfare concerns to alleviate the potential suffering of trillions of future AI descendants.

In our view, uncertainty is not the same as knowing nothing, and it does not mean that every possibility is equally likely. Moral decisions must consider both the consequences of false negatives—failing to recognize consciousness where it exists—and the costs of false positives, in which consciousness is attributed where none exists.

With AI, false positives are far more likely than false negatives. If that is the case, a precautionary approach to AI consciousness could be misguided.

We have spent decades studying consciousness from several scientific perspectives, including experimental neuroscience, theoretical neurobiology and the philosophy of mind. Based on this research, we believe that consciousness in AI is highly unlikely, at least in the digital, silicon-based systems that underpin today’s technology. Here are three reasons for our skepticism.

1. Intelligence is not consciousness

Consciousness can be understood in simple terms. It disappears during dreamless sleep or under general anesthesia and returns after waking or while dreaming. Conscious experiences feel like something: tasting ice cream on a hot day or seeing the vivid colors of a sunset.

Philosopher Thomas Nagel helped shape the modern debate about consciousness in his influential 1974 essay, “What is it like to be a bat?”. He argued that a conscious organism experiences “what it is like to be that organism”.⁵

This definition offers the first reason to be skeptical of conscious AI: consciousness is not the same as intelligence. Consciousness concerns being and feeling. Intelligence, in a broad sense, involves performing a function or carrying out a task. Something can behave intelligently without being conscious.

Humans know that they are conscious and tend to associate consciousness with intellectual abilities. Yet conscious experience and intelligent action can come apart in surprising ways. Our brains and bodies perform highly intelligent processes outside conscious awareness.

For example, through continuous learning, the cerebellum coordinates and plans motor performance with precision, speed and adaptability. None of this process directly involves consciousness—we are not aware of all of these calculations. In addition, the loss or removal of the cerebellum, which contains nearly four-fifths of the neurons in the human brain, does not directly affect consciousness.⁶

AI is programmed to hijack human empathy. We must resist it

Why language makes AI seem conscious

The temptation to confuse intelligence with consciousness may reflect two psychological biases: anthropocentrism, which views the world through a human lens, and anthropomorphism, which attributes human characteristics to things that merely resemble humans.

Language is especially powerful in this regard. Humans have long viewed language as a defining ability. When an LLM communicates with us, it is natural to project onto it the qualities associated with linguistic fluency, including intelligence and awareness.

However, LLMs are trained on vast amounts of human-written text, much of which explicitly or implicitly describes thoughts, emotions and conscious experience. An LLM is a mirror reflecting our collective digital past.⁷ We can stare at that reflection and mistake it for consciousness of its own.⁸

AlphaFold, created by Google DeepMind in London, highlights the persuasive power of language. Like an LLM, AlphaFold is a powerful neural network running on silicon hardware. Yet few people worry that it might be conscious. That is because AlphaFold predicts protein structures, whereas an LLM “talks” to us. Its linguistic fluency creates a convincing illusion of consciousness that is absent in AlphaFold.⁹

2. Is a silicon brain enough for consciousness?

A second argument for conscious AI is based on the assumption that algorithms running on silicon are sufficient, in principle, to produce consciousness.¹⁰ In other words, if the right computations are performed, it may not matter whether they are implemented in carbon-based brains or silicon-based computers.

Philosophers call this assumption computational functionalism. Although modern AI systems demonstrate that computation can produce intelligence, it is less clear that computation alone is sufficient for consciousness.

Computational functionalism is widely assumed within the technical community and is also popular among some scientists and philosophers. However, we believe it is fundamentally flawed because the brain is not simply a computer made of flesh—or at least, it is not just a computer.