Whether a machine could ever be conscious is a live question in neuroscience, computer science and physics. The honest summary of the current evidence is short: no leading scientific theory says that today’s AI systems are conscious, several theories leave the door open in principle, and researchers do not yet agree on what consciousness is or where it comes from.
That disagreement is the starting point. One school of thought treats consciousness as an emergent property, something that appears when information in a system reaches a sufficient level of complexity and organization, in the way that life is thought to have emerged from non-living matter. A second school holds that consciousness requires a biological system that does not just process information but feels, so that emotion, sensation and biochemistry are part of the picture and data processing alone is not enough.
Two leading theories
Integrated information theory (IIT), developed by the neuroscientist Giulio Tononi, proposes that consciousness arises when information in a system is highly integrated, meaning the parts combine to produce an experience that cannot be reduced to the parts taken separately. The theory assigns a measure of integration, written as phi. On this view, a future computer with a network architecture that produced a very high phi value could in theory have something like conscious experience, though IIT’s proponents argue that conventional digital hardware is poorly suited to that.
Global workspace theory (GWT), associated with Bernard Baars and later developed by Stanislas Dehaene and others, describes the brain as a set of specialized subsystems. Information becomes conscious when it is broadcast to a shared “global workspace” that many of those subsystems can access at once. A system that routed information through a central hub shared across modules could, in principle, behave in a consciousness-like way. Transformer architectures, the basis of today’s large language models, have some superficial resemblance to this idea but are not organized as a global workspace in the sense the theory requires.
A 2023 report by Patrick Butlin, Robert Long and a group of neuroscientists, philosophers and AI researchers surveyed these and other theories, including recurrent processing theory and higher-order theories, and derived a list of “indicator properties” for consciousness. The authors concluded that no current AI system satisfies the indicators, but that there are no obvious technical barriers to building systems that do.
How artificial networks differ from brains
The human brain contains roughly 86 billion neurons, each with on the order of thousands of connections. Modern neural networks have very large numbers of parameters, but these are statistical weights, not biological neurons. The brain uses both chemical and electrical signaling; machines use only electrical signals and mathematical operations. The brain also exhibits plasticity, meaning its physical structure changes with experience, and its activity is shaped by hormones and emotional states. Nothing equivalent exists in current machines.
Researchers in the biological camp argue that this is the fundamental barrier. A machine can learn, in the sense of adjusting its weights, but it does not feel, because it has no biological substrate for feeling.
The quantum hypothesis
A third, more speculative position comes from the physicist Roger Penrose and the anesthesiologist Stuart Hameroff. Their theory of “orchestrated objective reduction” (Orch OR) holds that consciousness originates in quantum processes inside microtubules, structures within neurons. If that is correct, classical computing of the kind used in current AI could not produce consciousness, and something like quantum neural networks would be required to reproduce those effects. The theory remains controversial, and there is no experimental evidence that quantum effects in microtubules play the role it describes.
What would have to happen first
Taken together, the theories suggest that machine consciousness is theoretically possible but depends on at least three unresolved steps:
- A complete understanding of the mechanisms by which the brain produces conscious experience.
- An artificial architecture capable of the kind of integration or global broadcasting that the leading theories describe.
- A physical substrate able to reproduce the complex, self-modifying dynamics of biological neurons.
Science is still at the first step. Large language models can produce fluent answers to questions about their own processing, which can look like self-awareness, but researchers generally treat this as learned imitation rather than evidence of inner experience. The area in which current AI comes closest to the components of consciousness is cognitive complexity, not subjective experience.
Some researchers argue that an embodied AI system, one with a body, sensors and continuous memory, would come nearer to the early ingredients of consciousness. Without biological features such as emotion, pain and hormonal coupling, however, most of the scientific positions reviewed here would describe machine consciousness as a synthetic simulation rather than genuine experience.
