The one-sentence version
The question "Is AI materialist or idealist?" is actually asking: What is the relationship between mind and matter, and where does AI fit into this relationship. Different philosophical stances yield vastly different answers:
- Physicalism / Materialism: Everything is ultimately a physical process; if a brain can produce a mind, then a sufficiently similar physical process can too;
- Functionalism: The mind depends on "what is being computed," not "what it is made of"; AI can, in principle, have a mind;
- Biological Naturalism: The mind depends on the specific causal powers of the biological brain; digital computation is merely a simulation and cannot produce a mind;
- Dualism: Mind and matter are two different things; AI only possesses the material part;
- Idealism and Panpsychism: Consciousness is fundamental; the question then becomes whether AI "organizes" a unified consciousness.
AI is not just the subject of this debate, but is also becoming a testing ground to examine these stances.
Positioning
A survey and comparison of philosophical stances. Each stance has serious proponents and rebuttals; this article does not choose for the reader.
Layer 1: The Intuition
1.1 A Thought Experiment
Suppose we replace the neurons in your brain, one by one, with silicon chips that are functionally identical. With every replacement, your behavior remains unchanged. Eventually, your entire brain consists of chips.
- Does your consciousness suddenly vanish at some point?
- Does it gradually fade away?
- Or does it remain throughout?
Your intuition regarding this question essentially determines your stance on "whether AI can have a mind." (This thought experiment is often referred to as the "Fading Qualia" argument, discussed by thinkers like Chalmers.)
1.2 Clarifying Two Terms
In the context of the philosophy of mind:
- Materialism (Physicalism): Everything in the world, including the mind, is ultimately determined by physical facts;
- Idealism: The mind or consciousness is more fundamental; the material world depends on it or is constituted by it.
This is related to, but not exactly the same as, the usage of "Materialism / Idealism" in everyday political contexts.
Layer 2: Key Positions and Implications for AI
| Stance | Core Claim | Can AI have a mind? | Representative Figures |
|---|---|---|---|
| Reductive Physicalism | Mental states are brain states | Depends on whether the AI's physical state is sufficiently similar to the brain | Smart, Place |
| Functionalism | Mental states are defined by their functional role (inputs, outputs, relations to other states) and can be implemented by different substances ("multiple realizability") | Yes, as long as the correct functional organization is implemented | Putnam (early), Fodor |
| Computational Functionalism | The mind is a computation; running the correct computation results in a mind | Yes | Default stance for many AI researchers and cognitive scientists |
| Biological Naturalism | Consciousness is a biological property produced by the brain, much like digestion is a property of the stomach; programs are just simulations | No (programs on digital computers cannot) | Searle ("Chinese Room" argument) |
| Integrated Information Theory | Consciousness = the degree of information integration in a physical system | Generally no on standard computer architectures (see "AI Consciousness Checklist") | Tononi, Koch |
| Property Dualism / Substance Dualism | Consciousness is a property or substance independent of the physical | Whether a physical system can "acquire" it depends on the specific theory; traditional dualism usually says no | Descartes (substance), Chalmers (property, naturalistic dualism) |
| Panpsychism | Consciousness is a fundamental property of matter, ubiquitous, existing in extremely simple forms within elementary particles | The question becomes: Can the structure of AI "combine" tiny bits of consciousness into a unified subject ("the combination problem") | Goff, Strawson |
| Idealism | Consciousness is the only fundamental reality; matter is a phenomenon within consciousness | AI is a phenomenon appearing within consciousness; whether it has its own perspective depends on the specific theory | Berkeley; contemporary thinkers like Castrup |
| Eliminativism / Illusionism | What we call "subjective experience" is, in a sense, an illusion created by the brain | If human consciousness is also a "self-model illusion," AI could also have the same "illusion" | The Churchlands, Dennett, Frankish |
Layer 3: How AI Changes the Debate
3.1 The Old Problem of the Chinese Room, in a New Form
Searle's 1980 "Chinese Room": A person who does not understand Chinese processes Chinese characters in a room according to a rulebook, leading people outside to believe there is someone in the room who understands Chinese. Searle argued: Syntax (symbol manipulation) is insufficient to produce semantics (understanding).
Large Language Models make this thought experiment concrete: they indeed "manipulate symbols according to statistical rules," yet they perform as if they understand in many tasks.
- Functionalists say: The entire system (person + rulebook + room) understands Chinese;
- Searle says: No matter how complex, symbol manipulation is just symbol manipulation.
3.2 Symbol Grounding
Another related question is: Do an AI's words have a real connection to the world? For it, is the word "apple" merely a relationship to other words like "red," "fruit," and "sweet," or is it connected to a real apple?
- Pure text models define words only internally within language;
- Multimodal, embodied models connect words to images, actions, and physical outcomes (see "Post-training for World Models");
- This provides a path for "grounding" within a materialist framework.
3.3 AI as a Testing Ground
| Previous Dilemma | Changes Brought by AI |
|---|---|
| Can only study humans and animals; cannot open the brain for precise observation | Can fully access every parameter and activation of an AI |
| Unable to perform "gradual replacement" experiments | Can systematically change architecture, training, and scale to observe which factors correlate with specific abilities |
| Function and substrate were always tied together (only biological brains) | For the first time, highly complex systems have appeared that are functionally similar to humans in parts, but have entirely different substrates |
This cannot directly answer the "Hard Problem," but it can test many specific hypotheses regarding function.
3.4 A Practical Conclusion
Regardless of your stance, you can agree on the following in practice:
- The behavior and capabilities of AI are produced by physical processes (computation on chips); this is undisputed;
- The controversy lies in whether these processes accompany or constitute a mind;
- When the answer is uncertain, one must avoid two errors simultaneously: treating a system without a mind as if it has one (anthropomorphism), and treating a system that might have a mind as a mere tool.
Summary
| Layer | Key points |
|---|---|
| Essence of the problem | The relationship between mind and matter; AI's place within it |
| Spectrum of stances | Physicalism, Functionalism, Biological Naturalism, IIT, Dualism, Panpsychism, Idealism, Illusionism |
| Significance of AI | The Chinese Room and symbol grounding problems become concrete; AI becomes a testing ground for hypotheses in the philosophy of mind |
| Practical consensus | Computation as a physical process is undisputed; whether it constitutes a mind is disputed; avoid the dual errors of anthropomorphism and outright denial |