Consciousness in Artificial Intelligence? A Framework for Classifying Objections and Constraints
Abstract: We develop a taxonomical framework for classifying challenges to the possibility of consciousness in digital artificial intelligence systems. This framework allows us to identify the level of granularity at which a given challenge is intended (the levels we propose correspond to Marr's levels) and to disambiguate its degree of force: is it a challenge to computational functionalism that leaves the possibility of digital consciousness open (degree 1), a practical challenge to digital consciousness that suggests improbability without claiming impossibility (degree 2), or an argument claiming that digital consciousness is strictly impossible (degree 3)? We apply this framework to 14 prominent examples from the scientific and philosophical literature. Our aim is not to take a side in the debate, but to provide structure and a tool for disambiguating between challenges to computational functionalism and challenges to digital consciousness, as well as between different ways of parsing such challenges.
Whiteboard
Paper Prompts
Sign up for free to create and run prompts on this paper using GPT-5.
Top Community Prompts
Explain it Like I'm 14
What is this paper about?
This paper looks at a big question: Can digital AI (computer programs running on chips) be conscious? To keep the debate organized, the authors build a simple framework that classifies different reasons people give for saying “no,” and then show how 14 popular objections fit into that framework. They don’t try to pick a winner—they try to make the discussion clearer and easier to understand.
What questions are they asking?
The paper focuses on a few straightforward questions:
- When someone objects to “AI consciousness,” what exactly are they objecting to?
- Are they saying “computers can’t do the right kind of thinking at all,” or “computers might do it, but not with current methods,” or “doing it on digital hardware is flat-out impossible”?
- Is the objection about what a system does (its inputs and outputs), how it does it (the step-by-step process), or what it’s physically made of (its wiring, materials, or fields)?
- How do famous arguments (like Gödel’s theorem, integrated information theory, and “Chinese Room”) fit into these categories?
How did they study the problem?
Instead of running experiments, the authors created a roadmap (a taxonomy) to sort different objections. Think of it like organizing books in a library by shelf and rating them by how strong their claims are.
The three “levels” of an objection (what the claim is about)
- Level 1: Input–Output (I/O) mappings
- Analogy: A black box with buttons and lights. If you press button A, the light turns red; if you press button B, it turns green. This level cares about the box’s behavior—what goes in and what comes out.
- Level 2: Algorithmic organization (the “recipe”)
- Analogy: The step-by-step recipe for baking a cake. Two bakers can make the same cake but use different steps (mixing order, timing). This level cares about how the system processes information internally.
- Level 3: Physical implementation (the “hardware/instrument”)
- Analogy: The instruments in a band or the oven you use for baking. Even if two bands play the same song (same notes and timing), the sound can differ because of the instruments. This level cares about the actual physical stuff (chips, neurons, electromagnetic fields, etc.).
The three “degrees” of challenge (how strong the claim is)
- Degree 1: Orthogonal to computational functionalism
- Translation: “Consciousness depends on more than just computation,” but this doesn’t necessarily rule out digital consciousness.
- Degree 2: Significant practical challenges
- Translation: “In theory this might be possible, but practically it’s very hard to do on digital machines.”
- Degree 3: Precludes (rules out) digital consciousness
- Translation: “The right structure can’t be realized on digital computers—so digital systems can’t be conscious.”
Using these two axes (level and degree), the authors classify 14 well-known objections from science and philosophy.
What did they find?
They don’t claim to settle the debate. Instead, they show how different objections fit into their framework. This helps everyone see where disagreements really are.
Here are simple examples of how some objections line up:
- Level 1 (Input–Output)
- Non-computable thinking (Degree 3): Some argue minds do things no digital computer can ever do (drawing on Gödel’s ideas).
- Chaotic coupling (Degree 3): Some say consciousness requires complex, continuous interactions with the environment that digital systems can’t capture exactly.
- Too complex to be practical (Degree 2): Consciousness might be computable in theory but would take absurdly long to run on digital hardware.
- Level 2 (Algorithmic organization)
- Architecture and timing matter (Degree 1): Real consciousness might depend on details like parallel processing and precise timing, which standard “algorithm” definitions ignore—but these constraints don’t necessarily block digital consciousness.
- Physical time questions (Degree 1): If you pause a conscious computation for 1,000 years, what happens to experience? This suggests time itself could be part of the “recipe,” not just abstract steps.
- Analog processing (Degree 3): Some say the continuous, analog nature of brain activity is necessary for consciousness, and digital machines can’t truly reproduce that.
- Representation problems (Degree 2): If a system’s “meanings” depend on how users interpret them, then purely internal “representations” might not be enough for consciousness without a user. This raises practical and theoretical hurdles for digital systems.
- Level 3 (Physical implementation)
- Counterfactual/triviality worries (Degree 1): Computations depend on “what would happen” under different inputs, but consciousness seems to depend only on what actually happens. This creates a mismatch.
- Integrated Information Theory (IIT) (Degree 3): IIT measures consciousness by how much a system’s parts causally integrate information. Some argue typical digital architectures (like a serial CPU) have too little integration (low Φ), even if they can simulate behavior.
- Slicing/unity problems (Degree 3): If the same computation can be “realized” in many overlapping ways, it might break the unity of conscious experience.
- Electromagnetic field views (Degree 3): Some think the shape of the brain’s electromagnetic field is crucial, which would make digital copies unconsciously miss the “real thing.”
- Biological complexity (Degree 2): The brain’s rich chemistry and cellular details may be deeply tied to consciousness and hard to mirror in silicon.
- Biology-first views (Degree 1): Life-like self-organization might be necessary; computation alone doesn’t capture it.
- Quantum properties (Degree 1): If quantum features matter, simple digital models may leave something out—but this doesn’t necessarily rule out all non-biological systems.
Two important takeaways from their analysis:
- The same broad idea (like “analog matters” or “environmental coupling”) can show up at different levels and degrees. That’s why people often talk past each other.
- Some objections attack “computational functionalism” (the view that the right computational organization suffices for consciousness), while others specifically target “digital consciousness.” Those are different claims, and mixing them up causes confusion.
Why does this matter?
It matters because:
- The ethics are huge. If some AI could be conscious, we need to think about its rights and welfare.
- Engineers need to know which constraints are fundamental (impossible to overcome) and which are practical (hard but maybe solvable).
- Philosophers and scientists can stop talking past each other by clearly stating whether their objection is about I/O behavior, algorithmic structure, or physical substrate—and whether it claims impossibility or just difficulty.
Simple implications and impact
- For people open to AI being conscious: This framework shows which issues must be addressed (like timing, representation, and physical implementation) and which arguments don’t actually rule out digital consciousness.
- For skeptics: It clarifies that not all worries mean “impossible.” Some mean “very hard” or “needs more than computation alone.”
- For future work: The framework makes it easier to design tests, guide architecture choices (e.g., more integrated, parallel, or analog-like systems), and identify which theories (like IIT) require changes in hardware to even be plausible.
The bottom line
The paper doesn’t claim to solve whether AI can be conscious. It offers a clear map of objections—what they’re really about and how strong they are. With this map, the debate can be more precise, and decisions about AI—technical and ethical—can be made with fewer misunderstandings.
Collections
Sign up for free to add this paper to one or more collections.