Quantum-like Cognition in Process Theories: An Analysis

This paper extends quantum-like cognition models to general probabilistic process theories, demonstrating that while standard Bayesian models are insufficient, broader classical instrument models can explain sequential decision data, suggesting that only joint decision experiments violating Bell inequalities can strictly rule out classical explanations.

The Gist

The Big Question: Do We Think Like Quantum Computers?

Imagine your brain is a computer. For a long time, scientists have tried to model human decision-making using Classical Logic (like a standard calculator: $1+1=2$, and the order of operations doesn't matter).

However, humans are messy. We make "irrational" choices. If you ask a person Question A then Question B, they might answer differently than if you ask Question B then Question A. This is called an Order Effect. In classical logic, this shouldn't happen.

Because of these weird quirks, some researchers proposed that our brains actually work like Quantum Computers. In quantum physics, the order of measurements matters, and things can interfere with each other. This paper asks: Do we really need the complex math of quantum physics to explain human thinking, or is there a simpler, classical explanation we are missing?

The New Tool: "Process Theories" (The LEGO Analogy)

To answer this, the authors use a new mathematical framework called Process Theories.

Think of a Process Theory as a giant box of LEGO bricks.

• Classical Models are like a specific set of LEGO instructions that only allow you to build flat, 2D pictures.
• Quantum Models are like a set that allows you to build complex, 3D structures with moving parts.
• Process Theories are the universal instruction manual that tells you how any set of bricks (Classical or Quantum) can be snapped together.

The authors use this "universal manual" to build models of human decisions and see which bricks fit best.

The Plot Twist: The "Magic" Classical Model

The paper starts by showing that many "weird" human behaviors (like Order Effects or the Conjunction Fallacy—where people think "Linda is a feminist bank teller" is more likely than just "Linda is a bank teller") can be explained by Quantum models. This is why people love the Quantum theory.

But here is the twist: The authors prove that you don't need Quantum physics to explain these tricks.

They show that if you allow your "Classical LEGO set" to have moving parts that change the state of the system, you can replicate all the weird quantum effects.

• The Old Classical View (Bayesian): Imagine a librarian who just reads a book and tells you the title, but never changes the book. If you ask the same question twice, the answer is always the same. This model fails to explain human quirks.
• The New Classical View (Instrument/Markov): Imagine a librarian who, every time you ask a question, rewrites the book before giving you the answer. The book changes based on what you just asked.
  • If you ask Question A, the librarian rewrites the book.
  • If you then ask Question B, the librarian reads the new version of the book.
  • If you ask B then A, the book was rewritten differently, so the answer changes.

The authors prove that any sequence of human decisions can be perfectly modeled by this "rewriting librarian" (a classical Markov model). You don't need quantum mechanics; you just need a classical system that updates itself.

The Two Paths Forward: How to Really Prove Quantum Thinking

Since simple classical models can mimic quantum effects, how do we prove that human thinking is actually quantum? The authors suggest two ways to catch the "Quantum Brain" in the act:

Path 1: The "Strict Librarian" (Measurement Models)

In the real world, we often assume that when we ask a question, we are just "measuring" a pre-existing opinion, not changing the person's mind.

• If we force our models to act like strict librarians who cannot rewrite the book (only read it), then Classical models fail again, and Quantum models win.
• The Problem: Even Quantum models have a flaw here. They struggle to explain why people give the same answer if you ask the exact same question twice in a row (Response Replicability). So, this path is a bit of a dead end.

Path 2: The "Entangled Twins" (Bell Inequalities)

This is the "Nuclear Option." To prove something is truly quantum, you need to look at Parallel Decisions.

Imagine two people (or two parts of one brain) making decisions at the exact same time, without talking to each other.

• Classical Logic: If they are independent, their answers should follow a specific mathematical limit (the Bell Inequality). It's like two dice rolls; no matter how you roll them, the correlation has a maximum limit.
• Quantum Logic: If they are "entangled" (connected in a spooky, non-local way), they can break that limit. They can coordinate their answers in a way that is mathematically impossible for two independent classical dice.

The Challenge: The authors say, "If we can find real human data where two people make parallel decisions, don't talk to each other, and still break the Bell Inequality, then we have proof that the brain uses non-classical (possibly quantum) mechanics."

The Catch: Human brains are messy and interconnected. It is very hard to find a situation where two decisions are truly "parallel" and independent enough to test this. Most current experiments fail this test because the "twins" are actually influencing each other.

The Conclusion: What Does This Mean for Us?

1. Don't panic about Quantum Brains yet: The "weird" ways we make decisions (changing our minds based on order, falling for logical traps) can be explained by a sophisticated Classical model where our mental state updates itself. We don't need to invoke quantum physics just to explain these quirks.
2. The Real Test: To prove the brain is quantum, we need to find a scenario where human decisions violate Bell's Inequality (the "spooky action at a distance" test).
3. The Future: Until we find that specific data, the "Quantum-like" label might just be a fancy way of describing a very complex, self-updating classical system. The authors suggest we should keep looking for those "parallel decision" experiments, because if we find them, it would be a massive scientific breakthrough.

In a nutshell: Humans are weird, but we might just be "weirdly classical" (like a librarian who rewrites books) rather than "weirdly quantum." To prove we are actually quantum, we need to find a way to test if our minds can do the impossible "spooky dance" that only quantum particles can do.

Technical Summary

1. Problem Statement

The field of quantum-like cognition proposes that human decision-making, reasoning, and concept formation are better modeled using the mathematical framework of quantum theory (specifically non-commuting operators and interference) rather than classical probability theory. This is motivated by empirical phenomena such as order effects (where the sequence of questions changes outcomes), interference effects, and conjunction fallacies, which violate classical logic and Bayesian probability.

However, the necessity of invoking quantum physics specifically remains controversial. Critics argue that these effects might be explainable by more general classical models that allow for state disturbance, rather than requiring the full apparatus of quantum mechanics. The paper addresses the fundamental question: Do cognitive effects strictly necessitate quantum models, or can they be reproduced by more general classical probabilistic models?

2. Methodology

The authors employ Process Theories (formally, Symmetric Monoidal Categories) as a unifying framework to analyze cognitive models. This approach allows for a diagrammatic and categorical comparison of classical and quantum systems without being tied to specific physical implementations.

• Process Theory Framework: The authors work within a probabilistic process theory $\mathcal{C}$. Systems are objects (wires), and processes are morphisms (boxes).
  • Classical Objects: Represented by thin wires, corresponding to finite sets and probability distributions.
  • Quantum Objects: Represented by thick wires, corresponding to Hilbert spaces and density matrices.
  • Composition: Processes can be composed sequentially (serial) or in parallel (tensor product $\otimes$).
• Instrument Models: A decision is modeled as an instrument: a process that takes a hidden mental state $S$, produces a classical outcome $X$, and updates the state to a new $S'$.
  • Classical Instruments: Probability channels $P(X, S' | S)$.
  • Quantum Instruments: Collections of Completely Positive (CP) maps summing to a trace-preserving map.
• Measurement Models: A restricted class where instruments are "canonical updates" induced by measurements (e.g., Bayesian models for classical, POVMs/Projective measurements for quantum).
• Diagrammatic Reasoning: The authors use string diagrams to visualize and prove properties regarding commutativity, interference, and state updates.

3. Key Contributions

A. Unification of Models via Process Theories

The paper provides a high-level, diagrammatic account of cognitive effects (order, interference, conjunction fallacies, QQ-equality, response replicability) within a single formalism. This clarifies that these effects are properties of instruments (processes that update states) rather than intrinsic properties of "quantumness."

B. The Existence of Classical Instrument Models (Theorem 46)

The most significant theoretical contribution is the proof that any sequential decision data satisfying "no-signalling from the future" can be modeled by a classical Markov instrument model.

• Result: Even simple deterministic classical instruments (functions mapping state to outcome and next state) can reproduce all standard cognitive effects (order, interference, conjunction fallacies, QQ-equality).
• Implication: The mere presence of these effects in sequential data does not rule out classical models, provided one allows for state disturbance (updating the hidden mental state). Standard Bayesian models (which do not update the state) are ruled out, but general classical instruments are not.

C. Limitations of Measurement Models

The authors analyze the "Measurement Model" restriction (where instruments are fixed updates of specific measurements).

• Classical Measurement Models: Equivalent to Bayesian models; they cannot exhibit order or interference effects.
• Quantum Measurement Models: Can exhibit order and interference.
• The Conflict (Theorem 40): Even quantum measurement models (Projective or POVM) cannot simultaneously satisfy Order Effects and the Response Replicability Effect (RRE) (the requirement that repeating a question yields the same answer). This suggests that if one restricts models to "measurements," neither pure classical nor standard quantum models are sufficient to explain all data.

D. Bell Inequalities as a Strict Test for Non-Classicality

To strictly rule out classical models when allowing arbitrary instruments, the authors propose moving beyond sequential data to parallel (joint) decision models.

• Parallel Composition: Modeling two decisions made simultaneously (or on distinct mental aspects) using the tensor product $S_1 \otimes S_2$.
• Bell-CHSH Inequality: The paper argues that if cognitive data violates a Bell inequality in a parallel setting, no classical parallel model (based on local hidden variables or classical probability channels) can reproduce it.
• Significance: This provides the first rigorous structural argument for non-classicality in cognition. If such data exists, it forces the adoption of non-classical theories (quantum or other GPTs).

4. Key Results

1. Sequential Data is Insufficient for Non-Classicality: Any sequential decision dataset (e.g., asking Question A then B) that satisfies no-signalling can be perfectly fitted by a classical Markov model. Therefore, order effects and interference in sequential data do not prove the brain is quantum.
2. Deterministic Classical Models are Sufficient: Even simple deterministic functions can model conjunction fallacies and interference effects, provided the internal state is updated based on the decision.
3. Measurement Models are Too Restrictive: Restricting models to "measurements" (Bayesian or POVM) creates a trade-off: Quantum models can fit order effects but fail Response Replicability (RRE); Classical models fail both.
4. Bell Violations are the "Smoking Gun": The only way to strictly rule out classical models (even general ones) is to find joint decision data that:
   • Satisfies no-signalling (independence of marginal probabilities).
   • Violates the Bell-CHSH inequality.
   • Is naturally modeled as a parallel process (not just a sequence).

5. Significance and Future Directions

• Reframing the Debate: The paper shifts the burden of proof. It is no longer sufficient to show that quantum models fit data better than Bayesian models. One must show that data cannot be fitted by general classical instruments (which is impossible for sequential data) or that data violates Bell inequalities in a parallel setting.
• Clarification of "Quantum-like": The term "quantum-like" is often misused. The paper suggests that many "quantum" effects are actually just "non-Bayesian" or "state-disturbing" classical effects. True quantumness requires non-separable (entangled) states in parallel compositions.
• Experimental Challenges: The authors identify the primary hurdle for future research: finding cognitive experiments that naturally model parallel decisions and satisfy no-signalling. Current cognitive systems are highly interconnected, making it difficult to isolate independent variables required for a Bell test.
• Alternative Theories: The framework opens the door to exploring General Probabilistic Theories (GPTs) beyond standard quantum mechanics (e.g., Spekkens' toy theory) that might offer "quantum-like" behavior (disturbance, interference) without requiring full quantum entanglement.

Conclusion:
Tull and Ozawa demonstrate that while quantum models offer a natural language for certain cognitive phenomena, the standard sequential data used in the field does not strictly necessitate them. Classical models with state updates can explain these effects. To definitively prove the need for non-classical cognition, researchers must design experiments involving parallel joint decisions that violate Bell inequalities, thereby ruling out all classical probabilistic theories.