Are mental properties supervenient on brain properties?

Joshua T Vogelstein, R. Jacob Vogelstein, Carey E. Priebe

Abstract

The "mind-brain supervenience" conjecture suggests that all mental properties are derived from the physical properties of the brain. To address the question of whether the mind supervenes on the brain, we frame a supervenience hypothesis in rigorous statistical terms. Specifically, we propose a modified version of supervenience (called epsilon-supervenience) that is amenable to experimental investigation and statistical analysis. To illustrate this approach, we perform a thought experiment that illustrates how the probabilistic theory of pattern recognition can be used to make a one-sided determination of epsilon-supervenience. The physical property of the brain employed in this analysis is the graph describing brain connectivity (i.e., the brain-graph or connectome). epsilon-supervenience allows us to determine whether a particular mental property can be inferred from one's connectome to within any given positive misclassification rate, regardless of the relationship between the two. This may provide motivation for cross-disciplinary research between neuroscientists and statisticians.

Introduction

For decades, philosophers have debated supervenience: the idea that there can be no change in mental states (thoughts, memories, emotions) without a corresponding change in physical brain states. While neuroscientists often take this for granted, there has been no formal way to test it empirically. This paper, published in 2011, bridges analytical philosophy and statistical machine learning by formalizing the philosophical "mind-brain supervenience" conjecture as a statistical hypothesis.

The challenge is twofold: turning a philosophical "logical" relation into a "probabilistic" one that handles noisy biological data, and determining if we can actually prove, using data, that the mind is derived from the brain. The authors address these gaps by framing supervenience as a pattern recognition problem. If the mind truly supervenes on the brain, then a perfect "decoder" should theoretically be able to predict a mental state from a brain state with zero error.

Formalizing Supervenience

The authors move from philosophy to math by defining supervenience as a property of a joint distribution of minds ($M$) and brains ($B$). Specifically, $M$ statistically supervenes on $B$ if the probability of two different minds existing given the same brain state is zero.

Visualizing Supervenience: (A) and (B) satisfy supervenience, while (C) violates it because a single brain state maps to multiple mental states.

As shown in the figure above, supervenience allows for multiple brain states to map to the same mental state (B), but forbids a single brain state from mapping to multiple mental states (C). The authors prove Theorem 1, which states that statistical supervenience is equivalent to a Bayes error ($L^*$) of zero. In other words, if supervenience holds, the best possible classifier will never make a mistake.

$\epsilon$-Supervenience and Testability

Recognizing that "zero error" is an impossible standard for empirical science, the authors introduce a relaxation called $\epsilon$-supervenience. A mental property is $\epsilon$-supervenient if the best possible prediction error is less than some small value $\epsilon$.

To test this, they suggest using "universally consistent" classifiers like $k$-Nearest Neighbors ($k$-NN). As more data is collected, these classifiers are guaranteed to converge to the lowest possible error ($L^*$), regardless of the underlying distribution. However, Theorem 2 presents a sobering result: it is mathematically impossible to "prove" perfect supervenience. One can never reject the possibility that the error is slightly above zero. Thus, we can conclude with high confidence that a mental property is $\epsilon$-supervenient, but we can never be confident in the negation.

Simulation and Results

Because measuring the "entirety" of a human mind and brain is currently impossible, the authors used a synthetic "Gedankenexperiment" (thought experiment) based on the C. elegans connectome (a 279-neuron wiring diagram). They simulated a mental property (odor-evoked behavior) depending on a specific "signal subgraph" of the brain's connectivity and generated synthetic "brain-graphs" with added Poisson noise.

Convergence of the error rate $\hat{L}$ toward the Bayes error as sample size $n_i$ increases, demonstrating the testability of $\epsilon$-supervenience.

In this simulation, the authors demonstrated that they could reject the null hypothesis (that error $\ge 0.1$) with 99% confidence using only a few hundred training samples. As shown in the figure, the estimated error $\hat{L}$ decreases as the number of samples $n_i$ increases, eventually falling below the threshold for $\epsilon$-supervenience.

Key Insights and Limitations

  • Supervenience is not Causality: Supervenience is a relationship of dependence, not necessarily causation. It implies the mind cannot change without the brain changing, but not necessarily that the brain "causes" the mind.
  • The "Slow Convergence" Problem: While $k$-NN is guaranteed to find the truth eventually, there is no "universal" amount of data that guarantees success. For complex human brains, the required data might be astronomical.
  • Measurement Noise: Mental properties (like psychiatric diagnoses) are often poorly defined, creating an artificial floor for the minimum achievable error rate.
  • Connectomics as the Physical Basis: The "brain-graph" is a powerful physical property for these tests, as it captures the structural essence of the nervous system.

This framework provides a rigorous statistical test for supervenience, moving the conversation from "armchair philosophy" to empirical hypothesis testing. It could eventually be used to search for "engrams" or the neural correlates of consciousness by testing which specific brain properties are most $\epsilon$-supervenient with specific mental states.

Summary updated

Abstract

The "mind-brain supervenience" conjecture suggests that all mental properties are derived from the physical properties of the brain. To address the question of whether the mind supervenes on the brain, we frame a supervenience hypothesis in rigorous statistical terms. Specifically, we propose a modified version of supervenience (called epsilon-supervenience) that is amenable to experimental investigation and statistical analysis. To illustrate this approach, we perform a thought experiment that illustrates how the probabilistic theory of pattern recognition can be used to make a one-sided determination of epsilon-supervenience. The physical property of the brain employed in this analysis is the graph describing brain connectivity (i.e., the brain-graph or connectome). epsilon-supervenience allows us to determine whether a particular mental property can be inferred from one's connectome to within any given positive misclassification rate, regardless of the relationship between the two. This may provide motivation for cross-disciplinary research between neuroscientists and statisticians.
BibTeX
@inproceedings{vogelstein2011are,
  author = {Joshua T Vogelstein and R. Jacob Vogelstein and Carey E. Priebe},
  title = {{Are mental properties supervenient on brain properties?}},
  booktitle = {Scientific},
  year = {2011},
  doi = {10.1038/srep00100},
}
Figure