The difficult legacy of Turing's wager

Andrew Thwaites, Andrew A. S. Soltan, Eric Wieser, Ian Nimmo-SmithJ. Comput. Neurosci 2017 · OpenReview

Abstract

Describing the human brain in mathematical terms is an important ambition of neuroscience research, yet the challenges remain considerable. It was Alan Turing, writing in 1950, who first sought to demonstrate how time-consuming such an undertaking would be. Through analogy to the computer program, Turing argued that arriving at a complete mathematical description of the mind would take well over a thousand years. In this opinion piece, we argue that — despite seventy years of progress in the field — his arguments remain both prescient and persuasive.

Introduction: The Inverse Problem

The central challenge in modern neuroscience is the "inverse problem": how do we move from observing the brain's inputs (sensory data) and outputs (behavior) to identifying the exact mathematical equations that govern its internal processing?

Early computing pioneers at a console, representing the era when Turing first proposed his 1000-year wager on modeling the mind.

In recent years, massive initiatives like the European Human Brain Project (HBP) and the US BRAIN Initiative have received billions in funding with the goal of constructing accurate mathematical models of the brain. There is a growing sense of optimism that "solving" the brain is a matter of decades, not centuries. However, revisiting Alan Turing’s 1950 "wager" suggests that the complexity of the brain might defy our current technological trajectory.

Turing’s Wager and the Search Space

In his 1950 essay, Computing Machinery and Intelligence, Turing proposed that guessing the internal calculations of even a simple computer program from its inputs and outputs is nearly impossible. He argued that to determine the calculations of a simple program, an investigator might need to test more candidate equations than there are atoms in the observable universe.

For the human brain, which handles tens of thousands of concurrent inputs, this "combinatorial explosion" is even more severe. While modern neuroimaging (fMRI, EEG) allows us to see "inside" the machine, these measurements only serve to constrain the search space. They do not eliminate the need to guess the underlying equations.

The Bottleneck of Hypothesis Generation

While supercomputers can test thousands of equations per second, they are poor at generating plausible hypothetical equations for "intrinsically complicated" processes like language comprehension or face recognition. This highlights a crucial distinction between the Turing Test (imitating human behavior) and Turing’s Wager (mathematically modeling the brain). Modeling is a significantly harder "sub-problem." If you can model the brain perfectly, you pass the Turing Test by default; the reverse is not true.

The Kymata Atlas, a database maintained by the University of Cambridge, archives equations used to describe human cognition. Notably, for complex features like "morality," "consciousness," and "ego," the database remains effectively empty, illustrating the gap between our observational capabilities and our mathematical understanding.

A Shift Toward Realism

The optimism of the early 2010s regarding whole-brain simulation has faced significant hurdles. The Human Brain Project was eventually forced to scale back its "complete model" ambitions following scientific criticism, shifting toward more practical, clinical objectives.

The authors argue that the most effective path forward is not "comprehensive" modeling, but rather:

  • Focusing on tractable sub-problems related to specific clinical conditions or technologies (e.g., hearing aids, prosthetics).
  • Seeking basic underlying principles of cortical function to reduce the mathematical search space.

Conclusion

Turing’s wager remains a powerful plea for realism. While we cannot predict if a "paradigm shift" in mathematics or artificial intelligence might eventually bypass the brute-force search problem, identifying the basic biological principles that govern the brain remains the best filter to narrow down the unimaginably vast number of possible equations.

Summary updated

Abstract

Describing the human brain in mathematical terms is an important ambition of neuroscience research, yet the challenges remain considerable. It was Alan Turing, writing in 1950, who first sought to demonstrate how time-consuming such an undertaking would be. Through analogy to the computer program, Turing argued that arriving at a complete mathematical description of the mind would take well over a thousand years. In this opinion piece, we argue that — despite seventy years of progress in the field — his arguments remain both prescient and persuasive.
BibTeX
@article{thwaites2017difficult,
  author = {Andrew Thwaites and Andrew A. S. Soltan and Eric Wieser and Ian Nimmo-Smith},
  title = {{The difficult legacy of Turing's wager}},
  journal = {J. Comput. Neurosci.},
  year = {2017},
  url = {https://openreview.net/forum?id=5vHXAeuOm4},
}
Figure