Bounded substrate comparison

Benchmark performance and biological plausibility

How should task utility and resemblance to biological mechanisms coexist?

engineered evaluation

Task performance

Valid claim: Meets this metric on this benchmark.

model-to-biology comparison

Biological plausibility

Valid claim: Matches these selected biological observations under this test.

Comparable axes

  • Declared model
  • Observed outputs
  • Uncertainty and alternatives

Non-equivalences

  • High accuracy does not imply brain likeness.
  • Biological resemblance does not imply engineering utility.

Comparison procedure

  1. 1.Score task independently.
  2. 2.Test biological predictions independently.
  3. 3.Publish a two-axis result.

Prohibited inference

Do not collapse benchmark rank and biological plausibility into one intelligence score or use success on either axis to certify the other.

Connected concepts and sources

Neuromorphic computingNeural coding and time
  1. NeuroBench: Advancing Neuromorphic Computing Through Collaborative, Fair and Representative Benchmarking · National Institute of Standards and Technology