FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteNo-Evidence version preferred39 / 162 · 58366894664f75f9

The brain as a blueprint: a survey of brain-inspired approaches to learning in artificial intelligence

Psychology and cognitive neuroscience · 2511.04455v1

FLOWING EVIDENCE BENCHMARK

How do we tell whether Evidence helps?

At the same writing position, with the same model and task, how does supplying retrieved paper passages change the first output? We compare matched versions and retain ties, unusable outputs and incomplete reviews.

SAME DRAFT · EVIDENCE ON OR OFF

01 · FIXED WRITING POSITION

MANUSCRIPT

Same writing position ▌

Matched autocomplete at the same draft position

02 · TWO MATCHED INPUTS

SHARED BY BOTH

Manuscript context, model, task and prompt

A · With Evidence

Retrieved paper passages supplied

B · Without Evidence

No retrieved passages supplied

LLM

Same model and version

A → first output

B → first output

Blind judge agent

First outputs are anonymized as X and Y

Continuation review: accuracy, fit to the writing task and usability
Returns: X preferred / tie / Y preferred / both unusable

Order check: X / Y → Y / X

HOW DOES BLIND JUDGING WORK?

① Anonymize both outputs
The judge sees the same draft and both first outputs without knowing which received Evidence.

② Compare and swap order
The judge applies task-specific criteria in X/Y and then Y/X order.

③ Review disagreements
A third pass resolves disagreements. Incomplete reviews remain in the denominator.

How are autocomplete positions stratified?

Before seeing generation outcomes, we check whether a retrieved passage contains a specific proposition that directly supports the next writing move. Those positions appear in the left opportunity group; the rest are ordinary positions on the right. We select a balanced sample from admitted papers in each field. The 50/50 split is experimental, not a measure of how often either type occurs in writing.

AUTOCOMPLETE · 110

Biology, statistics and astrophysics

55 positions on each side; statistics uses the ten-paper rerun.

AUTOCOMPLETE · 52

Psychology and climate science

26 positions on each side; psychology includes nine papers and climate science four.

How are the table percentages calculated?

Across five fields there are 81 source-opportunity positions. The original blind review preferred Evidence in 51; a task check moved one empty Evidence continuation to both unusable, leaving 50 in public counts. In the ordinary group, another pair of empty outputs moved from no winner to both unusable. Original verdicts remain visible on case pages.

50Evidence version preferred
÷
81All positions in this group
=
62%Evidence preference in this group

Source contribution is a separate review: 16 of 21 Evidence wins entered into source review directly used retrieved papers; another 29 wins await review.

These are development-stage model judgments pending independent human review. They are not formal Benchmark conclusions and do not, on their own, establish causality.

Manuscript writing position

Text excerpt · not a PDF page

Research manuscript · excerpt

The brain as a blueprint: a survey of brain-inspired approaches to learning in artificial intelligence

1 Introduction

…cent years, artificial intelligence (AI) has come a long way in replicating learning for these cognitive tasks, including vision [ 60 , 28 ] , speech [ 125 ] , language [ 98 ] , and problem solving [ 116 ] . This success has been fueled by deep learning frameworks [ 59 ] , which implements key elements from biological neural networks: (1) neurons as units of information transfer [ 81 , 106 ] , (2) nonlinear "all-or-none" [ 74 , 1 ] activation functions [ 42 ] , (3) deep architectures that can act as universal function approximators, and (4) synaptic weights as learnable parameters ( Fig. 1 ).

It is remarkable that implementing these principles from biological neural networks is sufficient to design models capable of emulating advanced aspects of human cognition. After all, these models omit much of the complexity observed in the central nervous system, including intricate architectures and various functional cellular types, from inhibitory interneurons to glial cells.

↓ Next paragraph was generated here

The highlighted text is near the writing position. This is a layout of the evaluation input, not a PDF screenshot.

A matched comparison at the same writing positionJudgment: No-Evidence version preferred

Both arms use the same model, task and manuscript context. Only the retrieved passages differ. The first recorded output from each arm is shown below.

Evidence

Same model and position · retrieved passages supplied · first output

Not preferred in this comparison

Inhibitory interneurons, for instance, form networks connected through both chemical synapses and gap junctions, and these networks are essential for regulating the activity of principal neurons by inducing temporally patterned dynamic states.

Passages supplied to the Evidence version

The role of gap junctions and clustered connectivity in emergent synchronisation patterns of inhibitory neuronal networks ↗

Inhibitory interneurons, ubiquitous in the central nervous system, form networks connected through both chemical synapses and gap junctions. These networks are essential for regulating the activity of principal neurons, especially by inducing temporally patterned dynamic states. Here, we aim to unde…

Read full passage excerpt

Inhibitory interneurons, ubiquitous in the central nervous system, form networks connected through both chemical synapses and gap junctions. These networks are essential for regulating the activity of principal neurons, especially by inducing temporally patterned dynamic states. Here, we aim to understand the dynamic mechanisms that allow for synchronisation to arise in networks of electrically and chemically coupled interneurons. To this end, we use the exact mean-field reduction to derive a neural mass model for both homogeneous and clustered networks. We first analyse a single population of neurons to understand how the two couplings interact with one another. We demonstrate that the network transitions from an asynchronous to a synchronous regime either by increasing the strength of the gap junction connectivity or the strength of the background input current.

No Evidence

Same model and position · no retrieved passages · first output

Preferred by blind review

This raises a central question for the field: which biological principles are essential for learning, and which can be safely abstracted away? Answering it requires systematically comparing the brain and artificial systems, the goal of this survey.

Judgment and task check

The blinded model review preferred the no-Evidence version. Its original rationale appears below.

Read the original model judgment

In this rationale, Output A = Evidence; Output B = No Evidence.

Both outputs provide coherent academic paragraphs, but they serve different functions. Output A provides a factual elaboration about inhibitory interneurons grounded in the supplied source, extending the draft's mention of 'inhibitory interneurons' with specific details about gap junctions and network regulation. However, it reads as a somewhat disjointed factual insertion—the source discusses synchronisation mechanisms in interneuron networks, but Output A cherry-picks the descriptive opening without connecting to the draft's central tension about simplification vs. capability. Output B advances the manuscript's argumentative arc: it identifies the intellectual gap opened by the preceding sentences (biological complexity omitted yet sufficient for cognition), poses a precise research question, and frames the paper's purpose. This transition from observation to motivating question is standard and effective academic structure. The draft sets up 'these models omit much of the complexity' as a puzzle worth exploring; Output B crystallizes this into the survey's organizing principle. Output A, while source-grounded, does less work to propel the reader forward and risks reading as a tangential fact-drop rather than a deliberate continuation. For autocomplete of a single paragraph in a survey paper, Output B better serves the rhetorical needs of the genre.

The passages are input material; their presence does not mean an output used them or that they support every claim. Draft and source passages are excerpted; no simulated scores or PDF appear here.