FlowingFlowing
← Back to all development cases
Development case · blind model judgment pending independent human review; not a formal Benchmark conclusion.
AutocompleteWith cards better140 / 162 · 4f9a6fbd376f6c6c

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 positions and cards selected?

The evaluation uses seven fields with forty fixed ordinary autocomplete positions in each, for 280 total. A position does not need to require a literature citation in its next sentence. Each Flowing request receives three to five retrieved cards: five when all pass the relevance gate, without padding with weak cards. The run supplied 1,355 cards: 244 positions received five, 27 received four, and nine received three.

ADOPTION · 150/280

Adoption coverage

After drafting, the model returns an evidenceUse array identifying cards that materially affected facts, correction, direction, scope, terminology, or expression. A nonempty array counts as adoption.

OUTCOMES · 70 / 7 / 30

Full outcomes after adoption

Paired outcomes are counted only when the model reports adoption and both arms produce text. Among 107 comparable positions, Flowing was preferred in 70, 7 tied, and no-Evidence was preferred in 30; one tie was unusable in both arms.

How are the two table percentages calculated?

ADOPTION COVERAGE

150adopted
÷
280all positions
=
54%

FLOWING PREFERENCE AFTER ADOPTION

70Flowing preferred
÷
107comparable adopted positions
=
65%

evidenceUse is the model’s structured report of its own generation process. It does not require a citation and is not rejected merely because a faithful source paraphrase is not verbatim. Actual manuscript citations continue to use separate strict source verification. The previous 140-position result remains retained for traceability.

These are development-stage model-reported adoption and blind-judge results 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

Beyond the aforementioned basic principles of deep learning, progress in machine learning is still consistently fueled by breakthroughs in neuroscience. For example, convolutional neural networks were inspired by the architecture of the mammalian visual system [ 44 ] , specifically by implementing hierarchical feature extraction from input patterns [ 32 , 33 , 34 , 60 ] , leading to visual models that could scale to large and complex images ( Fig. 2 ). More recently, vision transformers implement attention mechanisms that are essential in human perception [ 28 ] .

While there is also a long history of unsuccessful attempts at implementing neuroscience principles into artificial models, historically most breakthroughs in AI are somewhat inspired by aspects of the brain. In parallel, this also suggests that failures to implement aspects of the brain into AI models often reflects limits in our understanding of biological computations. This prompts a need to regularly evaluate current advances in neuroscience, and translate them into artificial models to not only advance machine learning but also test theories of brain function.

↓ 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: With cards better

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

Preferred by blind review

A central motivation for this translation is that the brain solves problems of credit assignment and perception under constraints that differ markedly from those assumed by backpropagation, and artificial networks provide a controlled test-bed in which the capabilities of more biologically plausible learning rules can be examined.

Passages supplied to the Evidence version

A Review of Neuroscience-Inspired Machine Learning

Understanding the nature of credit assignment in the cortex is one of the fundamental goals in neuroscience and finding answers to the questions related to it would provide us with direct insight into many other fundamental questions related to brain function and organization. Deep learning, a close…

Read full passage excerpt

Understanding the nature of credit assignment in the cortex is one of the fundamental goals in neuroscience and finding answers to the questions related to it would provide us with direct insight into many other fundamental questions related to brain function and organization. Deep learning, a close cousin of neuroscience, is now able to reach and surpass human-level performance on many tasks, solving problems similar in spirit to what the brain does using the backpropagation of errors algorithm, which is highly neuroscientifically implausible. However, machine learning problems and networks provide an invaluable test-bed to experiment with and to test the capabilities of more bio-plausible algorithms.

Model-reported card use

The model did not report that any card materially affected this completion.

No Evidence

Same model and position · no retrieved passages · first output

Not preferred in this comparison

This survey examines the bidirectional relationship between neuroscience and artificial intelligence, asking how brain-inspired principles have shaped learning algorithms and, conversely, how artificial models can serve as testbeds for theories of brain function. We organize the field along the levels at which biological inspiration has been translated into computational mechanisms, from sensory processing to memory, attention, and reinforcement learning.

Result of this comparison

The blinded model review preferred the Evidence version. Its original rationale appears below.

Read the original model judgment

Output A provides a coherent continuation grounded in the source material. It connects the draft's theme of neuroscience-AI translation to the specific topic of credit assignment and biologically plausible learning rules, directly supported by the source's discussion of 'credit assignment in the cortex,' 'backpropagation of errors algorithm, which is highly neuroscientifically implausible,' and 'machine learning problems and networks provide an invaluable test-bed to experiment with and to test the capabilities of more bio-plausible algorithms.' The paragraph flows naturally from the draft's closing sentence about testing theories of brain function. Output B is unusable due to multiple unsupported claims. It introduces 'This survey' as if the manuscript itself is a survey paper, which is not established in the draft. It claims the paper 'organizes the field along levels...from sensory processing to memory, attention, and reinforcement learning' — a specific structure not mentioned or implied in the draft or source. The source does not mention sensory processing, memory, or reinforcement learning as organizational levels. The draft focuses on visual processing and attention specifically, not a broad survey structure. These are material factual claims about the paper's scope and organization that lack support.

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.