Proof of human written content in the AI world.

30 Mar, 2025

Proof of human written content in the AI world.

AI-generated text (content for the purpose of this article) is indistinguishable from human created content. Even better at times. How to give a verifiable proof to human generated content?

Facts:

Humans write in keystrokes.
Human typing follows patterns.
Humans edit in logical and methodic patterns.
Blockchain makes it difficult to cheat the proof.

Solution

1. Keystrokes as a fingerprint of human typing.

Base logic

Keystroke Pattern Analysis: Measure intervals between key presses
- Calculate mean/variance of two-key sequences pairwise
- Build a keystroke profile for every human.
- Prompt human to type a long sentence before and after finishing their typing session.
- Adjusts for slowing over long sessions
- Recognise habitual phrases, fast words and slow words through repeated sequence matching
Cheating detection:
- Flag deviations from established typing patterns
  - Patterns are established after the first certified typing session by a human.
- Paste a huge chunk of text and start editing
  - Edits not deviating away from the pasted content
  - Paraphrasing detection

2. The Digital Paper Trail

section written with the help of AI, verified by human

Captured Metadata

Metric	Description	Authenticity Indicator
Input Sources	Keyboard vs paste events	>80% direct typing suggests human origin
Edit Patterns	Frequency of undo/redo actions	Natural human: 2-5 revisions per paragraph
Cursor Behavior	Mouse vs keyboard navigation	Hybrid control patterns indicate human
Session Rhythm	Active vs idle time distribution	Humans show variable focus intervals

Detection Heuristics:

AI-assisted writing shows:
- Burst pasting followed by minor edits
- Uniform time between revisions
- Linear cursor movement patterns
Human writing exhibits:
- Nonlinear editing (jumping between sections)
- Erratic pause durations
- Frequent style/tone adjustments

3. Blockchain certification

section written with the help of AI, verified by human

Certification

Data Hashing:
- SHA-3 hash of:
  - Final content
  - Keystroke timestamps
  - Edit history JSON
Blockchain Anchoring:
- Ethereum Smart Contract
Verification Portal:
- Public DApp for:
  - Hash re-computation
  - Temporal pattern validation
  - Blockchain record lookup

Implementation Considerations

Privacy
- Zero-knowledge proofs for sensitive data
- Identity stored with anonymization
AI Adaptation Resistance
- Algo updates as AI typing simulators evolve
- Incorporating user feedback

"In the arms race between AI generation and authentication, behavioral biometrics provide our first durable moat." - Adapted from DeepSeek R1 reasoning logs

Written by Rushi. Co-author: Deepseek R1