Skip to content
Violetta Liszka

AI Lab · Method / Human–AI Workflows

Human–AI Workflow Experiment

A documented Human → AI assistance → Human verification cycle, applied as a working method across projects.

Fig. 01 — working cycle

Human → AI → Human

  1. 01

    Question

  2. 02

    Frame

  3. 03

    AI assistance

  4. 04

    Verify

  5. 05 · rejection gate

    Reject / refine

  6. 06 · authorship

    Human decision

  7. 07

    Output

  8. 08

    Reflect

Return path — rejected material re-enters at 02, with the constraints rewritten

Question → Frame → AI assistance → Verify → Reject / refine → Human decision → Output → Reflect

01

Context

A working method, documented so that it can be inspected and reused — not a product or a proprietary system.

02

Question / Challenge

Where in a piece of work should AI sit, and what must stay with the person?

03

My role

Method design and documentation across the projects where it is applied.

04

Process

  1. 01Define the question the work has to answer.
  2. 02Set constraints: scope, sources, format and what would count as a wrong answer.
  3. 03Use AI assistance for research, drafting and structuring.
  4. 04Verify against sources and reject weak or unsupported claims.
  5. 05Produce the evidence and record what was changed.
  6. 06Reflect, and adjust the constraints for the next cycle.

05

AI contribution

Research compression, drafting and structural suggestions inside a fixed brief.

06

Human judgement

The question, the constraints, the verification and the decision. Every claim that could not be traced was removed.

Contribution map

Human

  • Research question
  • Criteria and constraints
  • Source verification
  • Rejection and refinement
  • Ethical judgement
  • Final decision

AI

  • Research summarisation
  • Drafting
  • Structuring
  • Implementation assistance

07

Output

A documented methodology applied across the AI-related projects in this portfolio.

08

AI disclosure

Why AI was used
To test where assistance improves a piece of work and where it degrades it.
What was verified
Every factual claim produced inside the cycle, against primary sources.
What AI got wrong
Fluent but unsupported assertions, invented citations, and premature closure on a single framing.
Rejected or changed
Anything untraceable; drafts that answered a different question from the one set.
Human contribution
The question, the constraints, the verification and the decisions.
Ethics & authorship
The method is described as practice, not technology. No proprietary system is claimed.

Skills demonstrated

  • Workflow design
  • Verification practice
  • Documentation
  • Critical evaluation

Gallery

Fig. 01 — working cycle

Human → AI → Human

  1. 01

    Question

  2. 02

    Frame

  3. 03

    AI assistance

  4. 04

    Verify

  5. 05 · rejection gate

    Reject / refine

  6. 06 · authorship

    Human decision

  7. 07

    Output

  8. 08

    Reflect

Return path — rejected material re-enters at 02, with the constraints rewritten

Question → Frame → AI assistance → Verify → Reject / refine → Human decision → Output → Reflect

Fig. 01 — the working cycle, drawn as a site-native research diagram.

—

Reflection

The rejection step is the one that carries the weight; without it the cycle simply launders plausible text.

Related projects