Use of AI
Woodrush Solutions uses AI tools in a controlled, risk-based way. A person checks every output, accountability stays with the director, and the level of use is disclosed, in line with the ICE Code of Professional Conduct (Rule 1, guidance xi).
AI can make errors. For that reason, the checks below are part of how the work is done, not an afterthought. Each piece of work carries two short codes: one for the role AI played, and one for the sensitivity of the information involved.
A code is written as ai followed by s, for example ai2s1.
ai2s1
The number rises with the amount of AI involvement.
The higher the number, the more input and checking is asked of the author.
| Code | What the AI does | What the author checks before use |
|---|---|---|
| ai0 | Nothing. No AI is used. | None needed. |
| ai1 | Hygiene only: checks sentence construction and clarity of reading, and applies a house style. No new content. | Reads the change list and accepts or rejects each change. |
| ai2 | Challenge and review: compares versions and raises comments and questions to supplement an individual review, in place of a traditional “cold towel” review (a last look at a completed draft with a fresh, clear mind), and suggests ideas to help ensure personal bias is removed. The author remains the author. | Answers the questions raised and decides what to change. |
| ai3 | Research and synthesis: prepares base research to a defined brief, which the author then reviews and challenges iteratively. | Checks the base sources, verifies every sourced claim, resolves provenance flags and answers the AI's questions before any conclusions are drawn. |
| ai4 | AI-drafted, human-directed: the AI drafts to the author's direction and the author edits. | Everything in ai3, plus a checklist of statements to verify, a one-line justification for using AI, and agreed wording for the recipient. |
What information is involved decides which tools may be used.
| Code | Information | Rule |
|---|---|---|
| s0 | Public, or entirely the author's own material. | Any tool may be used. |
| s1 | Client-confidential. | Approved tools and settings only, with no training on inputs. Detail is kept to the minimum needed, and a self-check is run before anything is entered. |
| s2 | Contractually restricted. | No AI is used. |
Five rules that apply to every piece of work.
A person is accountable for every output. Accountability cannot be delegated to AI.
Work is never described as checked by AI. The author verifies; the AI assists.
AI makes mistakes. Its comments are there to assist, and are used only once checked to a reasonable level.
AI use must be justified and proportionate to the risk, with care over how data is selected so that bias is not introduced.
Where a client's contract limits or prohibits AI, that overrides the codes above.
A general statement on every email, and the specific code on the work itself.
Email footers carry a general statement and a link to this page. Reports and other deliverables carry the specific code for that piece of work, and say that the author has verified the output and, where AI supplied facts, the sources.
If you would like to know how AI will be used on your project, or need it restricted or excluded, please say so at the outset and it will be recorded in the terms of engagement.
Get in Touchhugh@woodrush.com