跳到正文
菜单

机构使用AI需要一份书面规定。

老板说办公室没有用AI。隔壁标签页开着聊天机器人的那位顾问,没有人问过。

Visatory · · 阅读 5 分钟

本文尚未翻译。下方正文为英文,上方的标题和摘要则不是。

机构使用AI需要一份书面规定。
Photograph by Zoshua Colah on Unsplash.

At ten past six a counsellor has a bank statement, a sponsor's letter and a family expecting a summary of their finances on tomorrow's call. She opens a free chatbot on her personal account, uploads the scans and asks for a table. It is a good table. It is also a sponsor's account number, balances and home address, handed to a service the agency has no contract with, from an account the agency cannot see, and nothing in the file records that it happened.

Ask the owner and the answer will be that the office does not really use AI. That answer is nearly always wrong, and it is wrong because nobody has told the staff anything. A ban does not work either. In an office under deadline pressure, a ban moves the chatbot to a personal phone. What works is a rule short enough to be read, specific enough to be followed, and written before the first complaint rather than after it.

Sort the work by what goes in and what comes out.

Where agencies have an AI policy at all, it tends to be a paragraph about using the technology responsibly. That is no help to somebody with a file open at six o'clock. Two questions do the work: what is being entered, and what happens to the thing that comes out. Asking how to structure a follow-up email is a different act from uploading a transcript. A draft a manager reads is a different act from a message sent straight to a family.

Start with what cannot be taken back.

Depending on the product and the account settings, what goes into a consumer chatbot may be stored, reviewed by people or used to improve the service. Once a passport has gone in, the agency cannot retrieve it and cannot tell the family where it went. A bad draft can be rewritten. That cannot, so it is the first thing the rule has to settle.

The one-page rule.

Copy this, change what does not fit your office, put the owner's name and a date at the bottom, and have every member of staff sign it.

Never entered into any AI tool.

  • Passports, national ID cards, visa pages, birth and marriage certificates.
  • Bank statements, sponsors' letters, tax and income papers.
  • Medical results, police certificates and anything about a student's health or family circumstances.
  • Institution contracts, commission rates and passwords of any kind.

Only in an agency-approved tool, with names and numbers removed.

  • Transcripts, test reports and refusal letters, where a counsellor needs help reading them.
  • Case notes being tidied or summarised.

May be drafted with AI, then read by a person before it leaves.

  • Email and message templates that contain no student's details.
  • Translations of general checklists and office procedures.
  • Marketing drafts, checked for claims the agency cannot make.
  • Summaries of public rules, with each fact then checked against the official page.

Never decided by AI.

  • Whether a student is eligible, how much money must be shown, which deadline applies.
  • Which course to recommend to a family.
  • What goes on a visa application or a declaration to an institution.

Stays the student's own work.

  • Personal statements, study plans, genuine student statements and scholarship essays are written by the student. Staff do not run them through a chatbot to improve them.
  • Students are told to read each institution's published position on AI and to follow it.
  • Feedback from staff is given in their own words, and the file keeps the student's drafts.

Who signs off, and what is recorded.

  • One named person approves tools and keeps a register: tool, account owner, data settings, date last reviewed.
  • The counsellor who sends an AI-assisted message is its author and answers for it as if they had typed every word.
  • Any figure, date or requirement is checked by a person against the official source, and the source and the date go in the file.
  • A slip is reported the same day: what was entered, into which tool, by whom. Nobody is punished for reporting one.

Two of those lines deserve a second reading. The one about the student's own work matters because a statement polished by a model sounds like a model, and the student may then have to sit in an interview defending sentences that are not theirs. The one about reporting matters because a rule that punishes honesty produces an office where every slip is hidden, and hidden slips are the ones a family finds out about first.

The line about checking every figure is there for a plainer reason. A chatbot will state a funds requirement with complete confidence and be a year out of date, and it will not say so.

A paid tool is a contract, not a logo.

The usual next step is to buy a subscription and consider the matter closed. A subscription solves nothing until somebody has read what it says. The owner needs to know whether entered data is used for training, how long it is kept, who in the company can see it, and how it is deleted. Staff need to log in through an account the agency controls, so that when somebody leaves, their history does not leave with them. And even in an approved tool, the habit should be to remove names and numbers first, because the safest data is the data that never went in.

Begin with an honest question.

At the next staff meeting, ask without blame what people have already put into a chatbot. The answer tells the owner which line of the rule to write first, and whether something has already happened that needs dealing with. The meeting will be uncomfortable once. The alternative is a parent asking how their bank statement came to be on a service nobody in the office can name.

在 Visatory 中

专为留学机构打造,数据存放在独立的数据库中。

用您自己的学生数据免费试用 14 天。无需绑卡,停用时可导出数据。

全部文章