AI is here to stay

Artificial Intelligence (AI) is here to stay; it is too late to wind back the clock. It is a waste of time and energies talking about how bad it is but do put your energies into putting in some constructive guardrails. Every entity, no matter how small, should have an AI Policy. Start off small with a one pager (or commensurate with the size of your entity and your other policies). 

You will need to decide beforehand how much AI use you will allow in your entity, but good luck if your intention is to not allow its use at all. In my opinion, its use is too widespread for it to be banned completely. I know many schools have chosen to allow AI but establish limits, require AI logs and guidance on how to best use it.

Here are some important considerations on AI use.

1. AI is only as good as the data that informs it. AI sweeps the web (or the data that has been fed into the AI software. The web holds a lot of good and valuable information so a balanced view can often be retrieved. The Internet Archive alone holds more books than you or I could possibly read. The problem most of us have is that we are too limited by or in:

   ·       our questioning of the web

   ·       the websites we frequent for information

   ·       The algorithms of our website browsers

   ·       Commercial and marketing strategies 

   
   

2. If we learn to ask the right questions in both our internet searches and AI, we open many more doors. In asking the right questions, and keeping track of the questions, we can also identify influencing biases to ensure greater accountability. A common practice is to ask AI to produce something in a particular style. I experimented with this by asking for instructions on making a cup of tea in the style of Tennyson, which instantly produced amazing results. By referencing a style, you are foretelling the preferences and thus declaring biases. Something that is not as easy to achieve with a simple internet search.

3. AI can be wrong. The term used for this is hallucinations and is more common than you think, as AI can extrapolate anticipated information. AI software can be challenged when information is questionable but of course the problem is we don’t always know when AI has made assumptions. So do maintain an element of cynicism when using AI and challenge it frequently.

   
   

4. AI information should be checked with someone taking ultimate ownership of the data. A common process for entities is for a staff member to prepare a document which will ultimately be endorsed and ultimately owned by a board. Good practice is for such a document to have an explanatory memorandum (some sort of instruction document titled item paper or similar) to accompany and explain the document. If AI is used to create parts of that document, then this would be the time to explain the type and extent of AI use. This enables the board to understand the final product and take ownership of the document. It also means a paper trail exists if any issues arise later.

So, with these considerations in mind, my suggestion is that Artificial Intelligence policies should contain the following elements.

• The extent to which AI can be used by staff.

• How AI material is to be referenced, logged or explained.

While many may choose to put much more information in their policies, I believe this would be sufficient to provide guardrails for many, especially small entities, with appropriate monitoring and resourcing doing the rest.

Having your head in the sand is not the answer, having appropriate guardrails in place is essential.