AI with Ashley: What Should I Not Put Into AI?

ALEXANDRIA, VA — After a very successful introduction to this new column, AI with Ashley (I’m Ashley: an Alexandrian, mom, florist, co-founder and journalist), my mom read the article and wrote in with her first question. Thanks, Mom!

My mom, Laura, is in her early 70s, but she has always been a pioneer adopter of technology. She worked for Booz Allen and ran the database for a private all-girls high school for more than two decades. She loves ChatGPT and uses it to ask all sorts of bureaucratic, and sometimes medical, questions.

She wanted to know: “What shouldn’t I put into it?”

This is an excellent question, and one worth thinking about.

A Chatbot Is Not a Diary

Chatbots are remarkably easy to talk to. That is part of what makes them useful. You can ask an AI to help write an email, explain a confusing document, plan a vacation, brainstorm a birthday party or untangle a problem. After a while, typing into that little box can start to feel less like using software and more like talking to someone.

And yes, I have named my chatbots…

But a chatbot is not a diary, a safe-deposit box or a confidential conversation with your doctor or lawyer.

So, what shouldn’t you put into one?

Start with the obvious: passwords, PINs, authentication codes, Social Security numbers, passport numbers, detailed stock portfolios, bank-account information and credit-card numbers. An AI generally doesn’t need these things to help you. And if it does, I’d take a long pause before continuing.

Look Before You Upload

Be careful with documents, too. Before uploading a medical record, tax return, legal document, school record or business file, look at what is actually on it. Names, addresses, birth dates, account numbers and other identifying information may have nothing to do with the question you want answered.

Often, you can simply remove them. Don’t you love copy and paste!

Suppose you want help understanding a medical report concerning your child. The AI probably doesn’t need your child’s full name, birthday, school and home address. You can ask, “Can you explain these results for a 10-year-old?” and provide only the relevant, de-identified information.

Don’t Forget Other People’s Privacy

The same principle applies at work. Don’t paste confidential client information, proprietary company material, personnel records or anything else you have an obligation to protect into a consumer AI tool simply because asking the AI would be convenient. If your employer provides an approved AI system, follow its policies about what information can be used.

And remember other people’s privacy. A piece of information may be yours to know without being yours to share.  We all hate gossip even when it comes in the electronic form.

It is also worth remembering that what you type into an AI system should not automatically be treated as legally privileged or beyond the reach of litigation. Depending on the circumstances, digital records may potentially become relevant in a lawsuit or other legal proceeding.

Know Which AI You’re Using

There is an important caveat here: AI systems aren’t all the same. Different products and account types have different policies governing how information is stored, retained and used. Privacy settings matter. Before using AI for particularly sensitive work, find out what kind of system you’re using and what happens to the information you give it.

You can start by asking the chatbot to explain its privacy and data controls, but don’t stop there. For anything important, follow the links to the company’s actual privacy policy and settings so you can verify the details yourself.

Ashley’s Five-Second Privacy Check

So, before you press Enter, try what I call Ashley’s Five-Second Privacy Check:

Does the AI actually need to know this?

If not, leave it out.

Please help us continue the community conversation and email me your AI questions at [email protected]. I try to stay away from political questions, but I’ll do my best to answer the others!

This column was written in part with the assistance of artificial intelligence and reviewed, edited and approved by the human author.