by Larry Magid
This post first appeared in the Mercury News
There is a lot of angst about AI, including the ways it can invade our privacy. Almost every AI safety article, including ones I’ve written, warns users about the risks of sharing personal information with an AI chatbot.
Most major AI companies have privacy policies saying they don’t sell personal information, but that doesn’t mean everything you share is entirely private. Depending on the service and your privacy settings, companies may use your conversations to train their AI models, and in some cases, human reviewers may have access to what you share. There is also a risk, however remote, that personal information used in training could turn up in responses to other users. And, as with any stored digital information, there is always the possibility of a leak or data breach.
Protecting your privacy
If you’re concerned about privacy, there are things to consider when using an AI chatbot. One is training. Can the company use your conversations to improve its models, and can you opt out? Another is memory. Can the chatbot remember personal details across conversations, and can you turn that off? How long does the company retain your conversations, and can you delete them? Can you have conversations that aren’t saved to your regular history or used for training?
You can ask your AI service these questions, but be sure to request links to the company’s official privacy policies and verify the answers before relying on them.
The good news is that major AI services give you some control over your privacy settings. The instructions for adjusting these settings are too long for this column, so I’ve created a short guide, broken down by service, which you can find at connectsafely.org/ai-privacy-settings/.
Weighing the trade-offs
The more an AI chatbot knows about you, the more helpful it can potentially become. But that convenience comes at a cost. You’re trusting a company with information you might not otherwise share, and even with privacy protections in place, there is no guarantee that your information will remain completely private.
Before deciding to lock down your privacy settings so AI remembers nothing about you, consider the advantages of allowing it to remember what you’ve shared. How much to share is a personal decision that requires weighing the benefits against the risks.
The benefits of sharing
For example, someone who shares their medical records, medications and health conditions with an AI chatbot can use it to better understand their health and prepare questions for their healthcare providers. Knowing which medications a person takes can also help an AI system flag possible drug interactions or suggest whether certain symptoms might be medication-related. Of course, AI can make mistakes, and even if it has your medical records, it may not have the whole story, so any medical information should be verified with a health professional before making decisions.
The same applies to other aspects of life. Sharing information about investments, retirement income and expenses can help an AI chatbot provide more personalized answers to financial questions. Some services, including ChatGPT, can connect to financial institutions to access information about investments, bank balances, loans and other accounts. This can help you understand not just which mutual funds or ETFs you own, but what stocks and other securities those funds actually hold. For example, you might discover that several funds in your retirement account own many of the same stocks, leaving you less diversified than you thought. AI can help uncover these overlaps and give you a clearer picture of where your money is invested.
Telling it what technology you own can make troubleshooting easier because you don’t have to explain your equipment every time you ask for help. And sharing information about your professional work, interests and preferences can make AI responses more relevant and useful.
I often use AI to recommend recipes and sometimes give it prompts like “keep it simple.” But you can also share your family’s food preferences, dietary restrictions and allergies, or what you have on hand, to help it come up with recipes that will work for you.
Finding information
You can also connect some AI services to your calendar and email accounts, which can be incredibly handy when you’re looking for information about upcoming travel or trying to find records from years ago. While email and calendar services have their own search functions, AI lets you ask much more detailed questions in everyday language, such as “Find the times when Sally and I went to dinner” or “What was the name of that restaurant where I ate in Los Angeles a few years ago?” Instead of searching through countless emails and calendar entries, you can simply ask a question and let AI look for the answer.
This can be especially helpful during tax time if you need to unearth deductible expenses. For example, most e-merchants send order confirmations with the item, date of purchase and price. While you could find this information by searching, AI can produce a spreadsheet with much of what you’ll need at tax time. If you give it access to your email, calendar, credit cards and other accounts, it can check multiple sources to give you a more complete picture.
As always, carefully check the information. Even though AI often gets things right, I have found serious mistakes and omissions.
This incredible technology can be very powerful and extremely useful, but like almost everything in life, it comes with risks. Risk management is something you already do every day. You know that walking is, for most people, good for you, but you also know there’s a risk of falling. Kitchen knives and stoves are essential if you want to cook, but you also know how to avoid getting cut or burned, even though you’ve probably experienced both at some point in your life. The same is true with AI. The goal shouldn’t be to avoid it because of the risks, but to understand those risks, take reasonable precautions and enjoy the benefits.