I work in AI – here are the things I’d never use it for

Earlier this month, Jacob Coxon, an AI worker, resigned from Anthropic and declared on X: “The people building AI earnestly believe that it could kill us all by the end of the decade.” And Andy Burnham warned world leaders at the UN this week that “fundamental decisions on this technology are too important to be left to chance“. It’s easy to be drawn into the hype that artificial intelligence is going to destroy us all, and tech is often built faster than all the ethical knots it presents are worked out. But AI isn’t going anywhere, and plenty of us are already using it. To find out how we can make our own personal interactions with these tools safer and more reliable, The i Paper spoke to Lauren Ingram, a former Meta and Amazon marketer, who now runs AI Girl Gang, a community and podcast designed to get more women into AI. Don’t share anything you wouldn’t want elsewhere I think we will come to regret how much of ourselves we’ve given over to these technologies. I used to use ChatGPT to untangle relationships because it was this exciting, shiny and new tool, but I’ve become more self-conscious. It’s the Wild West out there – there are so many new AI tools coming out every day. So it’s hard to keep on top of which company has what kind of data policy or what’s OK to feed into these tools. My rule of thumb is I wouldn’t share information there that I wouldn’t be happy to share elsewhere online, because the more you’re sharing digitally, the less control you have over your information. That includes your children I personally wouldn’t share photos of my child with an AI chatbot, in the same way I don’t advocate for putting lots of photos of your children on social media. Firstly, they won’t have consented, and there are a lot of weirdos out there who have access to the same cheap AI tools as everybody else. You can personalise gifts using small AI companies, but I trust these even less than bigger ones. You don’t know what their data controls are, whereas bigger companies are under more pressure to provide safe AI, so tend to have a really clear privacy policy. Shorts I do – sometimes – put my photo into AI tools I resisted putting photos of myself in for a long time – really, how much do companies need to know about us as individuals? But AI can be helpful in preparing promotional images. However, I used Google’s Nano Banana because I wanted a photo to look like it was shot in a studio, and it made a couple of tweaks to my eyes to make me more generically pretty and made me a lot slimmer. AI is helping create unnecessary beauty standards instead of helping us fit them – I’m slightly dreading what’s coming next on that, because it’s so easy to kind of tweak and beautify images. If you’re not paying for it, you’re the product If you’re on the free version of an AI tool, by default, the things you type in may be used to improve future versions of AI. It doesn’t mean AI is literally memorising your conversation and repeating it to someone else – it’s more that your chats become part of the material used to make the whole system better. ChatGPT, Gemini and Claude are all made smarter every day through real-world examples of work people have done. You can turn that off, usually in the “settings” section. I have it off so it doesn’t use too much of my information. An exercise I did recently was to ask my AI what it knows about my family and my circumstances. It actually knew quite a lot. You can then ask it to delete that memory. It may not be off of servers for ever, but it means it won’t be used day-to-day. Assume your boss can see everything on your work computer If you’re planning to put more confidential information into AI, especially in a work context, it’s good to have the paid version of whichever AI tool, because it gives you better data controls. And if your company is paying it, you should be cautious with what you put in. In the same way I wouldn’t use my personal emails on a work computer, I think you should always assume that bosses can ultimately see things on your computer if they want to. I use AI, then always fact-check As a rule of thumb, AI gets things right about 70 per cent of the time, but you never know which 30 per cent is a little bit off, because when AI is wrong, it’s confidently wrong. It gets so many things right that you can be lulled into a false sense of security. So I always fact-check anything that’s got a name, a number or a date in it. And I’ll say, “Are you sure about that?” or “Is that definitely the case?” or “Is that information recent, and is it reliable?” Expect to be recorded I don’t think there’s a good established etiquette around recording work meetings, but if I’m speaking face-to-face with someone, I’ll ask, “Is it OK if I transcribe this using AI?” But sometimes online AI tools will dial in to the call, and that can feel more invasive than me just recording. I always assume any kind of online work call is being recorded and transcribed by someone’s AI. I’m freelance and I sometimes feed in transcripts of calls to try and understand someone’s prerogative in a situation, like when money is being discussed in a business meeting. I’ll ask, “Why does this company want this? What’s their skin in the game? Are they low-balling me?” I always bear in mind that AI won’t give a correct answer, but it can give a good answer. I have a safe word with my family to avoid scams AI makes it very easy to clone someone’s voice. The public needs to be aware that this will become a common scam, and I feel hyperaware that, as a podcaster, my voice is out there. So I’ve told my family that if I ever ring them saying I’ve been in a car crash and they need to send money to help, to put their emotions aside for a second and ask themselves: would I really need money in a rush? We have a family safe word agreed in advance to avoid getting scammed in those situations. Play with AI Recent headlines are so negative and extreme, and that definitely boosts the value of companies, but it just makes people scared, which helps no one. We don’t always need to be frightened of it. You can give it some really simple tasks to begin with, like if your goal is to get promoted by the end of the year, ask it, “What are the blind spots I may not have thought about when trying to get promoted?” I’ve got a tool called Town and it is like a human executive assistant that goes into my emails and calendar to pull out what I want to pay attention to, or where I’m at risk of dropping the ball. AI can be used to problem-solve I have a double-glazed backdoor and I wanted to put a cat flap in there. Normally I’d just put it off, but then I took a photo of the door and asked AI, “How would you put a cat flap in this?” It said, “Whatever you do, do not try to do this yourself, because you’ll be breaking the vacuum of the double glazing”. It told me what a glazier could do to help. How much googling of random websites would it have taken for me to find that information out? It’s a bit like having an extra dad to help you get started on a problem.