I'm Crap at Responding to Texts. So I Got ChatGPT to Do It.

If, by now, you’re still bothering to keep track of all the wonder inventions that are going to drag us out of shitty reality and into a utopian future, thanks for playing and I’m sure you’re very much already well acquainted with ChatGPT.

For the uninitiated, ChatGPT is an artificial intelligence software developed by OpenAI capable of doing everything from writing a shopping list for you to writing complex, video game building code. There’s a lot of hype around it. Some of it about right – the language model’s ability to do jobs in mere seconds is astounding – but there’s also been no shortage of overconfident tech-fluencers preaching about how it will change the workplace forever.

Sure, AI is great at making certain jobs easier, but what if I could use it to make my personal life easier? Earlier this year, a data scientist trained an AI language model to clone his group chat, training it on over 500,000 messages to understand his friend’s personalities and writing styles. So can AI alleviate WhatsApp anxiety for someone who is not a data scientist? And more importantly, how can it help me with the hundreds of unread notifications currently sitting on the top right hand side of my WhatsApp icon?

First things first: I should speak to someone who knows their shit, so I ring up James Maddison, a prompt engineer and founder of AI-powered marketing and branding agency Prompt, for some advice. 

“Basically, what you’d need to do is create a digital clone of yourself and I would recommend ChatGPT-4 as your starting point,” he says, suggesting the latest version of OpenAI’s language model. Twenty dollars and one expense request later, I’m ready to go.

What now? “Give as much data about you as possible,” Maddison tells me. Essentially, it’s time to get ChatGPT to know me better. To do that, I need to hand over as much of my personal data as possible. Despite having watched and vaguely understood Netflix’s 2019 documentary The Great Hack, I come to terms with handing over the keys to my brain in the name of great content. Fortunately, as Maddison reminds me, I can just send ChatGPT all my previous VICE articles!

“What you need to do,” he says, “is tell ChatGPT, ‘I want you to learn my style of writing, here are all my articles. I'll keep on sending you them and you'll keep on learning it and then I want you to craft a response in my persona’ and it should spit something out.”

Unfortunately, I can’t link an article directly without a plug-in. Instead, I copy and paste the articles into ChatGPT with Maddison’s prompt: “I wrote this, this is my writing style”. First, I choose an article that I wrote on dating in the metaverse – what better way to kick things off than to send ChatGPT something I wrote about its good friend, Web3?

“Thank you for sharing the excerpt from your first article on dating in the metaverse,” ChatGPT tells me. “It provides an interesting insight into your writing style and tone, as well as the topics you like to explore. I can see that you have a conversational and engaging writing style, with a touch of humor and a willingness to dive into novel experiences. This will be helpful in understanding how you might communicate in personal messages.” Cool! Also, because I know this robot cannot feel - and the compliment is coming from a purely objective point of view - I’m now feeling validated as hell.

Impressed, I double down and send another piece. To mix it up a little bit, I send an article I wrote about how your flatmates might be fucking up your credit score.

“Your writing style in this piece is informative, clear, and engaging, with a touch of humor to keep readers interested,” ChatGPT enthuses. “You have a knack for breaking down complex topics into understandable terms and presenting useful tips.” Probably the nicest compliment I’ve ever had on my writing – ChatGPT, you old flirt!

Now time to really lock in my personality with a third and final article: A list of all the awful shit Boris Johnson has said over his years. I copy and paste the piece into the chat box and hit enter.

ChatGPT replies: “Introduction: It's not just one off-the-cuff comment that's come back to haunt Boris Johnson” and proceeds to summarise the article. Ah shit, it’s forgotten everything we talked about and I have to start from square one. We’ve hit our first roadblock in creating my digital clone.

I’m starting to think that on my journey to alleviate the pressures of my social life, I may have accidentally hired an incompetent intern which I now have to spoon feed before getting them to reply to messages on my behalf. But I hold my nerve and repeat the process.

This time, ChatGPT remembers its task to create a digital clone of me and even apologises for the confusion! If our future robot overlords ever read this article, please note: I was very understanding.

Now the AI had a decent understanding of my writing style, it was time to hand over a bit more personal information. “Ask me as many questions as you need to better create a digital version of Ross Buchanan,” I order ChatGPT, “in order to create realistic responses to his personal messages on my behalf.”

The language model then asked me ten surface-level personal questions, including “what’s your favourite movie?” and “do you have a favourite sports team?”. It’s giving Single Dad Sets Up An Internet Dating Profile in a Noughties Sitcom, but I indulge them regardless.

“Thank you for sharing more about yourself, Ross!” ChatGPT responds, “With this information, I'll be better equipped to create a digital version of you and generate realistic responses on your behalf.” My clone is complete!

Now to put it to work. Unfortunately for me – but probably for the benefit of humanity – ChatGPT cannot yet gain access to my WhatsApp and carte blanche reply to all my messages, so I have to paste the messages I receive into the chat box for it to craft a reply.

First up, it was my best mate Pete’s birthday yesterday. Because I’m a shitty friend whose only way of remembering important events was Facebook (which I deleted in 2017), I haven’t messaged him yet. The anxiety has been building and the longer I leave it, the worse it gets. He’s just messaged me asking to hang out on Tuesday and I need to come up with a plan, but also come up with some sort of excuse that alleviates me of any responsibility.

Time to get ChatGPT to craft me the ultimate belated birthday message! It could probably do with an edit (I’m definitely not that generous with my emoji usage) but in the interest of efficiency, I paste it into my WhatsApp chat with Pete.

Safe to say that didn’t exactly go to plan!

Next up, my family WhatsApp chat has been popping off over the last few days but I’ve just been too busy to reply. For context, the group is mainly for my parents who want to be kept in the loop with what my nieces and nephews are up to. 

In the interest of the privacy of my family, I’m not going to copy and paste any of their messages into ChatGPT – I just need to assure everyone in the chat that I’m basically still alive.

Again, even though the response has zero rizz and way too many emojis, I copy and paste into the group chat without any editing. I’m annoyed that ChatGPT has made a promise on my behalf to “be more active in the group chat soon” but I’ll cross that bridge when I get to it.

“You alright, mate? Are you being held hostage or something?” my brother privately messages me immediately after I hit send. I feel as though, once again, ChatGPT may have dropped the ball on this one.

Finally, and with all hope nearly gone, my friend Doug has invited me to the screening of a movie tomorrow night. I’ve never heard of the director so I need to politely decline without coming across like a totally ungrateful and uncultured swine.

Holy shit. That answer actually doesn’t suck! It’s not quite me but I’m sure this time, we might be onto something. Holding my breath, I copy and paste it into our chat and wait for a response.

I can’t believe it actually worked! I’m pretty sure that my response was so weird that Doug will never ask me to hang out ever again but it’s going down as a win.

But even though it’s definitely possible to dupe your mates into having a conversation with a robot, there’s an uncanny valley element at play: The more realistic the conversation gets, the more people are aware that something just feels a little… off. There’s also so much heavy lifting involved in creating a digital version of yourself that you’re almost definitely not going to save any time or social labour. In short: If you’re feeling anxiety over being MIA in the groupchat, AI is probably not your friend.