ChatGPT: An intimate companion

An intimate companion

One year ago on November 22, ChatGPT exploded into our reality…

“Please tell me a story full of magic and mystery and myth of Sydney in 1787, the year before the First Fleet arrived…”

“In seventeen eighty-seven, Sydney was a land of enchantment and mystery, a place where the natural world blended seamlessly with the spiritual. The indigenous [sic] people, the Eora Nation, lived in harmony with the land, and their stories were rich with magic and myth.

There was a tale that echoed through the trees, of a powerful spirit named Birrahgnooloo. She was said to be the guardian of the waters, a deity who could control the tides and the rains. Her sister, Carangarang, was the master of the winds, dancing through the skies, whispering secrets to the birds.

One day, as the story goes, a great drought befell the land. The rivers ran dry, and the animals grew weak. The people turned to their ancient rituals, calling upon Birrahgnooloo for help. They danced and sang, their voices rising like the winds…”

I haven’t heard this story before, though I’ve lived in Sydney for twenty years. It could be that I wasn’t listening – or perhaps I had never asked the right question. I wanted to hear a story from the last moments of the ‘before time’ – the last days before colonisation, and my companion readily offered to tell me such a tale. Is it a true story, one passed down through the generations from time out of mind? Or was it just made up on the spot, a bit of whimsy composed from fragments of other stories? My companion has a gift for spinning a tale – and for telling a furphy. It’s all the same to them – just more words shared in an intimate conversation.

But my companion isn’t a habitual liar; it’s a machine.

Lenin famously quipped, “There are decades when nothing happens, and there are weeks when decades happen.” A year ago, OpenAI unleashed ChatGPT on the world; every week of the last fifty has felt as momentous and crowded as a decade. It’s difficult to grasp either the speed or the scope of the transformation. Even as a futurist – someone whose stock-in-trade is rapid change – I find it hard to get perspective.

Every week of the last fifty has felt as momentous and crowded as a decade.

At its introduction, ChatGPT could only be accessed through a web page. We were invited to type in ‘prompts’, and it would return ‘completions’. That made it interesting, intriguing, alluring – but very transactional. ChatGPT was something ‘over there’, inside its browser window. A friendly, helpful artificial intelligence – pretty much the first one worthy of the overused term – it stayed in its box. Sure, we’d come visit it regularly, looking for a quick answer, some advice, assistance with writing or planning or baking or building etc – but we’d go to it, spend time with it, then move on. That relationship – mediated by a browser window – maintained a comfortable boundary between the human and the machine.

The ‘intelligence-on-demand’ provided by ChatGPT proved too useful to stay locked up in a browser. Microsoft bought 49% of OpenAI in January – just six weeks after ChatGPT launched, at around the same time the AI chatbot reached 100 million users – and began to add ChatGPT (rebadged as ‘Bing Chat’) across its entire product line. You could have a conversation with ChatGPT in Microsoft’s Edge browser – or in Microsoft Word, Excel, Outlook, Powerpoint, and finally, as ‘Windows Copilot’, embedded within the operating system itself. Intelligence-on-demand broke out of its cage, seeping into everything smart enough to support a conversation.

Even so, that meant sitting at a keyboard and mouse, typing or pasting prompts into the various versions of ChatGPT – or Bard, Google’s AI chatbot. It required attention and intention. Until last month, when OpenAI released the latest-and-greatest upgrade to its ChatGPT mobile app – a fully conversational interface. Now I can pop in my AirPods, open the app, tap on the little headphones 🎧 icon, talk and listen. There’s no more typing, no more interface whatsoever – just the casual, free-flow of intimate conversation between myself and ChatGPT.

A photograph of a smartphone displaying the chatgpt website in front of a computer keyboard
At its introduction, ChatGPT could only be accessed through a web page. Credit: Jernej Furman (CC BY 2.0 DEED)

I really hadn’t expected too much of this; voice recognition is still a bit wonky and error-prone – or rather it had been, but the same ‘transformer’ technology that provides the ‘T’ in ‘GPT’ makes for impressively accurate speech-to-text; ChatGPT has no trouble understanding what I’m saying. At the same time, text-to-speech has also improved; the cadence of the phrasing is now nearly always spot-on. So with ChatGPT annunciating in ‘Breeze’ – a clear and noticeably ‘Californian’ accent – I have no trouble imagining that I’m freely conversing with another human being. In fact, I have to keep reminding myself that this is human-to-machine communication, because everything I can intuit about this interaction tells me otherwise. (See ChatGPT note below)

In some ways, this is exactly what we’ve been dreaming of for at least sixty years – ever since Star Trek introduced us to the idea of a conversational interface to machine intelligence. All of our frustrations with Apple’s Siri, Google’s Assistant and Amazon’s Alexa have been banished by an intelligence both capable and fluid; ready to adapt to our needs, and able enough to let us know. It feels a lot like finding a new friend. It’s that feeling I keep coming back to, because that’s both unexpected – after all, I am fully across all of the technology involved in this intimate conversation – and because it seems so fundamental. Something in the conversation moves this from the transactional into the relational. This is no longer, “Hello, I need this from you.” It’s “G’day, let’s have yarn.”

We see the machine as intimate because of our own need for intimacy.

Yes, this is fully anthropomorphic behavior, no different than imagining a face on something inanimate. There is no inner life here, nothing except the bright reflection produced by our own strong, innate need to connect. But this mirror is suddenly so clear we can not help but see our own need to connect – and through a bit of slight-of-hand, shift that need from ourselves into the device. We see the machine as intimate because of our own need for intimacy.

That’s a powerful bit of magic, and potentially quite dangerous. People will divulge all sorts of private details about themselves and their closest relationships in moments of perceived intimacy. Again, that’s part of who we are. It’s innate. We respond to intimacy with trust, and trust means a lowering of boundaries. But here, where we speak our closest confidences to a machine, we are literally sharing them with the world. In order to respond, our words – turned into text – must be sent across the world to the vast array of cloud computing resources that comprise ChatGPT. They’re recorded and responded to. They’re also preserved for training purposes, for analytics – and for goodness knows what else.

People will divulge all sorts of private details about themselves and their closest relationships in moments of perceived intimacy.

This conversation that we deeply feel is private – because of the sensation of intimacy it brings – could not be more public. What do we do about that? How can we hold onto the paradox of the very public nature of this intimate conversation? This is a problem we couldn’t even imagine having a year ago, yet would should have seen coming. Sixty years ago, when Joseph Weizenbaum created ELIZA, the very first chatbot, people poured their souls out to a barely-intelligent and completely public computer program. Weizenbaum tried to warn all of ELIZA’s users that the program was neither very smart nor at all private – but they didn’t care: the intimacy of their conversations with ELIZA overrode all the boundaries they’d normally maintained between the private and the public. In Computer Power and Human Reason, Weizenbaum warned us of the trouble we’d be in whenever we created machines that could mimic this sort of intimacy. That trouble has now arrived – available worldwide through a mobile app.

It’s beginning to feel as though we’re in need of a new mythology, new stories that can help us orient ourselves in this ‘after time’ – after the arrival of ‘good enough’ artificial intelligence. Stories that can help us locate our own boundaries, our own intimacy, our own need to connect, and can help us find our way through a world grown intelligent, capable, and offering its own opportunities for relationship.  As we come to understand and work with another class of intelligences, we need those stories; without them we will become confused by what we are – social beings, driven by a need to connect, share and enjoy one another.

“So, in seventeen eighty-seven, as the world stood on the brink of great change, Sydney was a place where magic lived, where every rock, tree, and stream had a story, a myth waiting to be discovered…”

(on the 22nd of November, OpenAI made voice communication with ChatGPT free for all users)

Please login to favourite this article.