Opinion: Improvements in generative AI are coming thick and fast. OpenAI’s GPT-4, was released last month, described by many as a significant improvement on GPT-3.5 powering the November 2022 release of ChatGPT. ChatGPT is also gaining a staged access to the live internet. Up until recently, its answers were limited by the cut-off date of its training data – September 2021.
Those with long memories may remember The Lawnmower Man, a 1992 sci-fi horror film in which a malevolent digital intelligence gains access to a primitive dial-up internet, signalling its escape by ringing all the world’s telephones.
The pace of these improvements recently led “1,100+ notable signatories” to endorse an open letter calling on “all AI labs to immediately pause for at least six months the training of AI systems more powerful than GPT-4”. If the labs don’t quickly comply “governments should step in and institute a moratorium”.
The Musk distraction
It says something about our obsession with celebrity that this call for a pause was turned by many media outlets into a story about Elon Musk. After attaching his name to the list of signatories, the story for the New York Times became “Elon Musk and Others Call for Pause on AI, Citing ‘Profound Risks to Society’”.
Musk has the inscrutability brought by the glare of extreme publicity. Can we really be sure why he paid USD$44 billion for Twitter, a loss-making social media company? One reason Musk may be calling for a moratorium is that he is not a current financial backer of OpenAI. His main complaint about ChatGPT up until now has been about its status as “woke AI”. For him, the moratorium may be about giving a non-woke competitor time to catch up.
The challenge to creative work
The real challenge lies not in AI’s wokeness, but in its impact on jobs. The open letter asks: “Should we automate away all the jobs, including the fulfilling ones?” Many of these fulfilling jobs fall into the category of creative work. They include graphic design artists, journalists, scholars, poets – people who formerly considered their work safe from automation.
Generative AIs won’t stop us from drawing pictures, penning poems, doing journalism, or writing philosophy articles. But how much will we be paid to do these fulfilling things in a world in which there is an alternative that is almost as good, and much, much cheaper. The exponential pattern of improvement characteristic of digital technologies exacerbates this problem. Creative workers should look at chess players who have seen digital technologies swiftly go from being sort-of-OK to being too good.
It’s tempting for creative workers to smugly list the failings of generative AI. ChatGPT is not creative. I recently asked it to write a review of the 1984 movie Terminator in the style of Jane Austen. “My dear readers, I must confess to being quite taken aback by the spectacle that was Terminator.” She, or ChatGPT impersonating her, didn’t enjoy it – “a rather crude and unrefined affair”.
The process that ChatGPT went through to produce that review bears very little resemblance to what Austen did when she wrote Pride and Prejudice. ChatGPT is basically a very powerful auto-complete engine. It was trained on 570GB of data from books, Wikipedia, and other writing from the internet.
If you ask a powerful computer to find patterns in that much text, you get much more interesting continuations than when your smartphone suggests “work” after you tap in “see you after …” The 570GB of data on which it was trained contained all of Austen’s writing and many discussions of it.
This means that ChatGPT has zero capacity to match Austen’s achievement as a writer. She was certainly not a generative AI who provided a market-friendly mash-up of 18th Century writing. Improvements of generative AI will probably offer more amusing mash-ups of Austen’s writing, but they won’t match her originality. The challenge for creative workers is to find a way to economically value this originality.
The future of creative work
The lesson from this review of Terminator is that creativity is easy to fake. I almost have enough there to start a review site “Jane Austen goes to the movies”. When ChatGPT gains full access to the internet it will offer amusing reviews of The Banshees of Inisherin, Avatar: The Way of Water, and Top Gun: Maverick.
It’s this originality and imagination that creative workers must double down on if we are to survive. If you fear losing your job to a fast-improving technology, then the pressure is on you to reconceive of what you do in a way that isn’t easy for machines.
My academic field of philosophy currently spends too much time training students to analyse academic articles and to write responses to them. The results can seem like ChatGPT’s mash-ups of Austen’s writing.
When ChatGPT’s successors get access to all the philosophical writing in paywalled academic journals, they will easily master this task. Truly novel solutions to problems such as the future of work and climate change require teachers of philosophy to value the originality of Austen.
Young minds come to university with a deep desire to address global challenges. We ill-serve students when we insist on training them to write and think in ways that GPT-6 will easily beat them at.