By Lauren Donley
A few weeks ago, my better half was tinkering on ChatGPT and asked it to replicate one of my recent columns for the Westsider. In an instant, my smugness gave way to impending doom as an article flashed up on the screen, very similar in structure to the one that had taken me hours to write.
My thoughts turned to the inevitable reality of being replaced by robots. But then I pulled myself together and began to ponder how I could use ChatGPT to work smarter, not harder. I mean, would anyone know if I outsourced this column to artificial intelligence…?
That’s a question for another day but in the interim, please enjoy these words and science tidbits that were painstakingly constructed by my human brain!
What the tech?
As Christmas tree season kicked off last year, Santa wasn’t the only phenomenon about to sweep the globe. ChatGPT, short for Chat Generative Pre-trained Transformer, was released in November and reached over 100 million users in just two months, comfortably outpacing the early growth of TikTok, which took nine months to accumulate the same number of users.
ChatGPT is the robot-brainchild of OpenAI, a San Francisco-based artificial intelligence (AI) company on a mission to create safe products that ‘benefit all of humanity.’ The company has been diligently working away on AI technology since 2015, supported by an early cash injection from Elon Musk.
At its core, ChatGPT is an AI chatbot that can be trained to respond to questions in a surprisingly human way. For example, here’s the opening paragraph of a 40th birthday rap it composed for me:
“Yo, listen up, it’s time to get ready; we’re throwing a bash, and it’s gonna be steady.
Come on down, don’t be late to the scene; ‘cause it’s a celebration, a big 4-0, you know what I mean?!”
Thankfully, the utility of ChatGPT extends well beyond spitting dope birthday rhymes. In theory, it can be used to compose any type of written content, including essays, social media posts, press releases, movie scripts and even computer code. But certainly not this newspaper article.
What the health?
Eight months on from its release, scientific research into ChatGPT has begun to trickle through, and the results are both fascinating and somewhat alarming.
In one study, ChatGPT answered open-ended and multiple-choice questions well enough to pass the three-step United States Medical Licensing Exam (USMLE) – a national standardised test that spans basic science, clinical reasoning, medical management, and bioethics.
Other studies have shown that when given simple instructions, ChatGPT can produce short medical reports and patient discharge summaries. Some errors were detected, but doctors generally agreed that the medical reports were factually correct, complete, and not harmful to patients.
And perhaps most notably for any writers among us, a study published in April found that when a team of scientists was given a stack of 50 abstracts – half written by researchers and the other half generated by ChatGPT – on a third of occasions the scientists were unable to spot the ‘fake’ abstracts. They also expressed surprise at how difficult it was to tell the two types of writing apart.
So, there you have it – maybe you wouldn’t have been able to tell after all. Until next month!
