Implicit in the phrase “The pen is mightier than the sword” is the idea that to persuade people with your writing takes a degree of skill. That a random jumble of words needs the energy of somebody who knows what they are doing to create order and sense.
But what if it’s machines that unjumble those words?
Despite what years of movies have told me about the Terminator or HAL, I refuse to believe that when computers become intelligent enough, they will turn evil and take over the world. What is more likely is that people will be dumb enough to believe computers and end up doing stupid things that destroy the world.
I had a chat with someone who was appalled at how an artificial intelligence (AI) research tool returned seemingly promising but ultimately fake URLs. “It’s so dangerous!” she exclaimed.
It’s true. Just ask Steven Schwartz, an American lawyer who has been called up by the court to explain how he used an AI tool to assist in writing court filings that cited six nonexistent cases. When he was asked to explain whether these cases were authentic, he said yes after referring back to the same AI tool for confirmation.
To me, the obvious issue isn’t that the AI tool spouted nonsense; it was that the person using it couldn’t tell the difference between good output and bad output. We need to learn to be able to tell the difference.
A professor in an American university set his undergraduate class the following assignment: Students were asked to use AI to generate essays using a given prompt, and then critically evaluate the output. In particular, they were told to fact check the information and critique the analysis.
Every one of his 63 students discovered that their generated essay contained some form of inaccuracy, and half of them said they had been unaware that AI could be wrong. Upon reflection, the professor’s students themselves pointed out the dangers of overrelying on AI to “get things right”. One student said: “I’m not worried about AI getting to where we are now. I’m much more worried about the possibility of us reverting to where AI is.”
Of course, the clever people making these AI tools have been trying to find ways to make sure the next iteration doesn’t make these kinds of mistakes, or at least highlights where it might have gone wrong.
For example, the folks at OpenAI have introduced a version of ChatGPT that can also browse the Internet to get the information it needs. If you ask it to write about something, it will annotate the resulting text to show you which website it found the information on.
Of course, you have to then read that original webpage source and try to determine if it is accurate and true. Although AI can help you draft the article and find the source webpages, you still ultimately have to do the fact checking.
It was pointed out that this was reminiscent of Wikipedia when it first became popular. We were told we shouldn’t use it because we don’t know how much of what was on it was true. Anybody could contribute to Wikipedia, and nobody knows whether the facts are reliable or not.
Over time, instead of becoming unreliable as a source of information, Wikipedia became a convenient port of first call for a quick overview. However, we learned that for critical portions, it really was better to find alternative sources to back up what was found in Wikipedia.
In the same way Wikipedia is not a source of truth, but a tool to help corral facts, these new AI writing tools are not about writing the truth, but about how to present facts it thinks it knows in a very human-like way.
If it’s so good at writing readable first drafts, then why don’t we use it to write stories? Well, Amir Muhammed, the founder of local publishing company Buku Fixi, is open to the idea, stating, “The kind of stories that do well are the ones that get positive word of mouth – by human readers. If something by an AI manages to achieve that, we won’t stand in the way of reader enjoyment.”
In fact, the main concern for book publishers seems to be the issue of authenticity, in this case of human intellectual property. Can an author claim to be the creator if he uses a mishmash of words generated by another? If yes, then if those “other words” he uses are found to be very similar to some already existing piece of work, can the author be accused of plagiarism?
When it comes to writing, the notion of “truth” carries different implications for legal submissions, academic essays, and works of fiction. For legal submissions and university essays, the concern is whether they are consistent with verifiable facts. For pieces of fiction, it’s whether the words evoke the writer’s intended meanings and emotions beyond a simple “cut and paste”.
In all cases, it should be obvious that the power lies not only with the pen and the person wielding it, but also the reader and how prudent we are in accepting those words.
In his fortnightly column, Contradictheory, mathematician-turned-scriptwriter Dzof Azmi explores the theory that logic is the antithesis of emotion but people need both to make sense of life’s vagaries and contradictions. Write to Dzof at lifestyle@thestar.com.my. The views expressed here are entirely the writer's own.
Already a subscriber? Log in
Get 20% OFF The Star Digital Access
Cancel anytime. Ad-free. Unlimited access with perks.
