I HAD this little trick I used when I was studying German. For the final exam, I knew we had to write a short letter. So I memorised an opening paragraph and a closing paragraph, and then all I had to worry about was the words in between.
These days, with modern technology, you don’t even have to memorise anything. In the same way the Internet has almost made print dictionaries and encyclopaedia obsolete, artificial intelligence (AI) is rapidly taking over the role of “good enough” reasoning.
Perhaps, despite the drop in Malaysia’s Pisa (Programme for International Student Assessment) scores, we have less to worry about than we think. Sure, according to the latest results, nearly two-thirds of Malaysian students cannot apply mathematics in everyday situations, which might seem like grounds for concern.
But the Pisa report also says that more than half of Malaysian students surveyed use AI weekly to help them learn. Malaysian students also do much better at using computers to solve problems than would be expected, given their performance in mathematics, science, and reading.
If our students are adept at using computers in their day-to-day tasks, that should stand them in good stead in a future when computers and AI are pervasive. A report by consulting firm Ernst & Young says that 93% of Malaysian employees are using generative AI to save time and improve performance, but only 12% currently receive sufficient training.
“But wait,” you may protest. This is fine if what we want is a generation that knows how to punch numbers into a machine. However, doing the hard thinking required to find real solutions to complex problems is something completely different.
But is it?
Recently, ChatGPT creator OpenAI announced that it had found a singularity in the three-dimensional Navier-Stokes equations. If you found that difficult to understand, perhaps this puts it into context: If the solution is verified, it could claim one of the six remaining US$1mil (RM4mil) Millennium Prize Problems.
The Problems were announced in 2000 as seven of the most difficult and fundamental unsolved problems in mathematics. One of them, the Poincare Conjecture, was solved in 2002-2003 by Russian mathematician Grigori Perelman, who built on the framework developed by American mathematician Richard S. Hamilton in the 1980s.
Perelman is considered an eccentric and has been described as someone who "looks like Rasputin, with long hair and fingernails" and spends his free time in the woods around St Petersburg. True to his reputation, in 2006 he rejected the Fields Medal (the equivalent to a Nobel prize in mathematics) and then the million-dollar Millennium Prize in 2010. When a reporter tried to ask him about it, he reportedly responded: "You are disturbing me. I am picking mushrooms."
All this is in stark contrast to the way OpenAI presented its solution on Sept 8. It was generated by approximately 10,000 autonomous AI agents working together for 88 hours, then verified by a separate AI model that spent another 17 hours formalising the result.
Then it was revealed that OpenAI may have rushed to solve the problem after hearing rumours that human mathematicians Tristan Buckmaster and Levent Alpoge in the United States were making significant progress on the related equations. To top it off, Buckmaster had been using OpenAI tools to develop his ideas and has alleged that OpenAI had used his prompts and chat sessions as part of the learning set for its own computer proof.
This has set off a debate in the mathematics community about the role of AI in solving mathematical problems.
Some have accused AI companies of claiming success based on the work of human mathematicians who have invested years in their research, with computers merely climbing the last few feet to the summit before claiming the credit. Others say it serves mathematicians right to finally experience what it feels like to have their jobs taken away by AI.
Personally, I pay particular attention to what Australian-American mathematician Terence Tao has to say. He is a Fields Medal winner and versatile polymath who openly championed the use of AI in mathematics when others were still unsure.
As early as 2024, he was advocating human-machine collaboration to push forward the frontier of mathematics. Yet his voice has also been among the loudest in advising caution about how AI works.
He says people might think what they want is simply the answer to a maths problem. But what matters is what you learn from studying the problem as a whole. “AI tools are like taking a helicopter to drop you off at the site. You miss all the benefits of the journey itself.”
Tao has been particularly critical of how AI companies race to the finish line for PR purposes, only to “dump carcasses of raw meat onto our communal village table” and leave the hard work of understanding and verifying the results to mere mortals.
I completely agree that if you write a mathematics paper proving something new, you should be able to explain it without overly relying on the words, “because the computer says so”.
This problem is so pervasive that Thomas Dietterich, the editor-in-chief of arXiv (pronounced “archive”, it’s a peer reviewed research-sharing platform open to anyone), has confirmed that he is seeing a trend of “authors submitting papers whose contents they likely do not understand” and is proposing ways of managing such AI slop.
While the opening and closing paragraphs of my German letters were not AI-generated, they were most definitely slop that I ladled onto my writing. Every sentiment I heralded in them had been prepared in advance and had absolutely no bearing on what I really felt or thought at the time. After all, it was just an answer for an exam that I had to get right.
This is precisely the problem with getting the answer without knowing how you got there. By hopping straight to the destination, you don’t know how to adjust the route, improvise a different path, or have the chance get distracted by some interesting fauna in the woods.
Being good at using AI and being good at reading, mathematics, and science are not mutually exclusive. The Pisa report notes that both Singapore and South Korea have implemented comprehensive policies to foster digital learning without sacrificing fundamental skills.
There are doomsayers who believe that AI heralds the beginning of the end of the world as we know it. That the human race will simply abdicate its responsibilities to the silicon overlords.
But if the point of being human is to persevere, learn, and improve ourselves for the generations to come, then we should recognise that and take responsibility for doing so – even if it means choosing the slower, less efficient journey.
