top of page

Iterations and the passage of time

13 hours ago
6 min read

Updated: 2 hours ago

In 1997, IBM’s Deepblue machine beat world champion Garry Kasparov in a game of chess[1].


The world was awe-shocked. It was the first ever defeat of a human by a computer. Kasparov accused IBM of cheating. Some said it was a publicity coup by the company. IBM later retired the machine.


The breakthrough made the headlines again nearly two decades later when Google’s Deepmind AlphaGo defeated human world champions in separate matches of Go. The feat also ignited strong reactions to how Deepmind had been secretly playing various players online, and training its database to maximize its chances of winning[2]. The subsequent live broadcast of AlphaGo vs Chinese champion Ke Jie had also been blocked in China[3]. 


So when the Navier-Stokes equations—a million dollar math problem—was solved by OpenAI recently, it was no surprise that it drew similar criticisms from various communities on the study of mathematics and the broader implication on humanity.


Computers have once again proven itself to outsmart humans, just in a different way and domain. 




Anything that can be reduced to a set of mathematical equations: rendering graphics and moving images on a screen, getting from home to the airport in the shortest time, or a game of chess. The list goes on... run enough iterations and it becomes ultimately solvable[4].


The paradox here is the comparison. 


To pit a human brain against a computer is like trying to outrun a speeding bullet. If the goal rests purely on speed, then there is no outcome whereby a human can reasonably win. Fortunately we live in a reality where the outcomes we care about are almost always never determined by perfecting a single parameter. 


No one feels threatened that a simple calculator can multiply two seven-digit numbers faster and more accurately than any human alive--because speed and accuracy at arithmetic were never really about being good with numbers.


What we actually value in a person who is good with numbers is judgment: that is knowing which calculation to run, spotting when an answer looks wrong, explaining the result to someone else, deciding what the resulting number means for decision-making.


The calculator wins the sprint every time—but it has never once been asked to do any of the things that actually matter around the arithmetic of it.


The “threat” here, so to speak, had always been personal.


The backlashes involving the solution to the Navier-Stokes problem occurred because boundaries had been crossed, interests were conflicted and livelihoods are at stake. If artificial intelligence hadn’t dipped its fingers in the lunchboxes of the mathematicians, the retaliation from the community might have been milder. 


Consider also the live translation captions being generated by AI in Zoom or other video conferencing tools. The feature doesn’t displace any jobs because no one would be employed to translate a casual conversation in real time. It also doesn’t replace the work of a language interpreter where accuracy is critical and the stakes are high.


More importantly perhaps, the technology is not adversarial.


An incorrect translation is at worst an occasional annoyance but humans are not “losing” to an AI like Kasparov in a game of chess. 



But our ongoing discovery of the speeds and limits in terms of what AI can do might just be a distraction from a deeper and more humanity-related problem. 


I am no mathematician. I do not know anything about the Navier-Stokes problem more than the engineering mathematics I had learned in university. 


What I do know is that we now live in a highly impatient world whereby organizations are building AI agents that can execute tasks faster than a human. And they can do it round-the-clock, effectively satisfying the new age of business owners and consumers that demand everything right now.


This fixation on immediate gratification have led us to squeeze every last second in the twenty-four hours we have every day, speeding things up dramatically. Patience becomes a drag and people now prioritize an instantaneous response over time-tested craftsmanship.


This comes down to what Fields Medallist, Terence Tao is saying as a "Severe Misalignment of AI in Mathematics" in response to the Navier-Stokes problem being solved by AI[5]. He describes it as:

“These companies are dumping carcasses of raw meat onto our communal village table and saying: ‘Here you go, I solved your food problem.’ And then they just leave,”

To be fair, this is not unique to the field of mathematics.


In other aspects of our daily lives, AI has significantly accelerated outcomes but never really solving the problem at hand.


Consider: reading, the accumulation of wealth, or the acquisition of knowledge and experience. 


Take wealth accumulation for instance: In a hypothetical setting, someone who successfully uses predictive algorithms to amass huge amounts of money within a short time might solve his financial problems but it does nothing to teach him the value of a dollar in a day’s hard work. 


Then there is also the conundrum when it comes to knowledge acquisition: e-books offered a newfound level of portability that conventional paperbacks could not provide. To top it off, there are even book summary apps such as Blinkist that condense the contents of an entire book, allowing a reader to digest its essence within fifteen minutes instead of spending the entire weekend being buried in hundreds of pages.


But if you are an avid reader like me, you would understand that nothing can replace the joy of reading a book from cover to cover, pausing and picking it up again. Part of what makes reading enjoyable and valuable is the gradual percolation of ideas, its synthesis into deeply personal reflections and how this eventually culminates into one's mindset and behaviour.


Irish philosopher, Edmund Burke once said,

“To read without reflecting is like eating without digesting.”

At the workplace, there is also the issue that agentic AI might take away the grunt work previously undertaken by the next generation of young graduates, depriving them of “going through the motion”. This rite of passage, according to The Economist, is essential to initiating them to the workforce[6].


"One argument for drudge work is its centrality to entry-level jobs. People who are new to the workforce know very little about the job they are meant to be doing. (The same is also true for lots of people at the end of their careers.) Grunt work has long been a way to fill junior employees’ time while allowing them to learn the basics of office life. Per photocopier ad astra. If AI takes on all the drudge work that has traditionally fallen to the newbies, employers may simply stop hiring them."


The article also puts forth the argument of balancing cognitive capacity whereby jobs that are less intense in nature may encourage the incubation of more creative ideas.


Interestingly but true, grunt work can also be effective in inculcating a sense of ownership:

"Much as nobody revels in doing their expenses, they can at least be done. You can press the submit button and feel minutely satisfied (until you are told the claim code is wrong)."[6]


“Practice makes perfect” means we acquire mastery by iterating. And iteration requires first-hand knowledge such as actually doing the work, walking the ground, and going out there to talk to people and learning the nuances of the trade, however mundane and time-consuming it may be—something that agentic AI misses.


Many desirable outcomes in the real world require repeated iterations and the necessary passage of time.


Development in AI doesn't necessary need to slow down. It is also not only about placing appropriate guardrails around privacy and personal information. It is about how the world would look like in the future with so much dependency on AI.


In Nick Bostrom's book Deep Utopia: Life and Meaning in a Solved World, he theorizes what life could possibly look like if AI turned out to be capable of solving all of our problems.


Without appropriate human iteration (and intervention), we risk creating a knowledge ecosystem that is highly transactional and built on potentially fragile foundations.


And God forbid we rely solely on deep research using ChatGPT, Claude, Deepseek or any other computer model to conduct a market feasibility study, design a new product or make the important decisions in life.






References:

[4] One of the foundational topics for artificial intelligence taught in universities is path finding.

bottom of page