Suits. Palaces. Compression.
On a busy morning at a highly competitive, cut-throat law firm, the air is filled with the smell of coffee and the flavor of snark, mixed with glares of office politics and cabal-like behavior amid the humdrum of an event , an interview. At this high-brow, elitist group that only hires individuals who radiate an equal amount of exclusivity, something unexpected happens. A drug dealer, running from the cops and for his life, serendipitously ends up in the interview room and gets interviewed by the most prestigious law firm in Manhattan. Surprisingly, despite never attending college, let alone law school, he gets hired! Blessed with a photographic memory, he can recite arguments from arcane cases and books better than seasoned veterans, just from having glanced at them once, years ago. How is this even possible? This is the incredible premise of “Suits,” the most-watched legal drama on Netflix this year.
Every now and then, Hollywood presents us with a genius possessing an eidetic memory, captivating us with their brilliance. The list is long and includes characters like Good Will Hunting, Mike Ross, Sheldon Cooper, and others. Yet, reality is quite different from this ideal. As a civilization, we no longer value great memories as much as we used to. Instead, we value computing machines with incrementally more memory and much more storage. There’s plenty of pedantic advice against the ill effects of rote learning and a vigorous amount of advocacy for teaching teenagers how to code, how to debug hardware, how to solve math puzzles, and getting them more interested in STEM subjects. Yet, when was the last time you enjoyed someone reciting Shakespeare to you? Or the last time you enjoyed a contest where two participants recalled the 50th decimal digits of pi? While we do have spelling bees, they’re more of an accolade collection exercise to reach elitism. Why do we value reasoning a bit more than memory now, and was it always this way? It’s possibly because, we now have computers. With their infallible memory and near-perfect recall, we don’t need humans for that anymore.
A few years ago, I came across a memory champion. After getting to know them a bit, I asked the obvious question: how do you do it? How do you remember every single license plate you barely glanced at, every alphanumeric sequence you saw, for just a second?
He revealed his well-known but fascinating secret: he builds memory palaces in his mind, a method dating back to ancient Greece. This technique associates information with vivid, spatial memories. For instance, to remember a sequence like a license plate, he would construct a detailed, interactive scenario involving elements of the number. This technique not only helps in retaining numbers but also in recalling them quickly by walking through these mental landscapes. To explain further, imagine trying to memorize your wife’s license plate number, 6956–231. Picture going to meet your wife on her birthday. You carry 6 gifts, including a bouquet with 9 flowers, and a box with 5 chocolates. You climb up a tiny staircase of 6 steps, stop (-) to take 2 deep breaths, walk 3 more steps, and knock once before she appears. This scene, punctuated with numbers, helps you recall her license plate by simply replaying this scene. A memory champion, like him, has built an entire curated repository of memories based on this technique.
When I first heard this story (years ago) as an early student of machine learning, I was led to think that memorization and generalization were two distinct ideas, one better than the other. A poor model memorizes the training data and can’t generalize beyond it, while a good model learns a compressed version of the training data to generalize better on unseen data. The quality of compression determines the quality of the model. But! This is exactly why, the above story blew my mind. The best memorizers in the world don’t actually memorize facts literally; that’s for mere mortals like us. They compress it in ways that ties it to their human experience, which eventually leads to faster recall.
What is an easily palatable, information-theoretically near-optimal, easy-to-recall format of human communication?
A story.
Have you ever attended a presentation where the presenter drones on about facts and figures versus when someone like Steve Jobs speaks eloquently, with just one bullet point on his slides? Which one sticks?
The better story.
So here’s the dichotomy: better memorization can only happen when the memorization mechanism can generalize, and effective generalization requires a ton of memory to be effective.
This is not a debate; it’s an equilibrium.
Thus, the skills we cherish reflect our current needs and technological capabilities, suggesting an evolution of valued human traits over time.
Amogh Mahapatra's Writings 