It's rare to encounter a book that genuinely earns the accolade of "best ever," even in our age of hyperbole. Rarer still is to find such work published not centuries ago, like a classic or in last week's bestseller list, but written during our times and more or less overlooked. Gary Marcus' Kluge, a book I stumbled upon with no prior knowledge, proved to be such a revelation. From its intriguing title to its final sentence, it's an exhilarating journey through the fascinating landscape of the human mind. In a field blessed with exceptional works like Thinking, Fast and Slow, Nudge, Predictably Irrational, or slightly farther apart from books like Freakonomics and The Selfish Gene, Kluge stands out as its own type of masterpiece. For those interested in a deeper dive or learning what sets the book apart, I invite you to explore my review at https://geninnov.ai/s/XAwRaA.
In retrospect, I'm grateful to have encountered Kluge at this particular juncture. It provides the perfect framework for understanding the seemingly chaotic evolution of GenAI. While some may lament the randomness and lack of clear direction in this field, Kluge illuminates how this multi-faceted GenAI development, reminiscent of the build-out of our brains, may ultimately prove to be one of its greatest strengths. Importantly, whether desirable or not, this haphazard development is an intrinsic characteristic of these transformative technologies that we must acknowledge and appreciate. To fully grasp its implications, we must first learn what Kluge is!
Kludge: Evolution Has No Foresight
"Evolution is neither clever nor penny-pinching." —Gary Marcus, Kluge: The Haphazard Evolution of the Human Mind
A kluge is an inelegant yet effective solution cobbled together from what's at hand—a patchwork response to a complex problem. Evolution, the book argues, turned the human brain into a quintessential kluge—messy, convoluted, yet astonishingly effective. At every point in the build-out of our brains, like all other things that define life, natural forces are optimized for the issues at hand. What one gets from stacking these solutions, which were ideal for winning the near-term survival race, eventually is neither elegant nor foreseeable. What one sees today is not perfect, but it is a marvel nonetheless for all its quirkiness and ability to keep evolving.
The Internet: A Digital Patchwork That Worked
"…the characteristics we hold most dear, the features that most distinctly define us as human beings—language, culture, explicit thought—must have been built on a genetic bedrock originally adapted for very different purposes." —Gary Marcus, Kluge: The Haphazard Evolution of the Human Mind
The notion of a kluge extends beyond biology. The Internet is perhaps the most prominent example of a technological kluge in human history. What began as ARPANET's decentralized communication network evolved through various protocols and standards, each solving immediate problems without a grand master plan. For instance, the early adoption of TCP/IP wasn't necessarily the most efficient protocol but offered the crucial advantage of reliability through redundancy. This led to numerous technological cul-de-sacs - from Gopher to WAIS, IRC to USENET, and CompuServe to AOL's walled gardens. Even in messaging, we witnessed the rise and fall of ICQ, MSN Messenger, and Yahoo Messenger, each adding features that seemed revolutionary at the time but ultimately proved to be evolutionary dead ends.
GenAI: A Kluge of Algorithms, Patches, and Promise
"Evolution is driven by the immediate. Gene-bearing creatures either live and reproduce or they don't." —Gary Marcus, Kluge: The Haphazard Evolution of the Human Mind
Amid the proliferation of meticulously planned apps and products over the past few decades, we have perhaps forgotten how anything can evolve organically and demographically from the ground up without any single organization or group fully grasping what they were building or why. GenAI, advancing at an explosive pace across diverse domains, is a kluge like few things in technology history except the Internet. Different players and researchers are independently optimizing solutions for whatever issues they take on, leading to a situation where everything appears clunky and uncoordinated. Those who love the structured paths cannot understand the motifs behind most developments and lament the lack of "killer apps." Others are puzzled by the messy nature of what is coming out and the complete absence of plans.
In the world of presentations containing precise long-term goals and conferences where everyone lays out their ten or twenty-year visions, GenAIs are queens of chaos and proudly flying blind.
Patchwork of Progress: Competing Approaches to Making GenAI Stick
"But perfection is clearly not always the way; the possibility of imperfection too becomes apparent when we realize that what evolution traverses is not just a mountain, but a mountain range." —Gary Marcus, Kluge: The Haphazard Evolution of the Human Mind
The downside of explosive evolution isn't just unpredictability; it's also the many dead ends it creates.
It’s telling, isn’t it? A few months ago, Retrieval-Augmented Generation (RAG) was a big idea, stitching together external information and raw model capabilities. Today, more streamlined alternatives like Google's NotebookLM or OpenAI's Memories appear to be providing completely different versions of how one can integrate personal data pools. The patches are already coming in—Perplexity's Pages, Claude's Data Analytics, OpenAI's Canvas—attempting to add coherence, usability, and practicality to the GenAI ecosystem. The tools overlap in scope, bumping into each other—each a little different, none definitive, all trying to plug the gaps created by this mushrooming wave of innovation. What will stick? What will emerge as a cohesive interface, like the browser for the early internet? Right now, there are no clear answers. It is easy to understand why many would want to begin exploring only when things are settled; it is difficult to explain why such settling may not happen for a very long time, and the costs could be high for those who dither their journeys for too long.
Agents are a Phase…Not Memories
"Memory is, I believe, the mother of all kluges, the single factor most responsible for human cognitive idiosyncrasy." —Gary Marcus, Kluge: The Haphazard Evolution of the Human Mind
Rather than getting confounded by terms like RAGs, Agents, Copilots, and others, the best way to think about the developments of recent weeks and months is that models develop memory through interactions and the ability to manipulate the environment. Killer App seekers will not find anything revolutionary any morning, but evolutionarily, the change is explosive at a level that defies description.
The use cases are evolving rapidly. In fact, they’re arriving so fast that it is difficult to decide which one may endure and which will only improve into something else. Consider something as simple as setting reminders. Today, it's difficult to set a recurring reminder for Diwali if you're bound by the Gregorian calendar. But GenAI agents could soon remember that every year—a small, effective optimization for personal relevance. More intriguing still, think about conditional reminders that these new-age agents might manage—like presenting your New York friend with a bottle of champagne and your LA friend with flowers if the Yankees win the World Series, but swapping the gestures if the Dodgers triumph.
This writer has got tremendous insights about the self by asking the models to identify various personal blind spots based on the questions I have asked them. With group and team memories collecting individual members' actions and streams of consciousness in unstructured ways, organizational productivity tools are about to undergo a revolution. Personally, healthcare, education and social aspects could be the first ones to undergo a change while GenAI will continue to radically alter the way we engage almost any professional from any service sector.
From Imprecision to Imperfections: Don't Wait for Killer Apps or Settled Paths
"Evolutionary inertia occurs because new genes must work in concert with old genes and because evolution is driven by the immediate." —Gary Marcus, Kluge: The Haphazard Evolution of the Human Mind
"Part of the answer stems from how our prelinguistic ancestors evolved to think about the world: not as philosophers or mathematicians, brimming with precision, but as animals perpetually in a hurry, frequently settling for solutions that are 'good enough' rather than definitive." —Gary Marcus, Kluge: The Haphazard Evolution of the Human Mind
Notwithstanding the messy entanglement of the Internet beneath the surface of impeccably designed and executed usecases, few real life participants had to bother about the cludginess of the architecture. The GenAI world demands embracing imprecision and flawed outcomes. The folks with media megaphone who make an industry out of every error they find have a useful function in throwing lights on what needs improvement, but they should not be used to learn the true nature of the new technology and its astounding evolution.