The Tally is Humans 1, AI 0 — At least for Now

Image: Wisdom 2.0

Two days ago, I sat alongside 800 other folks at Wisdom 2.0 in San Francisco, watching Sam Altman, the creator of ChatGPT, and his Buddhist meditation teacher Jack Kornfield discuss Mindfulness, AI and The Future of Life on the Planet. AI it would seem, is poised to destroy both humanity and our pale, blue dot of a planet, in the process dwarfing human intelligence to morph into an all seeing Eye of Omniscience. Setting Sam Altman’s ambivalence (and barely concealed pride), it’s interesting to explore the limits of both AI and human intelligence. In this post, we’ll see how human intelligence is full of hidden possibilities that each one of us may be able to tap into. You’ll even read about one (or two, depending on your appetite for breaking the law) methods of tapping into your own secret reservoir of invention and creativity. As for AI, at least the current generation, hype seems to have overshadowed reality.

Associative Thinking, Knowledge Models and Reasoning

Sydney, like all LLMs, works by association, by piecing together words in sequences based on patterns it has seen in training data. Sydney’s coherent-sounding utterances are not based on a coherent underlying knowledge model. As humans, our language abilities are similarly based on association. On the other hand, we do form knowledge models — computer interpretable models of knowledge or standard specifications — and are able to reason over these knowledge models using predicate logic to arrive at precise conclusions.

It’s inexpensive for humans to engage in associative thinking and millennia of evolution have enabled our brains to excel at it. It’s much more neurologically expensive for humans to reason. That’s why we pick up language organically, but it takes us years of formal education to learn to reason. Anyone who has sweated over high school math and calculus will know what I am talking about. Typically, only those of us with the financial resources and good fortune to have access to years of formal education and good teachers (!) are able to develop more advanced abilities to reason. Even then, it doesn’t come easy to us.

Nobel prize winning economist Daniel Kahneman in his 2012 best-selling book Thinking, Fast and Slow distinguishes between two modes of human cognitive functioning, one lazy (and easy on our brains) and fast, the other more effortful (or hard on our brains) and slow. In the context of language and reasoning, you might say that we are able to speak without thinking, but struggle to speak thoughtfully.

(Languages like Sanskrit straddle the gap between associative and knowledge models. Sanskrit is sometimes compared to a computer language because of its highly structured grammar and syntax, as well as its precise and logical nature. For this very reason, Sanskrit never took off as a language of the masses. Instead, derivates like Hindi with far fewer rules and more suited to organic, associative learning have almost entirely replaced Sanskrit.)

AI can reason effectively about very narrow problem domains like Chess (Alpha Zero) or Go (Alpha Go), but general reasoning — over a broad domain like all of mathematics or about the world at large — is beyond the ability of today’s technology. Humans excel at general reasoning, but very few excel at reasoning deeply over a specific domain. So, are both humans and AIs doomed to struggle when it comes knowledge and reasoning, though for different reasons? Not so fast. As you shall see below, you and I may have latent abilities for both deep and fast reasoning and creativity over complex domains like math and science that far exceed what AI is currently capable of.

The Case of Shakuntala Devi, the Human Calculator

When I was in high school back in Chennai, India, the “mental calculator” Shakuntala Devi (1929–2013) visited our school. Shakuntala Devi excelled at arithmetic problems, such as computing “the 23rd root of a 201-digit number in 50 seconds. Her answer, which was 546,372,891, was confirmed by calculations done at the US Bureau of Standards by the UNIVAC 1101 computer, for which a special program had to be written to perform such a large calculation, which took a longer time than for her to do the same.”

She also demonstrated the multiplication of two 13-digit numbers — 7,686,369,774,870 × 2,465,099,745,779, correctly answering 18,947,668,177,995,426,462,773,730 in 28 seconds. Here’s the answer ChatGPT provided for the same problem, illustrating the limitations of LLMs: 18,952,052,704,314,516,712,345,330. Which is the wrong answer. ChatGPT got it wrong because it “fudged” the answer by scanning through its database of billions of documents for patterns that match the text I fed into it, and then applied a probabilistic model on top of the match it found to “generalize” the answer.

Coming back to Shakuntala Devi. Clearly, Devi could reason — in the sense of using a logical process to arrive at a precise answer — far more efficiently than most humans. What combination of knowledge representation and reasoning engine in her head enabled her to do so remains a mystery. All we can say is that her brain was wired differently. However, she had no special talent for advanced mathematics — theorems and the like. Here’s the case of another remarkable human who did.

Srinivas Ramanujan, The Man Who Knew Infinity

The Indian mathematician and prodigy Srinivas Ramanujan (1887–1920) was a prodigy whose method hints that advanced reasoning abilities might be organically available to at least some humans…“with almost no formal training in pure mathematics, Ramanujan made substantial contributions to infinite series, continued fractions and number theory, including solutions to mathematical problems considered unsolvable.”

What’s remarkable, Ramanujan would often come up with theorems with no proof whatsoever, claiming that the answers were given to him in dreams by the Goddess Namagiri. Upon seeing his theorems on continued fractions like the one below, the famed Oxford mathematician GH Hardy remarked that “the theorems defeated me completely; I had never seen anything in the least like them before”, and that they “must be true, because, if they were not true, no one would have the imagination to invent them”.

Figure: One of Srinivas Ramanujan’s theorems on Infinite Series and Continued Fractions

Where Ramanujan defies our understanding of how reasoning abilities are developed is that he had almost no formal training in mathematics. He did not have to arduously cultivate reasoning skills and build a knowledge model over two decades of school and college like the rest of us. His insights seemed to come to him from some mysterious internal process that he later had to confirm through the laborious (by his standards) process of developing formal mathematical proofs.

While Shakuntala Devi was content with computing the answers to problems presented to her, Ramanujan’s mind churned out a never-ending series of new and ground-breaking theorems. Ramanujan’s feats of reasoning were also feats of creativity. And creativity requires associative thinking, or the ability to make connections and associations between seemingly unrelated concepts or ideas. Humans have the innate ability to draw associations across seemingly unconnected fields, say a complex geometrical pattern and an advanced theory in physics (more of this later on down) that today’s AI is incapable of. Perhaps the human brain accomplishes this feat using conceptual and semantic webs, by storing facts and experiences in clusters or concepts and linking them to similar concepts in an ever-expanding and growing brain-wide web, which is a topic for a future blog post.

In the meantime, let’s consider how you and I might be able to tap into our own inner Ramanujan like him and other geniuses of our times.

Edison, Einstein, Kekulé and Jason Padgett

Ramanujan’s claim that he received his theorems in dreams is confounding, unless you consider how other geniuses like Edison and Einstein arrived at their insights.

Edison was well-known for tapping into the hypnagogic state between sleep and wakefulness to find answers to problems. When attempting to solve a particularly hard problem, Edison would nod off to sleep on a chair holding a steel ball in his hand. When his hand relaxed sufficiently, the ball would fall to the floor and “wake him up with insights to his problems”. (A study reported in this 2021 science.org article validates how the state between sleep and wakefulness can help problem solving).

Einstein used deep visualization techniques akin to meditation for his gedankenexperiment or thought experiments. At the age of 16, he visualized himself chasing a beam of light to arrive at the intuition that the speed of light is constant, which formed the basis of his special theory of relativity.

The nineteenth-century German chemist August Kekulé claimed to have pictured the cyclic ring structure of benzene after dreaming of a snake eating its own tail. In 1854 or 1855, after falling into a dream reverie aboard a horse-drawn London bus, Kekulé saw ‘’the atoms gamboling before my eyes…the larger ones forming a chain, dragging the smaller ones after them but only at the ends of the chain’’, which he said inspired his discovery of the propensity of carbon to form straight chain hydrocarbons.

What if you are not born with a brain wired like Ramanujan, Edison, Einstein or Kekulé? All is not lost. Consider the case of Jason Padgett, a futon salesman who, after being hit on the head by muggers outside a bar in Tacoma, Washington, discovered a sudden talent for advanced mathematics and geometry. His portfolio of geometric art is on Pinterest and apparently reveals deep connections with advanced physics theories, as described in “A Visual Representation of why E=MC²”. While I can’t find any authoritative, academic verification of this claimed link to physics, Padgett’s art by itself is very impressive. It is the kind of creativity that requires pretty advanced spatial or geometric reasoning.

Tapping Into Our Inner Super Powers of Reasoning and Creativity

The question I’d like for us to consider is this: does the human brain have an innate ability for advanced reasoning and creative association that lies out just out of reach and can be tapped into even if we are nor born with it or experience an accidental rewiring due to a blow to the head?

Consider that Steve Jobs once suggested that Microsoft products would be better if Bill Gates “had dropped acid once or gone off to an ashram when he was younger”, that Francis Crick claimed to have envisioned the structure of DNA during an acid trip, or that John Lennon attributed the Beatles’ album Revolver to the group’s acid use. Of course, all three of them were already able to tap into their creative super powers without LSD, and dropping acid is not an option for most of us. Plus, most people who got stoned didn’t turn into geniuses.

Now, consider that Grigori Perelman, the Russian mathematician who was awarded the Fields Medal in 2006 for his work on the Poincaré conjecture, has said that his practice of meditation and yoga helped him to solve the conjecture. And that Brian Josephson, the physicist who won the Nobel Prize in Physics in 1973 for his work on the theory of superconductivity, is a practitioner of Transcendental Meditation (TM) and has written about the connections between science and spirituality. Sam Altman himself has cultivated a meditation practice over the past decade and has described how it has helped him to reduce stress, increase focus, and gain clarity of thought. Meditation — that’s more palatable than LSD for most of us, if only by a little.

Meditation shifts our brain waves from the normal Beta state (12–30 hz) to Alpha (8–12 hz) and Theta (4–8 hz) states, the latter two corresponding to relaxed awareness and dreaming respectively. In the brains of advanced meditators, Gamma waves (30–100 Hz), associated with high-level cognitive processing, are present. Gamma waves have also been found to be active during REM sleep, particularly during vivid dream states. (Interestingly, multiple studies published in have shown that LSD has similar effects, increasing alpha, theta, and gamma brainwave activity, while reducing beta wave activity.) I can say this from personal experience: on those days that I sit down for my one hour of morning meditation, I am far more creative and productive than on the days I don’t.

And if Meditation sounds like too much hard work or too woo-woo, you can shift your brain waves with good, old-fashioned tech, as I described in an earlier blog post titled “My Experiences Hacking My Way to Mindfulness with Technology.” Silicon Valley is full of startups that offer technical shortcuts to enhanced creativity, bliss and nirvana. Neurable is a Boston-based startup spun out of the University of Michigan’s Direct Brain Interface (UM-DBI) laboratory that is developing headphones to measure your levels of sustained focus; Muse is a “brain sensing headband” and companion app from InterAxon, a company with heavyweights from Stanford and Harvard Medical School on its board, that gamifies meditation. And Kernel, a company founded by entrepreneur Brian Johnson of Venmo fame (he sold it to Paypal for $800M) has built a $50,000 bicycle helmet that will measure your brain waves and much, much more as you meditate or ingest psychedelics, as suits your taste.

I would venture to say that meditation is a more reliable tool than LSD to tap into higher states of reasoning and creativity. I can also say that in my own forays in meditation, I haven’t yet tapped into that advanced Gamma brainwave state that would turn me into a Nobel or Fields Medal winner. Coming back to the question of whether the human brain can tap into such superpowers hidden just out of reach for most us, I would put LSD in the “accidental rewiring” category and meditation in the “right technique” category — the latter just requires sufficient sweat and discipline.

One thing is clear — we are only scratching the surface of what our brains and minds are capable of. While Ramanjan and Einstein may have been born geniuses, the case of Jason Padgett should give the rest of us hope. As for Sydney, ChatGPT and the current generation of AI, knowledge models and the ability for general reasoning are still out of reach…but who knows what is just around the corner?!