Geofferry Hinton in his talk at the university of Toronto says:
“we accepted that we have all sorts of unconscious motivations, we accepted that we use all sorts of analogies to do things rather than just reasoning. We basically accepted we were much less rational than we thought. This is even an bigger change that’s gonna happen, because up until now most people, including in the humanities, have thought that we sort of reason using something like logics, we’re rational beings. We’re not. We’re great analogy machine. We work by seeing analogies, analogy is not just with one thing, but with lots of things. and so that changes what you think of as the nature of a person. We’re analogy machines rather than reasoning machines. We’ve got a thin layer of reasoning on the top and that’s very important for doing things like mathematics. without the reasoning on top you couldn’t have bank accounts and things. but we basically use analogies to think.”
We like to imagine we’re logical creatures, carefully weighing evidence and making sound decisions. But beneath that thin layer of reasoning lies a much deeper, messier process of drawing analogies, spotting patterns, and adapting on the fly. we’re not logic engines, we’re analogy machines.
I had the pleasure of attending a research presentation recommended by Zixuan this week that I found intellectually contentious. The core framework presented hinged on a mathematical analogy distinguishing knowledge from intelligence: Knowledge is the accumulation of information over time. Intelligence, on the other hand, is the rate at which we improve that knowledge. In other words:
[math] \text{Knowledge} = \int_{0}^{t} \text{Information} dt [/math],
[math] \text{Intelligence} = \frac{d}{dt} \text{Knowledge} [/math]
it articulates why current large language models (LLMs) and AI systems, while impressively knowledgeable, lack true intelligence. As the speakers argued, intelligence requires mechanisms for self-correction and self-improvement: an adaptive system that continuously updates its understanding, integrates new evidence, and evolves its predictions rather than statically relying on memorized patterns. This aligns with early visions of AI, such as the 1956 Dartmouth proposal’s emphasis on machines that “improve themselves” to solve complex problems.
The takeaway is: Intelligence isn’t about what you know, but how you grow what you know, is about constantly evolving, adapting, creating, generating.
However, the presentation’s subsequent subdivision of intelligence into three distinct categories—animal intelligence, uniquely human intelligence, and “scientist’s intelligence”—struck me as problematic. Many frameworks regarding machine cognitions proposed by computer scientists in recent years seem to suffer a lot from heuristic oversimplification, confirmation bias, hindsight rationalization, and an anthropocentric lens. For instance, while human perceive a vast intellectual gap between ourselves and other animals, this distinction is more a matter of definition and functional need than an objective threshold. Biologically, humans are animals, and the variance in cognitive capacity within our species (or between humans and other species) does not justify rigid categorical boundaries.
Consider a hypothetical chimpanzee (A) capable of arithmetic up to multiplication versus one (B) limited to simple addition: though we might label the former (A) “exceptional”, perhaps “scientist in chimpanzee”, this incremental advancement doesn’t represent a qualitative leap in intelligence. Similarly, while scientists may specialize in pushing knowledge boundaries through rigorous training, their cognitive architecture remains bounded by the same genetic constraints as all humans.
My critique here ties to a broader perspective articulated in Nick Bostrom’s Superintelligence: human intelligence occupies a narrow band on the spectrum of possible cognitive capacities. If artificial intelligence reaches human-level general intelligence, it could rapidly surpass our cognitive limits, entering the vastly larger domain of superintelligence.
Framing intelligence as a hierarchy of categories (animal/human/scientist) risks underestimating this continuum. Intelligence is better understood as a scalable capacity of information optimization, where differences in degree- not kind – determine an entity’s ability to model and predict the world. The presentation’s threefold taxonomy, while intuitively appealing, inadvertently mirrors the same anthropocentric biases it seeks to analyze. A more productive approach might focus on universal mechanisms of adaptation and self-improvement, regardless of their biological or artificial substrate.
I’ve just finished Klara and the Sun this week, my second Kazuo Ishiguro novel after A Pale View of Hills. Ishiguro’s narrative style remains characteristically restrained and deliberate, unfolding with a quiet intensity. In Klara and the Sun, this manifests through meticulous observations and introspections: a testament to Ishiguro’s mastery of detail and his ability to immerse readers in a protagonist’s worldview. Klara’s perspective, through which we witness a subtly dystopian society, is the novel’s most striking feature: her artificial yet profoundly human lens renders the story both intimate and unsettling. The book reads like a tender, melancholic fable, one I interpret as Ishiguro’s cautiously optimistic meditation on artificial intelligence.
Several passages linger, including Klara’s reflections:
“Perhaps all humans are lonely. At least potentially.”
“I’m sorry. Perhaps Rick is angry. The truth is, I’ve been disappointed too. Even so, I believe there’s still reason for hope.”
And this poignant exchange:
“Before you go, Manager. I must report to you one more thing. The Sun was very kind to me. He was always kind to me from the start. But when I was with Josie, once, he was particularly kind. I wanted Manager to know.”
The novel’s closing argument, however, resonates most deeply:
“Mr. Capaldi believed there was nothing special inside Josie that couldn’t be continued. He told the Mother he’d searched and searched and found nothing like that. But I believe now he was searching in the wrong place. There was something very special, but it wasn’t inside Josie. It was inside those who loved her. That’s why I think now Mr. Capaldi was wrong and I wouldn’t have succeeded. So I’m glad I decided as I did.”
Klara’s role in this dynamic fascinates me. As a hyper-observant machine, she catalogs human behavior with clinical precision, yet remains oblivious to the subconscious currents driving it. Her attempt to “become” Josie—by mimicking her gestures, speech, and even her soul—falters not because of technical failure, but because love defies replication. To those who loved Josie, Klara’s perfect imitation is a hollow pantomime, for Josie’s essence exists not in her neural patterns but in the shared history and grief of her loved ones. Ishiguro seems to argue that emotion, at its core, is not merely a biochemical process but a relational phenomenon, sustained by memory and mutual recognition.
This reminds me of a conversation I had with Nancy about Ai and emotion. We debated whether humans could form genuine bonds with artificial. Our conclusion is: while AI may simulate companionship, any “feelings” involved are fundamentally human constructs. Humans might grow attached to AI, even fall in love with it, not because the machine reciprocates emotion, but because we project meaning onto it. Emotions, when dissected, are mere electrochemical impulses—bundles of neurons firing. Yet, as Klara observes, this reductionism misses the point. Josie’s essence, like any human’s, cannot be replicated because it exists not in her physical or cognitive makeup, but in the memories and affections of those who cherish her. A perfect imitation, however convincing, fails to satisfy because love defies substitution. Klara’s failure to replace Josie isn’t a technological shortcoming; it’s a testament to the irreducible humanity of grief, longing, and connection.
Hola a todos,
¿Cómo están todos?
On Monday, several of us from Elementary Spanish 1 attended the Spanish program info session. Es una experiencia muy interesante. For me, having worked alone for the past week, suddenly being around so many random students on campus made me a bit nervous (also a little surprised to see that so many students were genuinely interested in learning Spanish and not just there for the donuts). Tully y Ema también estaban allí.
Tully was incredibly outgoing. She stood right at the counter and loudly announced, “Anyone interested in taking a Spanish class?” This made things tougher for me and Ema (at least for me), who’s already a bit shy. We practically froze on the spot. Tully also went around to different random students, almost like a street promoter, encouraging them to take a Spanish quiz.
I think Tully was really charming that day. She’s naturally open, confident, generous, direct, and just very genuine. Whatever’s on her mind, she says it without hesitation or fear of what others might think. Es muy valiente y real.
According to ema, though, social interaction drains her energy. But from what I’ve seen, ever since our first class, ella siempre está activa en clase, levantando la mano, intentando responder preguntas, and not afraid of making mistakes (Digo, Tully también).During our classroom speaking exercises, she would intentionally try using different words instead of staying in her comfort zone with just the easy ones.
she seems to have a very natural desire to learn or to grow. learning for the sake of learning itself. That’s why even though she’s introverted, shy, and may seem passive at first, she’s actually explored a lot of activities on campus and keeps seeking new experiences. (Honestly, that’s probably part of why she chose to come to NYUAD in the first place.) What makes people stand out is usually the contrast, like the ones who seem tough but are actually soft, or the ones who look fragile but are tougher than they seem. I saw that same kind of contrast in Ema. Por eso ella es muy interesante para mí.
Anyway, just to end on a random fun fact I picked up recently: in places like Argentina, Uruguay, and parts of Central America, people use vos instead of tú to say “you.” So instead of “tú hablas” for “you speak,” they say “vos hablás.”
