Part 1: No Aesthetics, No AGI - Know Aesthetics, Know AGI
We are living through a sci-fi novel bungee-corded to disappointment. The whiplash is real. Amazon’s $40 in-ear translator has 144 built-in languages, sells for $39, and is far more user-friendly than Douglas Adams’ Babel fish – and yet AI can’t reliably email me a reminder to buy one before my next trip. My YouTube feed sparkles with this same sci-fi veneer, but too often it’s a monkey washing a cat in a sink, metaphorically speaking. The real and rapid improvements of silicon-based AI is head spinning, thrilling, and too often as useful as magic tricks performed during abdominal surgery.
For all its PhD knowledge, clicking the right button, selecting an option from a dropdown, or clearing a text field are often well beyond AI’s capabilities both technically and intellectually. While writing a recent essay, I asked my computer-use agent (the very latest in cool AI tools) to add page numbers. And It did so, beautifully, after first deleting the entire contents of the document. In the end, human-level intelligence will not arise from brute computation of language – or not language alone. Sensory input, what we see, hear, and touch, how we react to the world and how the world reacts to us, the beauty and pain of it all, the aesthetics of living is needed to make intellect effective.
Broadly construed, aesthetics is the root of human intuition and cognition – physical experience gained through perception, pain, play, and comfort is the canvas and oil upon which an elegant mathematical proof is painted. Large Language Models (LLMs) exemplified by AIs like ChatGPT are limited by their disembodied and word-centric nature. They excel at average-case prediction, but struggle to unshackle their “thinking” from shallow platitude because depth of understanding arises fundamentally from a mind situated in a body in a world. The statistical reasoning of LLMs is a pale reflection of human problem solving and the disentangling of anomalies. LLMs reason from existing data and are hobbled by it.
But techniques like reinforcement learning particularly in the blossoming domain of world models point the way toward an possible aesthetic grounding for AI, because without aesthetics or its fundamental components of sensory experience, AGI (Artificial General Intelligence) will remain unattainable. Conversely, by embodying AI and inviting an aesthetic sense, a genuine general intelligence capable of the long-tail challenges of a messy and illogical world becomes possible. In short: no aesthetics, no AGI – know aesthetics, know AGI.
(This post is part of a larger series that will be added to every few weeks)
