Juergen Schmidhuber
2008 · arXiv
I argue that data becomes temporarily interesting by itself to some self-improving, but computationally limited, subjective observer once he learns to predict or compress the data in a better way, thus making it subjectively simpler and more beautiful. Curiosity is the desire to create or discover more non-random, non-arbitrary, regular data that is novel and surprising not in the traditional sense of Boltzmann and Shannon but in the sense that it allows for compression progress because its regularity was not yet known. This drive maximizes interestingness, the first derivative of subjective beauty or compressibility, that is, the steepness of the learning curve. It motivates exploring infants, pure mathematicians, composers, artists, dancers, comedians, yourself, and (since 1990) artificial systems.
Metadata: arXiv. Abstracts retain their original authors’ rights.
Original HN comments, with labels from simple text rules. A label describes the whole comment and may miss context or irony. Read the discussion to make up your own mind. About the labels
7 selected excerpts · 7 total mentions
Schmidhuber did it first: Driven by Compression Progress: A Simple Principle Explains Essential Aspects of Subjective Beauty, Novelty, Surprise, Interestingness, Attention, Curiosity, Creativity, Art, Science, Music, Jokes https://arxiv.org/abs/0812.4360
Schmidhuber reached your conclusions first. Driven by Compression Progress: A Simple Principle Explains Essential Aspects of Subjective Beauty, Novelty, Surprise, Interestingness, Attention, Curiosity, Creativity, Art, Science, Music, Jokes https://arxiv.org/abs/0812.4360
If your job is to attract an agent that is playing a careful balance between exploration and exploitation, while on tight computational budget, then you want to attract them with some initial hooks (e.g. bright colors) and appease their computational risk assessment module with high compressibility (e.g. symmetries and self-similar patterns). https://arxiv.org/abs/0812.4360
First, the competition does place a time limit (in fact several resource limits). I agree that it may be too lax to judge consumer-oriented compression algorithms. That is not what this competition is about. The goal is maximizing compression ratio - with the aim of affecting AI. Since an AI system can be seen as a lossy compression where some worldly input is compressed into something that we hope mimics "understanding". Improvements in compression may translate to AI advances [0] and potentially vice versa. To draw a parallel away from computing: One could say we are compressors that experience the world and compress it to loose what we don't care about. Improving our compression is basically akin to "knowing more" or "remembering more of the right things". See Schmidhuber [1]. [0] http://prize.hutter1.net/hfaq.htm#compai [1] https://arxiv.org/abs/0812.4360
There's a paper by Schmidhuber [1] in which he argues, IIRC, that not the surprise as in high or just-right entropy is what satisfies the observer of e.g. art, but the fact that it teaches you a new, previously unknown pattern of regularity about the world which allows you to compress information better. I may be misconstruing it though b/c it's been a while since I looked into it. On another note, Brian Eno has cited Morse Peckham's "Man's Rage for Chaos"[2], which seems to be more in line with what you are saying and to claim that the purpose of art is offering surprising shifts in perspective that break up the otherwise overly strong drive towards order and predictability. 1. https://arxiv.org/abs/0812.4360 2. https://www.goodreads.com/book/show/4552798-man-s-rage-for-c...
https://arxiv.org/abs/0812.4360 Juergen Schmidhuber - Driven by Compression Progress: A Simple Principle Explains Essential Aspects of Subjective Beauty, Novelty, Surprise, Interestingness, Attention, Curiosity, Creativity, Art, Science, Music, Jokes (2002)
allow me to make your day: - Computational trinitarianism, or how 1.) propositions as types, 2.) programs as proofs, and 3.) the relation between type theory and category theory are one and the same thing [0] - Scott aaronson's excellent "Why Philosophers Should Care About Computational Complexity" [1] - "From Philosophy to Program Size" by Greg Chaitin. Arguably Chaitin's work (notably Chaitin's omega) is a quintessential example of what lies at the intersection of mathematics, computation, and philosophy.[2] - From the inimitable Juergen Schmidhueber, "Driven by Compression Progress: A Simple Principle Explains Essential Aspects of Subjective Beauty, Novelty, Surprise,…
Interestingness, Attention, Curiosity, Creativity, Art, Science, Music, Jokes" [3] - A computational biology paper that presents a meta-learning scheme for cognition and emotion. This paper's idea has really stuck with me.. That and the whole notion of the "Dark room problem" (you could, probably, really easily come up with a strategy for living life according to a hybrid of what this article presents + any kind of philosophy there is) - "Emotional Valence and the Free-Energy Principle" [4] [0] https://ncatlab.org/nlab/show/computational+trinitarianism [1] https://www.scottaaronson.com/papers/philos.pdf [2] https://arxiv.org/abs/math/0303352 [3] https://arxiv.org/abs/0812.4360 [4] http://journals.plos.org/ploscompbiol/article?id=10.1371/jou...
More from the same topics.
William P. Thurston
In response to Jaffe and Quinn [math.HO/9307227], the author discusses forms of progress in mathematics that are not captured by formal proofs of theorems, especially in his own work in the theory of foliations and geometrization of 3-manifolds and dynamical systems.
Sébastien Bubeck, Varun Chandrasekaran, Ronen Eldan +11
Artificial intelligence (AI) researchers have been developing and refining large language models (LLMs) that exhibit remarkable capabilities across a variety of domains and tasks, challenging our understanding of learning and cognition.
Martin Kleppmann
The CAP Theorem is a frequently cited impossibility result in distributed systems, especially among NoSQL distributed databases.
Eric Chisolm
This is an introduction to geometric algebra, an alternative to traditional vector algebra that expands on it in two ways: 1.