Modularity is critically important for AGI, but we should avoid a naive formulation of modularity.
The most developed notion of modularity comes from computer science. We have notions of encapsulation and all kinds of composition design patterns that are strategically employed to trade-off one concern over another.
Modularity thus is understood as a controlled coordination mechanism between interacting parts. Certain information is allowed to be malleable while other kinds must remain immutable.
Advanced modularity thus is a consequence of sophisticated protocols between agents. It is less about rigid boundaries between what's inside and outside an agent.
Advanced modularity is the dynamic exchange of information in an ecology of smart agents. This drives co-evolution and ultimately more complex emergent behavior.
In another life, I had explored interoperable protocol design. I was applying what I new into the field of deep learning. When I wrote the blog, I was uneasy about the relevance of the connections I made.
The impedance mismatch between protocol design and biological modularity is that the latter does not have a mind to design the actual protocols. Yet, we know that biology is unsurpassed in its adaptive behavior between interfaces.
Nevertheless, humans require conceptual scaffolding to reason about the complexity found in biology. We cannot make progress unless we use a more advanced vocabulary that expresses advanced forms of modularity.
Netflix's Queen's Gambit must be the ultimate mathematical nerd series. One cannot underestimate the detail of this movie and how it takes you back to a different time.
The movie takes place in the decade of the 1960s. Where a young child mother dies and is sent to an orphanage. What's incredible is how this movie reveals the changes in 60s, the technology, architecture, interior decoration, music and fashion style.
But that's just a slice of the movie, it's about a gifted chess player. Few may have noticed that her natural mother wrote a Ph.D. thesis in group theory. This reveals her unique innate ability.
We use the word causality as a means of understanding cognition but we don't really understand its distinctions. Let's look at what C.S.Peirce had to say about causality.
What @yudapearl says is that to understand a system one needs to hypothesize a model of the system and then see how this model is in agreement. Statistics is just one of the methods of testing. But it's not how one formulates the original model.
Peirce called this cognitive capability to hypothesize about the world as Abduction. Bayes rule is in fact a kind of abduction. When Bayesians talk about formulating priors, they are actually implicitly talking about an impoverished form of abduction.
(1) All technologies are combinations. Individual technologies are combined from components. (2) Each component of technology is itself in miniature technology. (3) All technologies harness and exploit some natural effect or phenomenon.
His framework is general enough so that we can recognize things that we don't conventionally consider as technology. These include culture, human organizations, processes, language and biology.
John Krakauer in a recent Learning Salon conversation focused on the huge gap between participatory learning and propositional learning. It occurred to me that propositional learning is a kind of hypnosis!
Coincidentally, today's currents events are a consequence of hypnosis. @scottadamssays was the first to notice Trump's apparent use of hypnosis methods. fortune.com/2020/09/27/don…