[the notes from the book were retrieved with kobogarden, with the purpose of aiding to create a map of the ideas the book left me. The full list of book highlights can be found here. Reading is still in progress.]

why this book

I had come across the author on the Mindscape podcast, and I was curious to delve further, with the expectation on possibly expanding my knowledge on the field of cognition and adjacent topics.

I haven't yet finished it, and currently stand somewhere at its midpoint. After the introducing chapters on what is educability and some other basic concepts, it springboards from the information processing perspective to superimpose our process of learning with processes from machine learning. It is a very computational-heavy perspective (an interesting one, too);

the definining characteristic of humanity

The mark of humanity is that a single individual can acquire the knowledge created by so many other individuals. It is this ability to absorb theories at scale, rather than the ability to contribute to their creation, that I identify as humanity’s most characteristic trait.

It starts by stating educability as a conceptual alternative to intelligence, as the latter is poorly defined (and consequently poorly measured); not only that, there is an argument for educability as the defining cognitive capability in our evolution, and what let us progress towards civilization (the author seems to have also coined the term civilization enabler.

educability can be modelled as information processing

To be able to build on the knowledge of our predecessors, we go farther — but since we are acting upon information, the author considers that the subject matter of educability is information processing: this perspective will bring us closer to Computer Science and adjacent notions of computability, even as far as to argue that just as the field of physics is concerned with describing and modeling the real-world, an analogous match can be made between the study of the brain and the field of information processing.

While the twentieth century saw unparalleled developments in the classical sciences, equally important and particularly in the work of Alan Turing in the 1930s, it saw the birth of the science of information processing. By that time, it was commonplace not to marvel that physical concepts that are not visible, such as energy or electric charge, could have useful meaning. The fact that the same held for notions of information processing and computation, terms that I shall use synonymously, was startling news. The import of this news was well understood by the early pioneers of computing, namely, Turing himself and John von Neumann. They sought immediately to use computation to study biological phenomena, such as the brain, cognition, and genetics. Each of these phenomena involves the transformation of information. Focusing on the information processing rather than the physical realization became a viable and necessary approach toward understanding these once a scientific approach to information processing had come into view. I, and many others, consider it self-evident that if we are to understand how the brain works, we will need to understand it in terms of information processing.

Later on, an experiment with sea snails, in which their behaviour is altered upon stimuli that makes them closer to an evolutionary need, is analyzed through the lens of a learning algorithm, and it serves as an example of how the learning process — educability — can concretely be taken as information processing.

Educability can then be defined as the product of seven capabilities:

In each of the next seven sections, I shall focus on a separate capability that will be used as a foundation stone for the model. These capabilities are learning from examples, generalization, large memory, symbolic names, teaching, chaining, and a Mind’s Eye. Each one is computational, in the sense [that] information is being processed to meet a specification, and this specification can be realized with feasible resources. The physical substrate on which the processing takes place, whether biological or silicon, is immaterial.

the definition of PAC learnability

In computational learning theory, probably approximately correct (PAC) learning is a framework for mathematical analysis of machine learning. It was proposed in 1984 by Leslie Valiant.

In this framework, the learner receives samples and must select a generalization function (called the hypothesis) from a certain class of possible functions. The goal is that, with high probability (the "probably" part), the selected function will have low generalization error (the "approximately correct" part). The learner must be able to learn the concept given any arbitrary approximation ratio, probability of success, or distribution of the samples.

from Wikipedia