top of page


I sigh because this book was not fun in the way the title suggested. In all honesty, I found it quite dry and long, I feel it could have been written in a quarter of the words it took. On the other hand, I'm fairly sure my notes are of little to no value to anyone else reading them without context, as I made no attempt to summarise the book at all, rather picked out concepts I found interesting.

Actionable takeaways

  • Arguments in favour of AI probably never becoming able to replicate the human mind, because we depend heavily on categorisation and this is a very non straightforward process (it can be something else though)

  • When we try to categories our notes into files, second brains the system is always imperfect, while our own brains seem to do this easily: we seem to have little insight into how we form categories ourselves, and when we try to apply our logical assumptions as to how this happens on a computer, we end up failing every time

  • Our ideas of what people should be doing when learning depend on our concept of learning itself, so you need to start from the bottom up to improe this process

  • If you accept that your knowledge is point-of-view, you need to accept that others points of view can be valid too

Main issue

  • Do meaningful thought and reason concern merely the manipulation of abstract symbols and their correspondence to an objective reality, independent of any embodiment (except, perhaps, for limitations imposed by the organism)?

On objectivisim

  • We will be calling the traditional view objectivism for the following reason: Modern attempts to make it work assume that rational thought consists of the manipulation of abstract symbols and that these symbols get their meaning via a correspondence with the world, objectively construed, that is, independent of the understanding of any organism.

Why should we care about learning to learn

  • If we fully appreciate the role of the imaginative aspects of reason, we will give them full value, investigate them more thoroughly, and provide better education in using them.

Our ideas about what people can learn and should be learning, as well as what they should be doing with what they learn, depend on our concept of learning itself.


The way we group things and form connections is highly culturally dependent, there is an amamzing example of a language which divides all nouns into 4 categories, based on their connections to 1-men; 2-women; 3-food and 4-everything else.

  • Australian aboriginal language Dyirbal, which has a category, balan, that actually includes women, fire, and dangerous things. It also includes birds that are not dangerous, as well as exceptional animals, such as the platypus, bandicoot, and echidna.

How do we name things?

  • In his celebrated paper, “The Meaning of a Word," written in 1940 and published in 1961, Austin asked, “Why do we call different kinds of things by the same name?”

Although there is a biologically uniform way for people to understand “fuzzy” colours such as orange as the colour between different amounts of red and yellow, you would then expect and all people to see all colours the same way, identify the same colour as seeming the same across cultures but this isn’t the case

  • In Kay-McDaniel terms, this means that the fuzzy-set-theoretical functions that compute conjunctions and disjunctions for color categories are not exactly the same for all people; rather they vary in their boundary conditions from culture to culture. They are thus at least partly conventional, and not completely a matter of universal neurophysiology and cognition.

Can interpretation be universal?

  • Paul Ekman and his associates have studied in detail the physiological correlates of emotions (Ekman 1971; Ekman, Friesen, and Ellsworth 1972). In a major crosscultural study of facial gestures expressing emotion, Ekman and his associates disco; ered that there were basic emotions that seem to correlate universally with facial gestures: happiness, sadness, anger, fear, surprise, and interest. Of all the subtle emotions that people feel and have words and concepts for around the world, only these have consistent correlates in facial expressions across cultures. (Although in talking to strangers this wasn’t the case)

Our categories are not straightforward

Although we would all agree that both robin and chicken are birds, we categorise a robin much faster as a bird than we do a chicken - this suggests that there are endless nuances in the way we group things in our heads, and perhaps this can be one of the limitations in AI trying to replicate the human brain.

  • The pervasiveness of prototypes in real-world categories and of prototypicality as a variable indicates that prototypes must have some place in psychological theories of representation, processing, and learning. However, prototypes themselves do not constitute any particular model of processes, representations, or learning.

  • Her experimental ranking shows that subjects view robins and sparrows as the best examples of birds, with owls and eagles lower down in the rankings and ostriches, emus, and penguins among the worst examples.

  • Hinton (1982) gives a similar case from Mixtec, an Otomanguean language of Mexico. Mixte c has three causative morphemes: the word sá?à, and the prefixes sáand s-. The longest of these corresponds to the most indirect causation, and the shortest to the most direct causation.

  • Neutralization of contrasts can also occur in semantics. Consider contrasts like tall-short, happy-sad, etc. These pairs are not completely symmetric. For example, if one asks How tall is Harry? one is not suggesting that Harry is tall, but if one asks How short is Harry? one is suggesting that Harry is short. Only one member of the pair tall-short can be used with a neutral meaning, namely, tall.

  • The housewife-mother stereotype is therefore defined relative to the nurturance model of motherhood. This may be obvious, but it is not a trivial fact. It shows that metonymic models like stereotypes are not necessarily defined with respect to an entire cluster.

(A mother gives her child up for adoption and goes to work, she is still a mother, she is still working, but she is not a working mother - would a computer understand this?)

  • Cognitive reference points within a submodel show prototype effects of the following sert: Subjects will judge statements like 98 is approximately 100 as being true more readily than statements like 100 is approximately 98.

"He ran into the forest." "The road ran into the forest" (we use a long thing trajectory of human running as the commonality in these two categories)

  • Gould's discussion is particularly interesting: Some of our most common and comforting groups no longer exist if classifications must be based on cladograms (evolutionary branching diagrams]. . . . I regret to report that there is surely no such thing as a fish.

I do not believe that nature frustrates us by design, but I rejoice in her intransigence nonetheless.

(Gould 1983)

  • For example, many of us have two ways of understanding electricity—as a continuous fluid that flows like water and as a bunch of electrons that move like people in a crowd. Gentner and Gentner (1982), in a remarkable study of how people actually learn about and understand science, showed that it is common for people to have both of these folk models of electricity and to apply one in solving some problems and the other in solving other problems. As they observed, these are conflicting models, in that they give different results in a certain range of problems. One has to learn which model to apply in which range of problems. In short, if words can fit the world, they can fit it either strictly or loosely, and the hedges strictly speaking and loosely speaking indicate how narrowly or broadly one should construe the fit. For instance, take Kay's example: Loosely speaking, the first human beings lived in Kenya. In a strict sense, there were no such things as “the first human beings”-at least assuming continuous evolution. But loosely speaking, this expression can be taken to refer to primates with important human characteristics. And Kenya, if you want to be picky, didn't exist then. But loosely speaking, we can take “in Kenya" to be in the general part of Africa.

  • Presuppositions are background assumptions of ICMs. As such, they are not subject to the criticisms that have been made of the logical and pragmatic notions of presupposition. Second, they allow us to understand more clearly what went wrong with the notion of analytic truth.

  • Good examples of complex categories are often bad examples of component categories.

Becoming humble in our 'knowledge' of the universe

  • The problem arises in the following way: Since institutions are products of human cognition, institutional facts must depend on human cognition, which violates the Independence Assumption, which states that no facts can be dependent on human cognition.

  • It is our objectivist legacy that we view rationality as being purely mental, unemotional, detached-independent of imagination, of social functioning, and of the limitations of our bodies and our memories. It is our objectivist legacy that leads us to view reasoning as mechanical and to glorify those kinds of reasoning that in fact are mechanical. It is our objectivist legacy that leads us to view machines that are capable of algorithmic computation as being capable of human reason. And it is our objectivist legacy that we view it as progress when we are able to structure aspects of our physical and social environment to make it more like an objectivist universe. The advent of the digital computer has accelerated our attempts to make our environment and our society fit objectivist metaphysics.

(If the truth of something depends on our categorisation or grouping of it, then how can there be objective truth?)

  • And there does not appear to be any sort of direct relationship between the mind and the world of the sort hypothesized in model theory. Color categories exist in the mind, but simply do not correspond to anything like set-theoretical entities in the world. Metaphorically defined categories do not seem to correspond to anything that exists independent of human conceptual systems. And perception, which is often taken as characterizing the links between the mind and world, is not veridical; it does not even preserve the number of entities, since people can see one light movingwhen there are two lights flashing.

In short, the empirical studies discussed in this book suggest that, on all three grounds, there can be no justification for extending mathematical logic from the domain of mathematical reasoning to the domain of human reason in general.

(If you have a point-of-view-knowledge you need to accept that other points of view can be legitimate)

  • What he is saying is that we cannot have a privileged correct description from an externalist perspective. The problem is the external perspective—the God's eye view. We are not outside of reality. We are part of it, in it. What is needed is not an externalist perspective, but an internalist perspective. It is a perspective that acknowledges that we are organisms functioning as part of reality and that it is impossible for us to ever stand outside it and take the stance of an observer with perfect knowledge, an observer with a God's eye point of view. But that does not mean that knowledge is impossible. We can know reality from the inside, on the basis of our being part of it. It is not the absolute perfect knowledge of the God's eye variety, but that kind of knowledge is logically impossible anyway. What is possible is knowledge of another kind: knowledge from a particular point of view, knowledge which includes the aware that it is from a particular point of view, and knowledge which grants that other points of view can be legitimate.



bottom of page