I would recommend this book to everyone interested in AI. After all, it is about something more general. About the I, no matter if it comes with an A or not. It is about the common problems that lead to the development of that I, and how these are shared by biological and artificial systems. 5 ⭐️!!

A truly refreshing read that humbles the field of AI while putting it into a broader perspective.

From the basics of sensory input, valence, and motility Bennet takes us into neurochemistry, reinforcement learning, and credit assignment problems, showing that what we consider cutting edge research in machine learning is often something that evolution solved over one hundred million years ago.

The book is filled with clever experiments that separate the intelligence of worms, fish, & us, while showing that we still have much in common. When do we start recognizing patterns, & why? Why do the mechanisms that allow us to learn can lead to addiction? Why is imagination a key to intelligence?

Do you ever feel like running away when you are in an uncomfortable situation? Do you like dancing when things go your way? Well, so does this little worm.

These innovations start with deciding a direction of motion in a multicellular organism. One of the things I found refreshing, is that Bennet explains these mechanisms using experiments on c. elegans, a microscopic worm with about 300 or neurons, which can be used to understand primitive emotions.

In this theory, intelligence emerges in animals because, unlike plants who produce their own food, or fungi who eat what is dead, animals roam the environment searching for something to and eat. Animals grew intelligent because we started in the worse ecological niche & had to innovate to survive.

A Brief History of Intelligence assembles the many mechanisms that contribute to intelligence in chronological order. This is not just clever structure. By explaining the problem each mechanism solves you get a deep understanding of why intelligence emerged and how it differs among species.

Thanks!

This is one of the most fun papers that I have worked in a long time. If you want to geek out with the mathematical foundations of economic complexity theory, please give it a read.

We also used this opportunity to embed this model into a short-run equilibrium framework showing that prices increase concavely with the complexity of a product and that economies should converge to the wages of others with similar levels of complexity.

This allows us to easily generate output matrices as large as those observed in the empirical literature and calculate their ECI.

Our main model is a generalization of Kremer’s O-Ring model with factors that can be made specific to each economy and activity. In this model, the output of an economy in an activity is proportional to the probability that it is not missing the factors that the activity requires.

In the product space, the core-periphery structure tells us that capabilities are correlated (e.g. Singapore scores high across all capabilities & Mali scores low). In the case of the research space, the ring structure is given by the fact that each field is only similar to a few other fields.

We also show that our model explains differences in network structure, like those observed between the product space (a network with a core and a periphery) and the research space (which is shaped as a ring).

The idea that you can measure the presence of economic factors—no matter how they are defined--is an interesting departure. It provides a basis for the use of complexity measures in development. In the paper, we show this property is robust to noise & applies to other production functions.

In a single-factor model, ECI separates economies that are better or worse endowed with the factor. In a model with many factors, ECI tells us which economies have a higher probability of being endowed with many of them--regardless of what these factors are.