NB. I was sent this book as a review copy.
From Princeton University Press
Science so often lives in silos, or perhaps more accurately silos of Babel. We sit in our offices, ignoring a great deal of what is happening down the corridor from us, let alone in the building next door or the one on the other side of the campus. A lot of this is down to inertia, and the academic culture of mistrust. But a great deal of the problem is down to the technical languages that, over the years, we become fluent in, and as we do so we shut out those other scientists who may actually be talking about something very similar, just in a complementary language.
The self-assembling brain is a book unlike any other that I’ve read, in that it is very explicitly trying to deal with this problem. The topic is that of the creation of the physical structures in the brain during, particularly, development. It asks the question how neurons wire up correctly given the very limited genetic information that encodes for the development of the brain. This is a fascinating question in itself, but the book asks how different scientific disciplines who may be interested in this question can complement each other.
The book is written as a series of 10 lectures, covering different aspects of the question, and in between is a dialogue between a neuroscientist, a robotics engineer, a developmental geneticist and an AI researcher who discuss the lectures, and uncover their own biases as well as areas of knowledge and interest. Unlike in the dialogues in, for instance, Hofstadter’s books (which themselves are wonderful), these dialogues are based on the real coming together of such scientists in an exercise for which the book is essentially a sort of conference proceedings.
The book deals with information theory, genetics, algorithmic growth, evolutionary programming, biochemistry, questions of scale, development and function, AI and more. And each one is done in a way that each of the scientists can grasp, being as they are, not experts in the particular realm, but who have their own foundation and language with which to construct each of the ideas in these areas.
I believe that particular right now, where we are in the midsts, or perhaps beginnings, of a massive acceleration in AI, and treading on the first steps of potentially AGI, that it is vital that scientists from these areas really can talk to each other to figure out what is going on and the potential dangers.
The topic of the brain is a clear one where there are many people from different areas interested in the question, but the same must be true of many areas of science, and beyond, where the spreading of ideas between disciplines can only yield fruitful outcomes.
The writing itself is very clear, and the lectures go deep enough into each topic that you come away feeling that your appetite has been whetted but wanting to know more, and there is a healthy bibliography for each which one is very tempted to dive into.
Overall I hugely enjoyed this book, tearing through it in a day spent in a cafe, with bewildered waiters around me wondering how someone can spend that long engrossed in one book. For anyone interested in the brain, or AI, or any of the myriad of branches and subbranches of each, I would highly recommend this!
Possibly the “dangers of AI” isn’t quite the right phrase? Or at least, not until AI can replicate itself in a physical form that it may be able to shield from human interference (perhaps even from human awareness).
How much of the book is concerned with the social setting (institutions, some particularly influential individuals, their policies and goals and means, the political currents around each particular case etc)? These things will influence not only what AI may be able to do but also where and how and how much it gets used.
Hi Henri – the answer is very little – it’s really more about the neurobiology than about the sociology.