What do you think?
Rate this book
Design for a Brain; the Origin of Adaptive Behavior
This work has been selected by scholars as being culturally important, and is part of the knowledge base of civilization as we know it. This work is in the "public domain in the United States of America, and possibly other nations. Within the United States, you may freely copy and distribute this work, as no entity (individual or corporate) has a copyright on the body of the work. Scholars believe, and we concur, that this work is important enough to be preserved, reproduced, and made generally available to the public. We appreciate your support of the preservation process, and thank you for being an important part of keeping this knowledge alive and relevant.
298 pages, Hardcover
First published January 1, 1952
14 people are currently reading
386 people want to read
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 9 of 9 reviews
August 8, 2014
The centrepiece of Ashby’s Design for a Brain is undoubtedly the portrayal of the Homeostat, which is quite possibly the strangest machine ever made. It was built in 1948 from four surplus RAF bomb control switch-gear kits, each one refashioned into an electro-mechanical artificial neuron. The machine was presented to a baffled audience at the ninth Macy Conference on Cybernetics in 1952, where cyberneticist Julian Bigelow famously declared that, "It may be a beautiful replica of something, but heaven knows what."
The Homeostat has been described as a goal-driven machine without any goals. It has been described as a learning machine, but it is hard to fathom what it learns because the Homeostat embodies not only Ashby’s design for a brain, but also a design for the self-contained universe, or environment, that it inhabits. More confusing still, where the organism ends and the environment begins is purely in the eye of the beholder and is quite irrelevant to the operation of the Homeostat; both brain and anti-brain in a box. If the machine is disturbed by poking one of the gently oscillating needles on top of the bomb-controllers, this causes a flurry of random activity until it hits upon a combination that seems to calm it down. W. Grey Walter, a contemporary of Ross Ashby, and inventor of some of the very first autonomous robots, likened the Homeostat to a "fireside cat or dog which only stirs when disturbed, and then methodically finds a comfortable position and goes to sleep again”, leading him to supply the Homeostat with the moniker of “Machina sopora”.
Nevertheless, the Homeostat is a work of pure genius; one of the last great dinosaurs of an analogue world before Turing’s digital computer took over the world. Post-cognitive scientists today are only just coming to appreciate how the Homeostat creates its own umwelt through layers of couplings with its environment. This is Ashby’s discovery of a new kind of stability, called ‘ultrastability’, which makes its debut in Chapter 7.
This alone would be enough to secure Ashby’s place in the pantheon of cybernetics, but then he goes and drops another bombshell in Chapter 16. He applies his own ‘law of requisite variety’ to his machine - any system can adapt as far as it fits its environment - but no more. Once it hits this barrier then evolution will ruthlessly start to carve up a brain into independent subsystems that communicate with each other through the environment rather than directly with each other, matching one-for-one the constraints in its environment. With this astonishing idea Ashby is suggesting that advanced brains create barriers between internal subsystems to prevent them interfering with each other. The resulting ‘multistable’ brain is demonstrably more efficient than its richly interconnected counterpart. In the case of the richness of neural connectivity, neuroscience is only now re-discovering the fact that less is more.
It’s a great thing that more people are now re-discovering the work of Ashby and other early cyberneticists after being ‘lost’ for many years in the wake of the dominance of the brain-as-computer metaphor in Artificial Intelligence. Analogue is the new Digital.
The Homeostat has been described as a goal-driven machine without any goals. It has been described as a learning machine, but it is hard to fathom what it learns because the Homeostat embodies not only Ashby’s design for a brain, but also a design for the self-contained universe, or environment, that it inhabits. More confusing still, where the organism ends and the environment begins is purely in the eye of the beholder and is quite irrelevant to the operation of the Homeostat; both brain and anti-brain in a box. If the machine is disturbed by poking one of the gently oscillating needles on top of the bomb-controllers, this causes a flurry of random activity until it hits upon a combination that seems to calm it down. W. Grey Walter, a contemporary of Ross Ashby, and inventor of some of the very first autonomous robots, likened the Homeostat to a "fireside cat or dog which only stirs when disturbed, and then methodically finds a comfortable position and goes to sleep again”, leading him to supply the Homeostat with the moniker of “Machina sopora”.
Nevertheless, the Homeostat is a work of pure genius; one of the last great dinosaurs of an analogue world before Turing’s digital computer took over the world. Post-cognitive scientists today are only just coming to appreciate how the Homeostat creates its own umwelt through layers of couplings with its environment. This is Ashby’s discovery of a new kind of stability, called ‘ultrastability’, which makes its debut in Chapter 7.
This alone would be enough to secure Ashby’s place in the pantheon of cybernetics, but then he goes and drops another bombshell in Chapter 16. He applies his own ‘law of requisite variety’ to his machine - any system can adapt as far as it fits its environment - but no more. Once it hits this barrier then evolution will ruthlessly start to carve up a brain into independent subsystems that communicate with each other through the environment rather than directly with each other, matching one-for-one the constraints in its environment. With this astonishing idea Ashby is suggesting that advanced brains create barriers between internal subsystems to prevent them interfering with each other. The resulting ‘multistable’ brain is demonstrably more efficient than its richly interconnected counterpart. In the case of the richness of neural connectivity, neuroscience is only now re-discovering the fact that less is more.
It’s a great thing that more people are now re-discovering the work of Ashby and other early cyberneticists after being ‘lost’ for many years in the wake of the dominance of the brain-as-computer metaphor in Artificial Intelligence. Analogue is the new Digital.
February 10, 2016
It was hard to decide between a 4 and a 5 on this book. I think when it was written it was a totally awesome book, groundbreaking and visionary. After 60 years, it could use a little updating. Essentially, it proposes that random internal changes to a mechanism can explain adaptation/learning in animals (and of course machines). Now the ideas of dynamical systems and adaptive control are common place. The book could be reduced to a short paper that references common knowledge and proposes the random step function as a mechanism of adaptation.
I guess my greatest disappointment is not with the book, but with the field. Why haven't we developed this into a full-blown theory of cognition?
Regarding updates: There are several mechanisms in neural systems that go beyond the random step function. First, we have ways of assigning credit to unexpected events, and making select connections more plastic. Second, changes in connection strength are specifically directional, based on activity history. The combination of these provide for learning much more efficient than random search.
There are many regulatory loops across a range of scales in the brain. My intuition is that describing these mathematically, in terms of a large dynamical machine like Ashby envisioned, is the key to understanding the human mind.
I guess my greatest disappointment is not with the book, but with the field. Why haven't we developed this into a full-blown theory of cognition?
Regarding updates: There are several mechanisms in neural systems that go beyond the random step function. First, we have ways of assigning credit to unexpected events, and making select connections more plastic. Second, changes in connection strength are specifically directional, based on activity history. The combination of these provide for learning much more efficient than random search.
There are many regulatory loops across a range of scales in the brain. My intuition is that describing these mathematically, in terms of a large dynamical machine like Ashby envisioned, is the key to understanding the human mind.
April 8, 2007
Ashby was a giant and I think this book is out on the Internet in PDF form. Ashby brought Cybernetics to the UK.
November 16, 2024
A forgotten classic of cognitive science from one of the early Cyberneticists.
W. Ross Ashby was a contemporary of Norbert Wiener, whose research group coined the term "cybernetics". Ashby was also a contemporary of legendary early roboticist, W. Grey Walter who built some of the earliest autonomous robots to exhibit life-like behavior.
This book is Ashby's account of his attempt to understand what kind of a machine can learn. He starts with the obvious fact that a human brain is a machine that can learn and, from there, he tries to reverse engineer the requirements of a learning machine. Neuroscience was at a much more primitive state at the time, so Ashby largely treats the biological brain as a black box.
To test his theories, Ashby constructs a working electro-mechanical learning machine that he calls a homeostat that exhibits some of the same learning abilities as the human brain but on a much smaller scale.
AI since Ashby's time has come a long way but is continually being sidetracked by the "next big thing". At one time it was expert systems, then neural networks, then fuzzy logic. Each was identified as a key component of intelligent systems, was obsessively focused on and labelled as "AI" for marketing purposes. Eventually, in each case, researchers finally moved on to continue their research as it became clear that AI was more than just one clever algorithm like fuzzy logic. We eventually gained the new term AGI in an attempt to distinguish "real AI" from marketing jargon AI. (I actually owned a product or two that used the "AI" moniker thanks to fuzzy logic).
Today it's LLMs, Large Language Models, that are the new marketing jargon to claim something has AI. But we'll probably move on again in a few years. At the height of AI's newest buzzword bubble, it's refreshing to get back to the basics of what intelligence and learning actually is. Highly recommended for anyone interesting in cognitive science or artificial intelligence.
W. Ross Ashby was a contemporary of Norbert Wiener, whose research group coined the term "cybernetics". Ashby was also a contemporary of legendary early roboticist, W. Grey Walter who built some of the earliest autonomous robots to exhibit life-like behavior.
This book is Ashby's account of his attempt to understand what kind of a machine can learn. He starts with the obvious fact that a human brain is a machine that can learn and, from there, he tries to reverse engineer the requirements of a learning machine. Neuroscience was at a much more primitive state at the time, so Ashby largely treats the biological brain as a black box.
To test his theories, Ashby constructs a working electro-mechanical learning machine that he calls a homeostat that exhibits some of the same learning abilities as the human brain but on a much smaller scale.
AI since Ashby's time has come a long way but is continually being sidetracked by the "next big thing". At one time it was expert systems, then neural networks, then fuzzy logic. Each was identified as a key component of intelligent systems, was obsessively focused on and labelled as "AI" for marketing purposes. Eventually, in each case, researchers finally moved on to continue their research as it became clear that AI was more than just one clever algorithm like fuzzy logic. We eventually gained the new term AGI in an attempt to distinguish "real AI" from marketing jargon AI. (I actually owned a product or two that used the "AI" moniker thanks to fuzzy logic).
Today it's LLMs, Large Language Models, that are the new marketing jargon to claim something has AI. But we'll probably move on again in a few years. At the height of AI's newest buzzword bubble, it's refreshing to get back to the basics of what intelligence and learning actually is. Highly recommended for anyone interesting in cognitive science or artificial intelligence.
May 29, 2025
Design for a Brain is a follow-up work of An Introduction to Cybernetics by renown cyberneticist William Ross Ashby. The key to setting the right expectations before reading this book is to notice its subtitle: The origin of adaptive behaviour.
Originally published in 1952, Design for a Brain is not a treatise on all cerebral mechanisms but a proposed solution of a specific problem: the origin of the nervous system's unique ability to produce adaptive behaviour. It is based on the fact that the nervous systems behaves adaptively and the hypothesis that is essentially mechanistic in nature.
Ashby develops definitions and introduces us to the required terminology along the way. I found the book surprisingly easy to follow, especially from the beginning, as Ashby wanted to keep this work as a coherent unit. Nevertheless, those familiar with An Introduction to Cybernetics will recognize Ashby's clarity of style and pedagogical approach.
Concepts such as dynamic systems, stability, adaptation, the ultrastable system and its meaning in the organism as well as iterative systems are some of the core ideas presented in this book. Along the way, Ashby makes use of the homeostat machine as an artifact from which the aforementioned concept can be illustrated.
Lastly, the book contains an appendix in the form of 4 chapters where most of the concepts are defined and developed with mathematical rigor. So this makes the first part of the book readily accessible to the general reader, as no mathematical jargon clouds the exposition of ideas in bulk of the text.
All in all, this is a great introductory work to the concepts of dynamic systems and adaptation. All the more due to its emphasis on the holistic approach that characterizes the cyberneticists.
Originally published in 1952, Design for a Brain is not a treatise on all cerebral mechanisms but a proposed solution of a specific problem: the origin of the nervous system's unique ability to produce adaptive behaviour. It is based on the fact that the nervous systems behaves adaptively and the hypothesis that is essentially mechanistic in nature.
Ashby develops definitions and introduces us to the required terminology along the way. I found the book surprisingly easy to follow, especially from the beginning, as Ashby wanted to keep this work as a coherent unit. Nevertheless, those familiar with An Introduction to Cybernetics will recognize Ashby's clarity of style and pedagogical approach.
Concepts such as dynamic systems, stability, adaptation, the ultrastable system and its meaning in the organism as well as iterative systems are some of the core ideas presented in this book. Along the way, Ashby makes use of the homeostat machine as an artifact from which the aforementioned concept can be illustrated.
Lastly, the book contains an appendix in the form of 4 chapters where most of the concepts are defined and developed with mathematical rigor. So this makes the first part of the book readily accessible to the general reader, as no mathematical jargon clouds the exposition of ideas in bulk of the text.
All in all, this is a great introductory work to the concepts of dynamic systems and adaptation. All the more due to its emphasis on the holistic approach that characterizes the cyberneticists.
August 12, 2019
Obra muy interesante que intenta modelar un sistema ultraestable u homeopático.
February 2, 2022
Groundbreaking and remains in the lead in so many ways.
February 6, 2024
This is an amazing book. I’ve just completed a PhD in robotics and just about everything I encountered is here in this 1966 edition in embryonic form: embodiment, dynamical systems, complexity, frame of reference problem, causal networks, self-organisation, sparse networks, learning… the list just goes on and on. Ashby treats the whole of the problem of adaptive life (environment plus animal) rather than a bunch of isolated systems (perception, control…).
This holistic view is extremely rare in an over-specialised field. Those interested in AGI would be well advised to start with this book, rather than trying to follow the daily flood of papers dealing with this or that tweak to some rinky dink neural network architecture.
Ashby’s master stroke is to put all the maths into the appendix. The first benefit of this is accessibility: the patient reader with high school maths can follow the arguments without seeing a single equation. The second benefit is the way it has forced the author to explain the logic as clearly as possible. So many of today’s papers use maths as a way to avoid giving written explanations, and leave the readers scratching their heads. Ashby’s book is an absolute model of clarity.
Ripe for rediscovery - long live cybernetics!
This holistic view is extremely rare in an over-specialised field. Those interested in AGI would be well advised to start with this book, rather than trying to follow the daily flood of papers dealing with this or that tweak to some rinky dink neural network architecture.
Ashby’s master stroke is to put all the maths into the appendix. The first benefit of this is accessibility: the patient reader with high school maths can follow the arguments without seeing a single equation. The second benefit is the way it has forced the author to explain the logic as clearly as possible. So many of today’s papers use maths as a way to avoid giving written explanations, and leave the readers scratching their heads. Ashby’s book is an absolute model of clarity.
Ripe for rediscovery - long live cybernetics!
January 6, 2015
A classic work of genius!
Displaying 1 - 9 of 9 reviews