What do you think?
Rate this book
The Algebraic Mind: Integrating Connectionism and Cognitive Science
In The Algebraic Mind , Gary Marcus attempts to integrate two theories about how the mind works, one that says that the mind is a computer-like manipulator of symbols, and another that says that the mind is a large network of neurons working together in parallel. Resisting the conventional wisdom that says that if the mind is a large neural network it cannot simultaneously be a manipulator of symbols, Marcus outlines a variety of ways in which neural systems could be organized so as to manipulate symbols, and he shows why such systems are more likely to provide an adequate substrate for language and cognition than neural systems that are inconsistent with the manipulation of symbols. Concluding with a discussion of how a neurally realized system of symbol-manipulation could have evolved and how such a system could unfold developmentally within the womb, Marcus helps to set the future agenda of cognitive neuroscience.
242 pages, Paperback
First published February 19, 2001
About the author
Gary F. Marcus
14 books179 followersGary Marcus is an award-wining Professor of Psychology at New York University and director of the NYU Center for Child Language. He has written three books about the origins and nature of the human mind, including Kluge (2008, Houghton Mifflin/Faber), and The Birth of the Mind (Basic Books, 2004, translated into 6 languages). He is also the editor of The Norton Psychology Reader, and the author of numerous science publications in leading journals, such as Science, Nature, Cognition, and Psychological Science. He is also the editor of the Norton Psychology Reader and has frequently written articles for the general public, in forums such as Wired, Discover, The Wall Street Journal, and the New York Times.
Ratings & Reviews
Friends & Following
Create a free account to discover what your friends think of this book!
Community Reviews
Displaying 1 - 8 of 8 reviews
August 25, 2013
This was a really great book. While I am a psychologist, my specialty is in learning & memory, personality psychology, and psychobiology. I am, however, interested in cognitive science as well. This book compares two different ways of understanding the mind, neural networks/connectionism and the computational symbol-manipulation theory of mind. He argues that the best models are connectionist models which implement symbol manipulation.
I find myself persuaded by his arguments for the most part. He discusses a lot of the ideas in Rethinking Innateness: A Connectionist Perspective on Development and mentions Paul Churchland a few times as well. For a counter-point to this book, see Paul Churchland's more recent work Plato's Camera: How the Physical Brain Captures a Landscape of Abstract Universals . Marcus argues that we CAN think of cognition as being symbol manipulation with the caveat that it is implemented in sub-symbolic neural networks (the ACT-R architecture of cognition comes to mind, see http://act-r.psy.cmu.edu/ ) whereas Churchland is an "eliminative connectionist" and argues that our cognition is NOTHING like linguaformal symbols. Churchland is a fine philosopher, and though eliminativism is shocking at first, it does need to be taken seriously. That being said, I find that Marcus's book makes a compelling case for keeping symbol manipulation in our scientific tool kits for the time being.
Marcus also touches on the Nature-Nurture debate, although he goes into more detail in his later book The Birth of the Mind: How a Tiny Number of Genes Creates The Complexities of Human Thought . He also makes it clear at the end of the book that linguaformal symbol manipulation is NOT all there is to the mind. This book focuses on language primarily, and it is only language that he sees symbol manipulation as being especially important for. This is refreshing as a lot of cognitive scientists embrace a functionalist view of the mind where they see the whole thing as basically just a computer. See Consciousness: Creeping Up on the Hard Problem for a critique of functionalism as being a complete account of mind.
I find myself persuaded by his arguments for the most part. He discusses a lot of the ideas in Rethinking Innateness: A Connectionist Perspective on Development and mentions Paul Churchland a few times as well. For a counter-point to this book, see Paul Churchland's more recent work Plato's Camera: How the Physical Brain Captures a Landscape of Abstract Universals . Marcus argues that we CAN think of cognition as being symbol manipulation with the caveat that it is implemented in sub-symbolic neural networks (the ACT-R architecture of cognition comes to mind, see http://act-r.psy.cmu.edu/ ) whereas Churchland is an "eliminative connectionist" and argues that our cognition is NOTHING like linguaformal symbols. Churchland is a fine philosopher, and though eliminativism is shocking at first, it does need to be taken seriously. That being said, I find that Marcus's book makes a compelling case for keeping symbol manipulation in our scientific tool kits for the time being.
Marcus also touches on the Nature-Nurture debate, although he goes into more detail in his later book The Birth of the Mind: How a Tiny Number of Genes Creates The Complexities of Human Thought . He also makes it clear at the end of the book that linguaformal symbol manipulation is NOT all there is to the mind. This book focuses on language primarily, and it is only language that he sees symbol manipulation as being especially important for. This is refreshing as a lot of cognitive scientists embrace a functionalist view of the mind where they see the whole thing as basically just a computer. See Consciousness: Creeping Up on the Hard Problem for a critique of functionalism as being a complete account of mind.
April 23, 2022
An attempt to integrate two models of the brain: symbol manipulator and neural network.
Gary Marcus explores our understanding of how the mind works in terms of linguistics and symbol manipulation and compares it with a class of simple neural networks ("multi-level perceptrons"). Rather than continue the debate of "which model is right," Marcus tries to find ways to make them compatible. He highlights places where MLPs seem to be unable to reproduce human-like behavior. Then he discusses how adding some symbol-manipulating capacity could fix those issues, and ways to do that which don't seem to violate the spirit of these connectionist models.
I found this book fascinating, especially the way he thinks about embedding symbol-manipulation algorithms within the context of a neural network. Like Marcus, I suspect this is what the brain really does! Unfortunately, this book hasn't aged very well. It was published in 2003, and the state of the art in machine learning has advanced dramatically since then. Most notably, there are large sections of this book critiquing specific papers and models which feel quite dated. Those models are no longer the best examples to critique, as newer models are substantially more human-like in their performance (though, still lacking in particular ways). If anything, I see those sections as the start of a citation trail, to be followed by anyone who wants to see where the debate has gone since then. Despite this, I believe the theoretical discussions and the way of thinking presented in this book are still very relevant.
Whether you believe neural networks are the ultimate answer to ML, or whether you're troubled by the bizarre and inhuman mistakes they make, this book has some excellent food for thought.
Gary Marcus explores our understanding of how the mind works in terms of linguistics and symbol manipulation and compares it with a class of simple neural networks ("multi-level perceptrons"). Rather than continue the debate of "which model is right," Marcus tries to find ways to make them compatible. He highlights places where MLPs seem to be unable to reproduce human-like behavior. Then he discusses how adding some symbol-manipulating capacity could fix those issues, and ways to do that which don't seem to violate the spirit of these connectionist models.
I found this book fascinating, especially the way he thinks about embedding symbol-manipulation algorithms within the context of a neural network. Like Marcus, I suspect this is what the brain really does! Unfortunately, this book hasn't aged very well. It was published in 2003, and the state of the art in machine learning has advanced dramatically since then. Most notably, there are large sections of this book critiquing specific papers and models which feel quite dated. Those models are no longer the best examples to critique, as newer models are substantially more human-like in their performance (though, still lacking in particular ways). If anything, I see those sections as the start of a citation trail, to be followed by anyone who wants to see where the debate has gone since then. Despite this, I believe the theoretical discussions and the way of thinking presented in this book are still very relevant.
Whether you believe neural networks are the ultimate answer to ML, or whether you're troubled by the bizarre and inhuman mistakes they make, this book has some excellent food for thought.
March 13, 2020
Much of what Marcus wrote about neural networks in this book is dated -- deep networks can now perform tasks previously thought to be impossible, and perform them substantially better than Marcus' favorite symbolic manipulations and representations.
That being said, Marcus is a good writer, and much better suited to writing books than to tweeting criticisms about people who have actually solved problems in perception. In 2019 it seems that Marcus woke up from his slumber and decided that the only way for the "AI community" to become better is to listen to his criticisms of the advances in perception, even when those people are aware of the limitations of their methods to reasoning. The root of this problem in my opinion is the conflation of language understanding with intelligence (whatever intelligence means).
That being said, Marcus is a good writer, and much better suited to writing books than to tweeting criticisms about people who have actually solved problems in perception. In 2019 it seems that Marcus woke up from his slumber and decided that the only way for the "AI community" to become better is to listen to his criticisms of the advances in perception, even when those people are aware of the limitations of their methods to reasoning. The root of this problem in my opinion is the conflation of language understanding with intelligence (whatever intelligence means).
September 25, 2023
gary at his best
July 17, 2012
Pros: A lucid take on current computational neuroscience and the computational consequences of developmental programs which wire the brain. The last two chapters are very useful.
Cons: Even with the author's beautiful clarity of argument, I personally did not read many new hypotheses, but probably that was not what the author had set out to write.
Cons: Even with the author's beautiful clarity of argument, I personally did not read many new hypotheses, but probably that was not what the author had set out to write.
July 28, 2019
My biggest issues with this book were a) it felt like it was mostly critiquing something that science has moved beyond, and thus not dealing with the current questions surrounding AI models (at least that is my understanding of the science, and I recognize that is in part *because* of this book) b) too much of the book was dealing with technical details of different back-propagation models I didn't have the background knowledge to get much from. That being said, the book does do a good job of laying out certain distinctive features of human cognition any model would need to capture, and showing why it seems like we need to attribute some sort of symbol manipulation capacities to the brain. I wish I had a better understanding of bayesian theories and current AI developments to understand how they fit with his arguments. Various specific pieces of info about human cognitive capacities are fun, and I thought the stuff at the end about attributing symbols/representation to animals and how genes/biology need to be incorporated into discussion were both interesting. I think both of those sections, and the book as a whole, falls prey to a number of terminological ambiguities (specifically with level-explanations), and in many ways this book made me appreciate more the role of philosophy in cog sci debates (conceptual clarity is hard!).
March 23, 2021
Excellent book that exposes two initially distinct methods of cognition and and tries to bring them closer together.
November 8, 2012
Started reading this. Its a bit Dense. But it is somewhat Self Contained. I mean you can go through it, but it will most likely be time consuming. Good Philosophical arguments are presented along the way.
Displaying 1 - 8 of 8 reviews