For some artificial intelligence researchers, Minsky’s book is too far removed from hard science to be useful. For others, the high-level approach of The Society of Mind makes it a gold mine of ideas waiting to be implemented. The author, one of the undisputed fathers of the discipline of AI, sets out to provide an abstract model of how the human mind really works. His thesis is that our minds consist of a huge aggregation of tiny mini-minds or agents that have evolved to perform highly specific tasks. Most of these agents lack the attributes we think of as intelligence and are severely limited in their ability to intercommunicate. Yet rational thought, feeling, and purposeful action result from the interaction of these basic components. Minsky’s theory does not suggest a specific implementation for building intelligent machines. Still, this book may prove to be one of the most influential for the future of AI.
Customer Reviews
Understanding how the mind works:
This is a very unusual book. 270 chapters of one page grouped under 30 headings. Minsky tries to figure out how the mind works, by splitting what it does in one page very interesting bits. The discipline of one subject per page is unusual but effective. The brain is recognised as enormously complicated, but not so complicated that nothing can be understood about it. In fact it can only be understood by understanding its many different bits. Minsky built the first “randomly wired neural network learning machine” the SMRAC Such a type of machine has some kind of intelligence and some learning capabilities. Marvin Minsky is one of the pioneers of “artificial intelligence”. He found that to make these machines work intelligently it was useful to figure out how the brain solved the challenges of for example seeing in a useful way. To understand those brain processes Minsky delved into the workings of evolution.He found that you can learn a lot by figuring out how evolution in different steps increased brain capabilities. For example: why are we much more capable to remember faces of people rather than their names? Simply because vision is much older than language. Or why do people have strong egocentric tendencies? That is the result of a child having to learn how to survive. Why blind desire for prestige, money and sex? Our shared ancestry with chimpanzees etc
To me surprisingly the book contains many useful rules that can make you more effective. For example if you want to convince somebody it is better to use parallel than serial arguments. A serial argument is more vulnerable because if one link in the chain is weak you lose. Another example. “Virtually any problem will be easier to solve the more one learns about the context world in which that problem occurs”. Finally “Whatever happens, where or when, we’re prone to wonder who or what’s responsible?” That is why people are uncomfortable inside and outside organisations if they cannot find out who is responsible.
An unusual and very stimulating book.
a very interesting book:
This is a strange book, when i start read it i was astonished about the book organization. There are a lot of brief chapter that introduce the arguments. No one is full explained but it starts you thinking about that little part. At the end you have a full picture of reasoing as whole as a sum of little pictures
The emotional brain:
1. Study what seems least
2. Agents are the components of a system that give the system agency
3. Machines behave in a lifeless manner
4. People are not machines
5. The longer a conflict occurs with sub agents the weaker the agent is among competitor agents
6. Agents compete actively while interest is strong (If - Do, Condition - Response)
7. Agencies live in hierarchies and levels of administration reduce complexity and threshold pole the agents yielding a group positive or negative response.
8. High level agents control lower level agents
9. Pain reduces interest in long-term goals. Pain puts a focus on immediate problems (hunger and danger)
10. Pain repels and pleasure attracts
11. Limited exploitation creates agency: for example oscillation between work and sleep, limiting work in combination with anger, balancing hunger and pain with a focus on the problem of eating
12. Many questions are unanswerable
13. Unanswerable questions often create circular causality logic
14. Consciousness is doing and doing does not necessarily require understanding how the process works.
15. Cooperation requires complex interactions between agents. Competition is less complex and less productivity.
16. Learning Meaning has four types: Uniframe (several description combined into one), Accumulation (collections of samples, descriptions by experience, slow to make discovery by pattern match), Reformulation, and transformation.
17. Learning a new idea is possible as the individual accessing structures in the mind. Old structures provide a beginning reference and either are built upon or eventually bypassed in place of the new idea.
18. What we think is based on our spatial learning during youth about the world of space.
19. Accumulation: accumulation rarely feels satisfactory because we expect unity and disunity occurs in categorization, for example, birds fly, penguins are birds, penguins don’t fly. Rules are not perfect. Rules reflect that which is typical and describe the exceptions.
20. Reformulation: reformulation is finding new descriptions that make the problem easier. When we can’t solve a problem, we reformulate and seek escape by finding a new way to solve the problem.
21. Reformulation builds on what is already known
22. Reformulation connects things with goals in many different ways:
function-structure (Tables are for supporting things), end-means (If I want to reach higher, I can stand on a table), conclusion-premise (If you put something on a table, its height increases), effect-cause (I can reach higher because I start higher), and body-support (Tables hold things away from the floor).
23. The world of sensation: sensation -> reception -> recognition -> cognition
24. Each new idea must compete against a collection of skills associated with older ideas.
25. Memory: Conscience is concerned not with the present but with the past
26. Memory: Recall is not complete detail paradigm; instead, recall is done as memory fragments.
27. Memory: Whenever you answer a question without delay, it seem the answer was already in your mind.
28. Emotion opposites: fear-affection, attachment-dependancy, and hate-love.
29. Behavior can be modeled as sensors and effectors (sensor detect pattern and effectors can a systematic response)
30. Cross exclusion: An activity can suppress the activity of its competitor agent. Avalanche occurs when all agents equal compete for a resource. Agents must have a way to access resources cooperatively.
31. Each simple principle or mechanism must b controlled to operate within some limited range.
32. Memory: Create banks of memory and Associate problem agents with a separate memory banks and restrict each specialist agent to learn only while the goal is active. The agents could be interconnect and cooperative, however this is unlikely; and mostly likely they will be exploitive and competitive in acheiving their goals.
33. One gains from learning better ways to learn
34. When we violate standards we feel shame
35. In logic arguments are true or false. In real life arguments are strong or weak; we seek parallelism in our arguments as a reduncancy against failure; parallel reasoning is harder to break because there are more exists more than one way to answer a problem. We rarely need to know right or wrong and often prefer confrontation methods to fight out the best alternative.
36. The closest we can agree on meaning is in the expressions of mathematics.
37. Words: Language builds things in our minds. Words cannot be the substance of our thoughts. Instead, words control agents in our minds. It is the underlying emptiness of words that gives it potential versality. Language parts are divided into three categories: semantic, syntax, and grammer.
38. Words: Polynemes are k-line association domains for a word. A polyneme signals different agencies (color, shape, or texture agencies) too turn on process in their agences. Each agency must have a dictionary of words and memory to know how to respond to the polynemes, a bank composed of k-lines.
39. Context: Evidence is weighted threshold trigger of positive and negative values.
40. Context: Nemeic spiral: demand, inquire, vision, beliefs, social, language, shapes, touch, get, grasp, request, explain, touch, hearing, traits, physical, places, hearing, vision, move, and put.
41. Things: Whatever we may see or touch (nouns)
42. Differences: A discernible change by comparing two different things (verbs)
43. Cause: The cause of an action
44. Clause: A single phrase treated like a single word.
45. A gene expression is either off or on. Gene manufacture proteins and proteins produce specific chemicals. The cell contains many different types of proteins. Certain proteins move into the cell and serve messengers: 1. by altering other processes; change states of a specific gene 2. Certain protein combinations can turn genes on or off 3. Genes seem like small societies of agents. Certain cells emit specific chemicals and mobile cells follow the chemical scent.
46. How can genes build concepts in the mind? Genes only produce chemicals and how can a chemical create a concept. Genes determine the architecture of agencies destined to learn particular kinds of processes.
Cornucopia of Ideas:
This book is different from other books on many different levels.
For instance, its organization parallels its subject. The main theme of the book is that the mind consists of a network of non-intelligent agents organized into more and more highly complex agencies. - That the function of these agencies taken together is to perform the mind (mind is thought of as a process and not a thing - the same way that in chemistry fire was eventually recognized as being a process of oxydizing and not an element). In the same way, each page of the book has a single concrete point to make or an idea to present. These ideas reference each other in a networked fashion. As a result, the meaning of the book emerges as you consider the networks of these ideas. This is a very aesthetically pleasing concept for organizing a book as it is informative.
The book introduces a lot of different ideas to its readers - such as the currencies that the mental agencies use to measure the importance of various tasks or views. It talks about the educational value of humor - in a way I never considered. It talks about organization of memories.
One thing that I thought was oversimplified was Marvin Minsky’s description of how genes affect the development of the brain. He seems to treat the subject as though the environment has nothing to do with it. For example, we know that mother’s consumption of alcohol can drastically affect the development of the fetal brain. What mother eats during pregnancy and even before can also affect the development of the brain. For example, consumption of Omega-3 fats seems to have a very strong effect on the brain development. The brain change is also directed by the experiences that the person has. Vision will not develop in an infant that has not been exposed to light in its first weeks of life. I think that this is a rather weak point of the book.
makes you think about the process of thought:
The Society of Mind attempts to explain how the mind works. The author considers the mind to be a society of small mental machines that do simple things by themselves but combine to perform amazingly complex tasks like the walking and talking that we take for granted. As this book was written in the 1980’s I am sure that it is somewhat out of date. But the questions he asks are timeless. How does memory work? How do we sense space? How do we process conflicting ideas? How do we know when to replace a memory with a more accurate version? This book will make you think about these questions and more. The Society of Mind is worth reading just for the questions it asks.
