1. We've heard that you're working on the sequel to The Society of the Mind, The Emotion Machine, can you briefly describe what this book is about, and what you wish to convey?
The central idea is that emotion is not different from thinking. Instead, each emotion is a type or arrangement of thinking. There is no such thing as unemotional thinking, because there always must be a selection of goals, and a selection of resources for achieving them.
2. Is consciousness possible without emotions?
Consciousness is not a 'thing'; it is a just a word that we carelessly use for several different techniques. Their common feature is that in each type of consciousness, some parts of the brain describe what has recently happened in various other parts of the brain. The most common forms of this are recognized when those descriptions take the form of linguistic expressions or visual representations-but there are other ways for parts of the brain to represent what has happened in yet other parts of the brain. However, because we cannot describe all of these in the usual ways, then,
paradoxically, we are 'unconscious' of some forms of consciousness! I'd say that this has not much to do with emotions, except that each emotion engages different mental resources, and hence different forms of descriptions.
3. How would machine emotion be generated? (if these questions aren't already answered: What type of architecture do you think would allow for machine emotion? What type of architecture would allow for consciousness? Are they the same? or different?
The human architectures for these things are highly evolved and very complicated. There is nothing simple or basic about human emotions. In the new book, I will describe the architectures for several emotions-and each of them involves dozens of different resources. As for consciousness, that requires an architecture in which there are systems for making
descriptions of the contents of various types of short-term-memories. Again, we probably have many types of those memories, because almost every brain center has several short term memory systems.
4. Are there any types of emotion ? Is emotion boolean ? or not?
In infancy, the modes of thinking that we call emotions are rather separate and tend to be mutually exclusive. By the end of the first year, they overlap more, and the child shows mixtures of them.
5. Would the emotion that machines could be capable of, allow them to deserve preferential treatment (is a human no better than a machine in the future)?
Future machines will be capable of all sorts of emotions, and eventually they will invent new ones whenever this is found useful for solving different kinds of problems.
6. Could emotion/consciousness be evolved on a digital computer?
Certainly, because a digital computer can simulate anything-and simulated thinking is the same as thinking!
7. Would emotion allow machines to survive longer if they were presented any great threat (i.e. if someone is trying to kill you, you get angry and try to prevent them from doing so for your survival).
Absolutely. All the architectural systems that we use for emotions have evolved for accomplishing purposes. If the machine has any goal to survive, it will try to find ways to accomplish that-and it will help for it to have some that are built-in from the beginning. Also, any machine that needs a long time to solve a hard problem is likely to realize that
it must survive long enough to solve it!
8. Why or why not would our machines require emotion?
It will need some kinds of emotion for selecting its goals. If it has only one kind of goal, and therefore needs only one way to think, then that will be its only emotion.
9. In 1969 Papert and yourself published "Perceptrons", a paper detailing how perceptrons could not solve the Parity Problem and other simple pattern classification tasks. In 1989, you published the second edition of your paper - how had your opinions of perceptrons changed over those years?
In the early 1960s, we already realized that different kinds of problems might require appropriately different architectures. That is why we argued that no single, simple architecture would work very well. Certainly the theorems in that book remain true, and most of them apply equally well to loop-free machines with multiple layers. Generally, only the exponents of their exponential growth change. It is not true that new techniques such as back-propagation change our conclusions, because most of the theorems were about what loop free nets could (and could not) compute with ANY sets of coefficients-so it was never a matter of learning at all. Generally, the neural-network community has now silently recognized that more complex architectures are needed for harder problems, but they don't like much to talk about this.
