A good job of explaining some of the power of some modern neural network models. By the time we reach Chapter 6 the limits of these network models begin to intrude on Churchland's enthusiasm. He predicts that even the complexities of human language will eventually fall to a sophisticated syntactic analysis made possible by neural network models. Well, at least we can all agree when he suggests that this future success for AI researchers "will be some time in working itself out". No kidding. The real issue is: what will the AI researchers have to add to their current models to clear the hurdle of language? Churchland is silent.

Early in the book, Churchland informs us that "we have reached the present stage of functional success in our [current] artificial [neural] networks by trying to be as faithful to the neural reality as presently we can be". Ha. Ha, ha, ha. What a gem. None of the models he describes uses a realistic neural network model. Period. Churchland pauses and mutters, "It is an empirical question whether deeper faithfulness [to biological reality] will yield yet deeper success [in AI neural network modeling]. At the very least, it is a question richly worth pursuing." But not in this book.

Given the title of the book, you might imagine that this would be a book about brains. Wrong. The first part of this book is about a computer modeling technique that is able to mimic some of the things a brain can do at the functional level of pattern recognition. The second part of the book is what Churchland is really interested in, philosophical issues like the basis of human morality. About in the middle of these two topics is Churchland's closest approach to biology. In Chapter 7 we witness his tip of the hat to synaptic neurotransmission and neurotransmitters. With Prozac described as an effective regulator of neurotransmission and human behavior Churchland kicks all bystanders off of Freud's couch and down to the local pharmacy. On the way out the door from this brief visit to the world of biology, he admits, "We cannot socialize people just by admineristering drugs." and adds, "A deeper understanding of how biological neural networks do their work will help...." Hmm....that sounds like an interesting topic...maybe in the next chapter there will be some more biology? But no. That's it. Enough. Back to philosophy.

But all is not lost! The start of Chapter 8 is very sweet. Churchland compares the apparent inpenetrability of the study of consciousness to three historical examples of problems that once seemed impossible to solve:

1) identification of what keeps the planets in place (gravity),
2) identification of the nature of "star stuff" (the same as "Earth stuff")
3) identification of the physical basis of life (complex macromolecules, cells, genetic programs)

I am interested in the possibility that the fundamental difference between the neural network models of AI and the biological neural networks in your brain is heavily related to the issue of memory and how memory makes learning possible. Short-term memory gets mentioned in Churchland's list of key aspects of consciousness (since we obviously have a cognitive grasp of recent events). What about long-term memory and everything in between short- and long-term memory? Would we be anything more than simple perceptual automatons without long-term memories? Surely the full human consciousness that we are interested in includes the key aspect that we are likely to remember things that have been in our consciousness. People like to ask "what is the function of consciousness?" For me, consciousness is closely related to the filtering of sensory inputs (or our memory-generated internal world of thoughts) that centers our awareness on limited, understandable patterns occuring in our neural activity. This filtering is Arbib's fundamental issue of resolving redundancies of potential command. The brain filters and allows us to take meaningful action on a few simple issues at any one time. For brains, "take meaningful action" can take two major forms: motor output and memory storage. Those items that enter our awareness and move to the center of our consciousness are the neural activity patterns that are allowed to resonate, maybe trigger motor responses, and maybe modify the synapses involved so as to construct new memories and allow for learning. And just which neural activity patterns are allowed to resonate? Clearly, our long-term memories have an important role to play in our selective awareness and our streams of consciousness. Biological brains are structured so that memories are built into our sensory filters and so the use of our old memories influences the production of our new memories. Thus, if our consciousness depends on "training" our neural networks through interactions with a complex environment, then our consciousness depends on our memory systems. The whole point of our memory systems is to adaptively enhance our sensory filtering and our capacity to construct internal models of reality. Whether we are conscious of our physical surroundings or lost in deep thought, our memories guide the stream of consciousness and the selected information that resonates in our consciousness has a good chance of being stored in new memories. For me, memory is fundamental.

I never hoped to defend Dan Dennett's
homo-centric view of the importance of human language in human consciousness, but I think there is ground for compromise between Churchland's and Dennett's views. As I mentioned above, consciousness is very much centered on a restricted subset of the brain's on-going neural activity. I would be willing to say that when we concentrate and center our awarenes we have a nearly serial stream of consciousness ("nearly serial" as compared to the vast parallel torent of information flowing through the whole brain at any one time). I view language as a way of converting such"nearly serial" streams of consciousness into specific motor output (speech) that can be heard by another person and efficiently direct their consciousness. My view is that language is important for our current "normal" consciousness, but I do not put language at the center of consciousness, except for the case of certain amazing philosophers like Dan Dennett who may well spend their entire existance in an uninterupted linguistic stream of consciousness.

So, this book is good at what it does, but a dissapointment to those who are interested in the details of biological intelligence.

Go to John's Book Page.

Go to John's Home Page.