Most Helpful Customer Reviews
|
|
2 of 2 people found the following review helpful:
5.0 out of 5 stars
Accessible, clear, entertaining, and thought-provoking, 21 May 2007
Haikonen outlines the issues surrounding machine consciousness, and develops some theoretical and practical ideas towards realising the goal of cognition and consciousness in machines - initially via discussions and surveys of the philosophical and neural network aspects of the problem; towards the end of the book, he offers some tantalising glimpses into how to implement the project. (These "implementation themes" form the main topics in Haikonen's 2007 book, "Robot Brains: Circuits and Systems for Conscious Machines").
Haikonen seems committed to a representationalist cognitivism - i.e. the view that cognition deals in representations as it's main "contents" - though the notion of what inner representations are (and the attendant debate surrounding their philosophical status) is left mostly unanalysed. Late on in the book, he seems to go for a notion of representation as a dynamic pattern of neuron excitation, a plausible computational neuroscience approach (cf. Churchland & Sejnowski "The Computational Brain").
On sensation, the familiar philosophical issues - about what exactly these "sensations" are - are avoided in favour of a "standard neurobiology" approach (nociceptors, haptic sensors, images on a retina etc). Similarly, the account of what a "percept" would be - how to individuate, describe etc is left for the non-engineer to think about further.
There's an interesting and quite plausible analysis of pain qua subjectively felt pain - for Haikonen, pain is a system reaction, rather than a sensation ascribed to outside-body perceptions. Also his system-account of the emotions and motivation is quite modern (though again there are issues of identification and individuation here, given the complexity of the brain qua neural network and the body in it's tight integration with the brain).
In Chapter 7 "language and thought"... Haikonen shows he understands the inherent weaknesses of Chomskian "universal grammar" - e.g. biological implausibility, and the omission (via the failed attempt to separate semantics away from syntax) of "meaning as motive" for communication. His practical linguistic model, as he notes, is quite similar to that of Deacon ("The Symbolic Species"). Haikonen is also sympathetic to the Cognitive Linguistic axis as propounded by Langacker et al (see e.g Ronald Langacker "Concept, Image, and Symbol" on the Cora language), especially to the view there is no such thing as pure syntax, that all grammar and syntax always have both semantic, and cognitive, aspects.
Also interesting is Haikonen's emphasis on the importance of cross-modal linking in cognition, and on making language to some extent part of "conceptually based perception".
Haikonen notes in passing the power of language as a compression/de-compression tool for associations - this ability is a main factor in our species' "cognitive power-play" compared with other species. (See e.g. Hurley & Nudds (eds) "Rational Animals").
The chapter on speech is interesting, both as a beginner's guide to connectionist models of comprehension and syntax, and through considering the issue of speech production - connectionism, and other axes in linguistics (e.g. the Chomskian-Pinkerian), have concentrated too much on the perception-comprehension-acquisition side of cognition and language, and not enough on the motor-speech production side; whereas Haikonen realises that speech is an important "action-tool" which we use to create the environment according to our desires, including of course getting other people to do things for us - and that, therefore, any cognitive model has to have a stab at both wanting to speak and at producing (to some extent) syntactical speech.
Chapter 8 "consciousness", is a selected survey of theories in this area of philosophy. I'm biased here, against the starting point of so many thinkers in this area, viz. the notion that there is some non-relational "quale" as a property of experience, and the notion that self-consciousness is somehow the "most-privileged" kind of consciousness. Whereas it seems to me, trivially, that all animals are conscious (and - more controversially - that even plants or cells may be conscious in some meaningful sense). On the other hand, Haikonen's practical approach to machine consciousness focuses on the feedback, motor- and process-aspects of consciousness, rather than getting bogged down in Journal of Consciousness Studies' interminable debates, so this "survey"-chapter serves mainly as "useful background" to Haikonen's constructive project.
Haikonen's specifications for the cognitive machine perhaps involve more "pre-building" than our own neonate brains endow us with. However, with machine building we have an advantage over evolution, in that we know roughly what our goals are in future, so this is ok in the context. The motor impact and input of/to learning can be taken even further with machines and virtual machines - cf. Phillipona, O'Regan, Nadal paper 2003 "Is there something out there? Inferring space from sensorimotor dependencies" in Neural Computation, 15(9), on a mathematical proof for how a brain with an environment, sensory receptors and proprioception learns by doing that the world is n-dimensional.
Haikonen's approach is quite modular and functionalist, yet connectionist and "bottom-up" - give the machine a neural network with various functions, together with some interconnections between the modules to enable learning and e.g. cross-modal perception. If this approach can be implemented then it will certainly help us see the possibilities for machine-consciousness.
On machine consciousness in general, Haikonen is definitely on the right track. Consciousness is process-based, arises from the material spatial world we all live in, and essentially involves an organism's thinking, acting, and teleological goals, whether that organism is carbon-, silicon- or virtual-PDP-constructed. As with animals, cognition and perception do not have as their contents the physical carriers which instantiate and process information: we humans don't see changes in rhodopsin molecules in the retina, we don't observe the firing thresholds of our neurons - and neither does a cognitive machine. Rather, through crossmodal and intra-modal distributed connections (cf. papers in Spence & Driver (eds) "Crossmodal space and crossmodal attention"), we and the robots perceive and act on that meaningful world which our perception and actions also essentially constitute, and the self-referencing nature of this enterprise is what makes us (and our environment) dynamical, non-linear systems.
|
|
|
5.0 out of 5 stars
Easily readable and very informative, 10 May 2009
I'm going to keep this review short and sweet...
Haikonen provides an excellent cross section into the implementation of cognitive systems using biological examples as a foothold. Although written in three seperate sections, the first is really an introduction into AI and motives for its development. The second section looks to approach cognition by looking at our own biology/psychology with the third discussing practical implementations of such systems.
In comparison to some other books I've read on the subject, it's highly enjoyable and certainly whets the reader's appertite for further study of this highly complex field.
It's not a large book and I managed to get through it in just one weekend and I certainly benifited from doing so.
|
|
|
|