For fifty years what is now known as classical AI research and development has striven towards the goal of the thinking machine and to date the results have been disappointing. Although we have many systems which show some of the basic traits we associate with intelligence they cannot even remotely be classed as thinking. One of the successes of AI is the Industrial self-guiding vehicle, used for internal factory transport without tracks or lines/cables to follow. Such systems have been used to some degree but in the end it is the lack of Robustness which lets them down, resulting in a vehicle stuck in some position and unable to find a way out, often on a weekly basis.
This robustness seems to be something which the standard "top down" approach to design in all fields is unable to resolve. Instead it has been found that the "bottom up" approach with interaction with its environment is producing more promising results. This may be simply that, however good an algorithim is, it is still a man-made model and, as such cannot take into account those circumstances and conditions of which the designer is unaware.

Some of the reasons for this disappointing result are :-

a) There is no complete definition or understanding of Intelligence - how can you design it if you don't know what it is - even the much used IQ measurement of intelligence has been seen to be woefully incomplete and some would say dangerously flawed.

b) The goals set for AI were too high too soon due to apparent initial successes (and the need for further funding) - this caused poor results and consequential funding difficulties - so today classical AI research is mostly aimed at specific marketable applications such as medical diagnostic systems which have greatly improved but still do not think.

c) The model of how the brain works was wrong. It is becoming increasingly clear that a brain does not resemble our digital computers in organisation or function and that the best we can do with existing technology is to create an artificial cyberspace in a computer. Perhaps within this artificial cyberspace we can work on the real solution.

d) Tied in with all of these is the unsuitable nature of the "Top down" design approach for the task (as mentioned above) which dominates technology today .

All this assumes that the artificial equivalent of a person or even a dog is our goal (the toy makers seemed to think so). In fact if we consider lower animals to show a level of intelligence then, for many jobs, that would be sufficient - indeed Mark Tilden, the creator of the BEAM concept in robotics and patent holder for the Nervous Net (as opposed to the neural net as widely used in AI today) believes that it is this very low level "intelligence" or lack of it which is our best way forward in the quest for usable robotics.

The good news is that it seems that a robot does not need to think and that what is termed as intelligence in today's non-biological world, can get us a long way..

It seems that any device which contains a microprocessor is termed intelligent, from stereo music players to washing machines. The term "microprocessor" is, in fact, the latest marketing BUZZ word and will actually suffice for most robotics purposes.
"Intelligence" in a system could be defined as some form of interaction between the device and its environment, to a greater or lesser extent, the result of which is then passed to its human operator thus reducing the amount of required intervention by that operator ( an advanced form of the automatic setting).

So we are back to the need for a definition and understanding of intelligence ......

All the publications shown here are readable by the interested non expert (or layman )

Understanding Intelligence - Pfeifer & Scheier - This is a good thorough introduction to the new approach to AI and covers all aspects of the argument for it, with many examples and thought experiments to colour the story - a hefty book and recommended. Maths is avoided for ease of reading.

Intelligent Machines - experiments in artificial consciousness - Braitenburg - This is the discussion of the Braitenburg Machines which, stage by stage, outline how it is conceivable to create what would pass for consciousness in simple buggy type machines - referred to in brief in Understanding Intelligence - a small book but a big read. Maths is avoided for ease of reading.

Creation - Life and how to Make it - Steve Grand - The inventor of the Artificial life game Creatures - a pretentious title maybe but an excellent read and a concept-expanding experience - not written for an expert but one most experts should read.