This is quite correct: there has indeed been a huge amount of progress in many sub-fields of AI. One place to find pointers is The AITOPICS web site. There are many other sources of information about progress in AI on the internet though not all are equally reliable. Some more links can be found in two documents of mine:
All such overviews should be treated as incomplete because the field is developing all the time.
- What is AI?
(written for careers advisers in schools).
- What is AI?
(written for the UK's Quality Assurance Agency Computer Science benchmarking panel.)
My slides do not do justice to all this, because the point being made is a different one, namely that despite the progress that has been made, and is being made, in many sub-fields of AI, there is serious fragmentation in the field because:
- most people working in the various sub-fields (e.g. vision, language, planning, theorem proving, neural nets, etc.) have no interest in or knowledge about what is happening in the other sub-fields (though there are notable exceptions, of course).
- most people working in AI do not think about how to combine results from the various sub-fields in order to produce a complete human-like functioning system.
The fragmentation started when the number of people doing AI began to grow rapidly, in the late 1970s and early 1980s. It has continued since then. Up to around the mid 1970s most AI conferences tended to cover all fields of AI, and people who were doing AI research or were teaching AI usually made it their business to be aware of the main developments in all sub-fields. At that time the only text-books on AI attempted to cover the whole field.
I should also point out that people working in robotics at present are in a sense required to study ways of combining different capabilities. For example, depending on the sorts of robots produced it will be necessary to have some sort of perceptual mechanism, motor control mechanism, action selection mechanism, and a subset of planning, reasoning, learning and communicating mechanisms. Integration is a requirement for a complete working system. However, it is often the case that the robots are designed with very simple capabilities because they are not intended to be human-like but merely to perform some particular class of tasks, e.g. on a factory assembly line, or in a robot soccer team, and some people are explicitly aiming only to replicate capabilities of insects or other simple-minded creatures. However, there will, increasingly, be attempts to build more sophisticated and more human-like robots, not just in their superficial appearance and behaviour like the very engaging Kismet robot but rather in their deep cognitive capabilities, including the ability to think about and talk about what they are doing, and learn as a result, as described by John McCarthy
In particular, the RoboCup Rescue project is likely to require such developments in order to meet the full range of requirements, namely producing robots that can perform search and rescue operations in disaster scenarios. There will be a series of competitions, some involving simulated robots in virtual world scenarios, and some involving physical robots in mock-up disaster scenarios, or even real disaster situations, just as the RoboCup soccer competitions have included both simulated and physical soccer-playing robots.
An amazing sheep-shearing robot is described and depicted here.