Expert Systems and Logic Games
By ai-depot | June 30, 2002
The Way Forward: Learning?
State of the Art Expert Systems
In early research and Artificial Intelligence applications, this approach was dominant. To some extent, it is still used today — but to a lesser degree, since attractive alternatives are available. Expert systems are now also being created with more advanced techniques. Human influence is still more or less required, but can be limited to supervising a data-mining process to extract the rules.
Despite the increasing quality of Expert Systems, research and applications are moving slowing away from them. Indeed, they require time and effort of valuable experts, who could be spending their time on more valuable tasks within the company. As such, more flexible solutions are being defined that minimise the workload� for big problems this can save a huge amount of time, but for smaller problems it’s often best to stick to simple expert systems. This decision should be based on the requirements and made with extreme care, as it will greatly impact development times and required resources.
Learning from Data
Much of the current research is moving towards learning and statistical analysis. This allows a system to learn from logs and logs of historical data, or simply learn online via interaction with the environment. This approach allows the programmer to simply setup a framework, and let the system learn the appropriate behaviours by its self. This can potentially be more creative than an expert system, since its ability to learn will not be biased by experience… for example, a version of computer back-gammon managed to out-class even human experts, introducing a new strategy.
Once could argue that there is still a lot of expertise required in designing a learning system. And to some extent, this is true; a certain amount of knowledge of the problem allows a more efficient design of a working solution. However, the expertise lies more in the ability to code AI systems — a skill which can be recycled. Rather, the expert’s potential background knowledge of the problem is integrated into the design rather than the expert rules.
Conclusion
Expert Systems are a great way to solve a simple problem quickly and efficiently. However, this approach scales up linearly in human resources and time with the size of the problem. This is not especially a desirable property, and can be remedied by using more flexible learning approaches.
But once again, it’s all a question of requirements; the trick is to maximise the ratio of time to result, and that’s just a matter of experience.
Written by Alex J. Champandard.
Tags: none
Category: essay |
