EXPERT SYSTEMS
Primary Argument
The average life expectancy of an Expert System was 18 months as the inference engine weakness was that it lacked ‘human’ judgment to ‘know’ when to stop learning. As a consequence, it over learnt leading to decisioning distortions that are typically not obvious at first but then rapidly deteriorate. The problem was further exasperated as the Expert System is ‘black-box’ AI, which means there was no transparency of how decisions were made. The legacy of this issue has become more topical recently in the wider area of AI related to trust and ethical risks.
Secondary Argument
Knowledge acquisition required the coming together of two specialists: subject matter experts and knowledge engineers to develop the data and inference procedures. This created interpretation issues, plus high costs and elapsed time for knowledge acquisition.
Other Arguments
Wikipedia http://en.wikipedia.org/wiki/Expert_system/ summaries Expert Systems strengths and weaknesses quite well as:
Strengths:
• Provides consistent answers for repetitive decisions, processes and tasks
• Holds and maintains significant levels of information
• Encourages organizations to clarify the logic of their decision-making
• Never "forgets" to ask a question, as a human might
Weaknesses:
• Lacks common sense needed in some decision making
• Cannot make creative responses as human expert would in unusual circumstances
• Domain experts not always able to explain their logic and reasoning
• Errors may occur in the knowledge base, and lead to wrong decisions
• Cannot adapt to changing environments, unless knowledge base is changed
EXPERT SYSTEMS DECISION-TREES
Expert Systems use data driven Decision-Trees for forward and backward chaining. However, it is the use of data driven Decision-Trees combined with inference-based learning that led to the weakness of Expert Systems.
The use of Decision-Trees is fundamental to IF-THEN-ELSE so an alternative approach was needed that bypassed the weaknesses of Expert Systems.
DECISIONALITY DECISION-TREES
Many concept and technology challenges had to be overcome to provide an alternative set of technologies to Expert Systems.
One over riding challenge was that the tools had to be simple and natural to use. Though this has been achieved it does create the perception of familiarity from the past and being so simple how can it be any good.
This is our oxymoron as the technology is very simple to use but at the same time it can cope with the velocity, volatility and complexity of procedural knowledge.
Procedural knowledge is about IF-THEN-ELSE logic. Decision-Trees are a natural visualization for this type of logic. Of course, knowledge cannot be contained in one Decision-Tree. So by treating knowledge as a collection of Decision-Tree objects which can be linked enables an ecosystem to be created that can cover any breadth and depth. Indeed, this ecosystem of knowledge objects can grow and evolve without constraint.
This is quite a challenge as software convention is that the bigger a system becomes the more complex it is too build and there is a tipping point where it becomes untenable. Yet the ambition to cover all procedural knowledge that can continually evolve demanded a software solution that would have to be different. We solved this by:
• Automatically generating a stateless web-service from the Decision-Tree script – this means no database is used which is the opposite of an Expert System.
• Enable the Decision-Tree script to dynamically link with a web-service that required a handover to a different knowledge set.
This combination enables creation and change to happen at the ‘speed of articulation’ thus supporting the development of an ecosystem that can change and grow as fast as is needed.
At the same time this approach solved another problem and that is the usage of the knowledge. People only want to navigate down the pathways of their choice without having to be exposed to the landscape of the knowledge. We have found that knowledge interaction can go deep very quickly at the rate of say a step every 5 to 8 seconds.
This approach means the only limit to the number of pathways and outcomes that can be interacted with is a constraint of the Decision-Tree scripting and not a constraint of the technology.
Because we can automatically measure the interaction in terms of steps, flows, patterns and metrics the learning loops can be very fast but they are human controlled. This whole approach is white-box.
The combination of capabilities provides a sound foundation for the acquisition, application and usage of procedural knowledge that can be accessed by any connected device or application.
The DecisionFlows scripting tool embraces visualization, narration and simulation. It is designed to improve logical thought and the acquisition of knowledge.
The Decisionality Decision-Trees use forward and backward chaining techniques as do Expert Systems. The difference is Expert Systems are data driven whereas the Decisionality is driven by human interaction.
Here are some other points for consideration of advantages between Decisionality Decision-Trees over Expert Systems:
1. Decision-Trees are simple to script whereas an Expert System needs a knowledge engineer thus creating a skills shortage plus overheads for knowledge capture from subject matter experts.
2. Decision-Trees are an open system as it is easy to link the end of a pathway within a Decision-Tree to the start of another pathway in another Decision-Tree. An Expert System is a closed system as it is constrained by the design of its inference engine and application.
3. Decision-Trees are simple ‘natural’ programs that can adapt to complexity and chaotic conditions. Expert Systems are designed to cope with a complicated system, but have difficulty coping with the dynamics of complexity and chaos. (20 years validation Stephen Wolfram New Kind of Science).
4. Decision-Trees are stateless whereas Expert Systems are data driven, which has an embedded complexity cost as more data items are developed and used.
5. Decision-Trees are a white-box meaning it is transparent, simple to understand and interpret. People are able to understand Decision-Tree models and therefore are designed for organizational retention of knowledge. Expert Systems are a black-box that is complex to understand and has greater reliance upon tacit knowledge of a specialist that dilutes overtime.
6. Decision-Trees can have value very quickly even with a small number of nodes. Important insights can be gained from usage that often stimulates ideas for knowledge evolution that were not obvious at first. Expert Systems in comparison evolve slowly and to the point of decision contamination.
7. Decision-Trees can be created by subject matter experts without the need for software specialists as with Expert Systems.
8. Decision-Tree dynamic linking generates an exponential growth of pathways and outcomes compared to a far slow rate from Expert Systems that will then reach a plateau without further advancement being made.
9. Decision-Trees development enriches inductive and deductive reasoning as it focuses upon pathways and outcomes. This value is largely diluted with Expert Systems as it is in the hands of the knowledge engineer and not the subject matter experts.
10. The interactive use of Decision-Trees, via web-services, and the automated analyses of interactions provide short learning loops enabling Decision-Trees to be evolved rapidly based on behavioral insights. Expert Systems capabilities weaken overtime as mentioned.
11. The reuse of Decision-Trees and dynamic linking of Decision-Trees (via web-services) increases value overtime enabling exponential growth of pathways and outcomes in very short time frames.
12. Decision-Trees with stateless web-services do not create legacy systems unlike Expert Systems.
13. Expert Systems decreases in value overtime whenever there are high rates of complexity, velocity and volatility. The opposite is true for a Decision-Tree ecosystem.
14. Decision-Tree scripts have been extended to influence behaviour of usage so that other mediums like pictures can be used to influence decisioning. This is accomplished because of event-tags being available for each scripted step. This level of granularity is not available from Expert Systems.
15. As the Decision-Tree generates a web-service this can be accessed across all digital touchpoints. Expert Systems do not typically have this level of interoperability.
16. The construction of the Decision-Tree is simply not just focused on business logic but on good dialogue and choices to influence behaviour and decisioning. The automated analyses of behaviour enable the Decision-Tree to adapt to behavioral dynamics. Additionally, dialogue narrative can change according to pathways chosen. Expert Systems have limited capabilities for the nuances of different behaviors.
17. Decision-Tree source and its web-service representation are measurable as a knowledge asset. This is not the case with Expert Systems.
18. Decision-Trees can be developed in parallel with each one often being built in a matter of hours. Therefore the speed of building an ecosystem from existing procedural knowledge is very fast.
