Intelligent Systems Development
Intelligent Systems are systems with in-built "Artificial Intelligence" that is useful for monitoring and controlling the business, for discovering new knowledge, as well as in augmenting human decision making.
Intelligent systems add advanced functionality, often to existing systems that they integrate with.
Intellia can build some types of intelligent systems, however, other more advanced systems can be only managed by Intellia; with the help of advanced programmer(s) to program the artificial intelligence features that we cannot built ourselves inhouse.
Artificial Intelligence systems are usually going to be integrated with existing systems, or new systems, for example a transaction-processing system. There are also intelligent systems, i.e. artificial intelligence systems that can operate fully autonomously.
We are fully prepared to provide you with the latest state-of-the-art artificial intelligence programming support you might need.
Intelligent Business Systems
-
Management Support Systems
-
Decision Making, Systems, Modelling and Support
-
Data Warehousing, Access, Analysis, Mining, and Visualization
-
Modelling and Analysis
-
Development of Decision Support Systems
-
Online Analytical Processing
-
Enterprise Decision Support Systems
-
Knowledge Management
-
Organizational Learning
-
Artificial Intelligence Systems
-
Expert Systems
-
Knowledge Acquisition and Validation
-
Knowledge Representation
-
Inference Techniques
-
Development of Intelligent Systems
-
Neural Networks
-
Development of Neural Networks
-
Advanced Artificial Intelligent Systems
-
Intelligent Software Agents
Example of an Intelligent System
The Technology: Artificial Neural Network (for Insurance)
Integrated Intelligent Enterprise System
The schematic below is a copy of an enterprise class intelligent system that integrates statistical analysis, decision support systems and an artificial neural networks. This type of system configuration was first proposed by Benjamin and Bannis (1991), and it has become a quasi generic intelligent systems architecture that has been copied by many large business intelligence software companies, and it is found printed in dozens of textbooks on advanced information systems / artificial intelligence.
This type of system can be adapted to suit the needs other organizations, in particular across the finance industry but there are other applications too, perhaps you can think of examples of your own.
The programming of such a system to customize this concept for your organization, can be done relatively easily provided you have the right programmers that can program the individual parts, and make them connect to integrate the software components into a whole and functioning system. What is more important is that the correct requirements are established, and that is best done by a person thoroughly schooled in business analysis. You can ask us at Intellia Ltd, to see if we can be of any help.
The System Component:
Integrated Intelligent Enterprise System
Schematic Source: Benjamin and Bannis, 1991. "A Hybrid Neural Network / Expert System for the Property Casualty Insurance Industry". Proceedings of the IJCNN, Seattle, WA.
The System Component:
Artificial Neural Network
(ANN for Insurance Underwriting)
Underwriting
Zooming in on the artificial neural network component of the schematic above, Intellia Ltd went to an insurance consultancy, to ask a few questions. We wanted to know what information underwriters ask for before they even begin considering the policy to be underwritten, and then on what terms. It is common practice for insurance agents and brokers to submit this information to one or more insurance companies, and if these insurance companies accept the submission the insurance companies will in turn submit this information to re-insurance companies (reinsurers), and these reinsurers may take this business to an underwriter. The underwriter takes the information, and due to the underwriters supreme knowledge of the 'situation' to use common layman language, will decide whether or not the underwriter wants this business. If the underwriter decides to take the business, the underwriter will then have the policy premium money, this money in turn is invested by the underwriter on the global financial markets. As a rule of thumb, underwriters take no less than 20% of the money for their own operation and they expect a return on their investment to be 20% or more.
The Insurance Submission
The Decision-making Structure can be programmed in the form of a purpose built and purpose trained artificial neural network. The submission to the insurance consists of five parts, the five reports shown in the first column (input layer). The inspection report reports on the business premises and any assets or things that require insurance. The Agent report may among other things include a report on the person or director of the company that seeks insurance, private detectives may be hired to assess the moral character of the insurance policy holder, if that person has drug problems, gambling problems, former convictions, a history of mental illness or similar, then these will be reported in the Agent report. The Risk report assess all the risks that can or require to be insured against. The Exposure is the value of loss that is incurred when any of the risk scenarios eventuate. The Protection is the level of protection, usually monetary compensation, that the business seeks to cover for the loss that may be incurred based on the businesses exposures. We may categorize this information as 'criteria'.
The Neural Network Architecture
Each criteria that is reported on, carries a certain weight when insurance companies assess the business as a proposal to them. Also note that each criteria affects every other criteria when considering how much weight is assigned to the information. Eventually a decision needs to be reached as to whether the insurance company wants to insurance this business or not, and if yes, what the terms of the insurance policy and the terms of the premium payments will be. Hence the decision, put simply has three options, a) accept, b) modify, c) reject.
So we at Intellia figured that the basic general architecture of the neural network should consist of one input layer with five criteria (each criteria has many sub criteria, not shown here), - the system has one 'hidden layers' that processes each criteria against all other criteria, and that the output layer consists of exactly one node, the decision, and out of this node come three decision outcomes.
Schematic Above: the neural network architecture depicted above was researched and first proposed by Intellia Ltd, in consultation with insurance experts from an Auckland-based insurance consultancy.
Legal Issues Arising from the use of ANN's for Underwriting Decisions
What criteria reports insurance companies demand is their prerogative, how much weight they assign the to each item in each criteria report even more so, and this comprises their commercial secret, and this information is highly commercially sensitive. It goes without saying, that companies are perfectly within their rights to automate their manual processes, since they are automating merely the information that the insurance company now owns, though they need to preserve the information's confidentiality. So far so good.
The use of artificial neural networks for underwriting decisions arise from the fact, that neural networks can make decisions, often with deadly accuracy usually between 95% to 97% far higher than any team of human decision makers score by parallel comparison, but that only a human operator can "explain" to the end-consumer why or why not their application for insurance was rejected, needs to be modified, or is accepted at that particular rate or premiums to be paid. Consumers have the right to ask why or why not this or that decision was made, and there is no way yet, using currently available technology that a neural network can explain any decision it makes, it simply makes the decision based on a vast amount of trial and error combinations using vast amounts of information, it discovers certain patterns, and in a machine like fashion works through all the combinations of the pattern and eventually concludes a certain pattern, matches this pattern against it's training pattern (neural networks need to be 'trained' much like people), and then outputs a match, mismatch, near match, as in the case example above. Only a a team of human operators can actually try to begin to explain why a certain decision was made, based on their explicable human reasoning.
Why is it not permissible to use ANN's , shouldn't we change the law to make industry even more efficient and pass on the cost-savings benefits to end-consumers and or either corporate balance sheets?
Well, to be sure, consumers have the right to ask for an explanation and it is not good enough to simply say, "because the computer said so." The insurance company would have to explain then 'why the computer made this particular decision and not another decision' and that would necessitate giving the client a highly advanced lesson in mathematics, pattern recognition, followed by a copy of the exact weighting and training patterns that were programmed into the neural network, essentially the insurance companies entire commercial secret visa vi their neural network would be comprised. Another legal problem is that even with a copy of the neural network and all its particulars disclosed, it still does not amount to an explanation since a human being cannot replicate in thought what the machine did. Human beings cannot now retrace, to use an extreme example, 3 million recursions using various algorithms, and then arrive at the same decision as the neural network did, the human brain is not capable of that, and it would consume not one but several human lifetimes to work through this process, hence it is 'unreasonable' to say that this explanation can be humanly understood, is in a format that is usable, and this is simply not strategic for the company with respect to its customer relationship management.
The use of Neural Networks and the legality of their use, depends on which country your organization is chartered in, it is a question of what jurisdiction your organization is located in and what it's pertaining laws are.
Lawful application of Neural Networks
If a proprietor of a neural network wants to consult this ANN for his own purposes, or the purposes of his companies business, then that is allowable, provided there is no end-consumer involved who by implication has the right to have things properly explained.
The use of Neural Networks for the monitoring and control of equipment, machinery, and generally all purposes related to engineering is allowable. This is based on their safety record, their ability to do things humans cannot, and the fact that there is no need to ask a machine why the machine made a particular to an end-consumer. Of course technical specialists in engineering, carefully watch neural networks and adjust their training, and likewise those neural networks used in the finance industry that don't involve end-consumer explanation issues also watch their neural network carefully and train, and retrain as they need to.
