Responding to the e-Commerce promise

This article was published in 2000 in The Handbook of eBusiness (Warren, Gorham, & Lamont, 2000). I wrote this article in Frenglish ( I was still living in France) ....nothing really changed!

      e-Commerce and its thorns

         Sales across the Internet are growing rapidly, and Business 2 Business transactions are growing even faster. The entire e-Commerce infrastructure is dependent on secure and safe transactions. Maintaining a stable and efficient e-Business depends on effective security. In particular, eliminating fraud is foundational to the ongoing viability of many businesses involved in e-commerce, especially those buying and selling on the Internet. While much press attention has been given to end-user privacy and fraud issues, fraud can be a daunting problem at each stage in the supply chain, from materials procurement, to manufacturing, to distribution to the end user. Virtually all such transactions and communications will be handled through e-commerce within the next few years. The question merchants are asking is, “Is my future safe?”

          Early e-commerce businesses often developed proprietary, high-cost transaction processing systems. Today, many online merchants outsource e-commerce functions, such as credit card processing and fraud detection services. Fraud detection has become a major element within most companies' e-commerce strategies, particularly those doing business on the Internet, because detecting fraud provides a high return on investment. Conservative estimates foresee e-commerce fraud losses in the billions. Today, some web sites experience fraud rates in excess of 50%. There is a huge need for e-businesses use of nimble, inexpensive and adaptable solutions for fraud detection and prevention. Some state-of-the-art technologies, nonalgorithmic technology (Intelligent-Agents) at their core, have the edge to respond to these growing problems. 

          Older methods for detecting fraud use a rules and neural networks. these legacy technology are not adaptive and are inefficient in a world of ever-more clever thieves where new behaviors of fraud arise daily. 

         Fraud solutions based on neural networks offer limited applications. Some credit card fraud solutions use Backpropagation, one of the oldest and outdated neural network algorithms. This algorithm works in the manner of supervised learning, meaning that the system is fed many known cases of credit card fraud.  Then the system is able to detect fraud in future transactions, but only those like the ones it has learned. This technology does not have the ability to respond to the highly volatile nature of fraud. Same limits for rules.

 Getting a grasp on fraud

            In the e-commerce market, it is commonplace for companies to quickly enhance their systems to comply with evolving standards, respond to rapid technological change and address the continuous attempts of fraudsters to break in and abuse the system. In the area of fraud prevention, there is a scarcity of effective programming technologies that address the ever more prevalent security risks and new fraud types that are born daily. With Intelligent-Agent technology, companies can stop fraudsters in real time, thus eliminating the vicious cycle of fraud.

 Fulfilling the Promise

       Intelligent-Agents technology enables a quantum leap in artificial intelligence and fraud prevention. Companies armed with such technology can assist e-commerce backbone, business-to-business and business-to-buyer companies by drastically reducing the risks to their business, thus making e-commerce safer for all participants in the supply/value chain. Companies that employ such technology have the edge on empowering e-commerce and fulfilling its potential. Next-generation companies that can maximize the potential of e-commerce use technology that is:

·        Easier to implement and support

·        Adaptive - self learning

·        Less expensive

·        Scalable

A model E-commerce fraud prevention technology suite possesses these technologies: Goal oriented Intelligent-Agent metacontrol module, Constraint programming module, Fuzzy technology module, Multiple types of neural network, Simulation tool, Knowledge-based systems (Rules), Cased-based reasoning and learning module and Genetic algorithms.

What this means in the realm of fraud is that companies can 1) eliminate fraud at the first attempt 2) prevent fraud from occurring without massive amounts of data or algorithmic programming methods. One of the cutting-edge advanced technology solutions that can deliver on this type of promise is the adaptive Intelligent-Agent technology. 

 The Benefits of Non-Algorithmic Technology

Anyone impressed by the increasingly dazzling speed or colossal memory capacity of computers, will not be able to find, in these or any other astonishing computer traits, any manifestation of the slightest fragment of intelligence as long as computer programming remains purely algorithmic.

 An algorithmic program is a deductive set of successive operations applied in a fixed order. An algorithm enables the computer to repeat long suites of logical operations tirelessly and accurately, as long as the algorithm is correct. An algorithmic program will neither know how to take any initiative nor stray one bit from a fixed line of code. The programmer must dictate the precise succession of acts that the machine must accomplish.

The problem with algorithmic programming is best explained by example. For instance, you cannot ask a financial expert to predict all of the events that may occur during one year, a month or even a day. There are too many variables to code. This weakness of algorithmic programming holds true for all of the domains requiring the use of human expertise. Every algorithm requires an exhaustive enumeration! This excludes the vast majority of real-world business problems from the field of computer science.

Business problems that require a minimum amount of reasoning cannot be transcribed into an algorithmic form. This is also true of programs based on artificial intelligence, expert systems, neural networks, object-oriented languages, etc.

In the case of expert systems (Rules), you must predict these possibilities by writing all of the possible rules (obviously impossible). In neural networks you need to train your system and have many samples in order to have a satisfactory result. In object-oriented languages, you must foresee, know and program all possible methods (this which is also impossible).

The Example of Chess

Even worse, the presence of an algorithm doesn't guarantee its usefulness. For instance, an algorithm may be known for playing chess, but the complexity of the game makes it unusable even if we know the rules of the game.The number of adversaries is constant, all of the rules are known prior to beginning, the chessboard keeps 64 cases throughout the game and no new piece, a tank for example, will ever appear during the course of the game! However, chess is child's play compared to analyzing, in real-time, stock market fluctuations or dealing with unexpected events.

The Total Greater Than the Sum of Its Parts

A Intelligent-Agents system functions like a community of agents possessing an expertise, exactly like a human society. The knowledge or competence of an application built with such a system will be the sum of the competencies of the community of agents. In other words, the addition of a new agent corresponds to a new competence and the addition of a new agent doesn't imply the modification of a principal program: each agent auto-constructs their own interpreter.

 A Goal Oriented Language

An algorithm is a program that you write in Lisp, C, C++, etc. A typical statement would be:

         If we have this

                    then do this

                             else if this

                                         or

                                                 switch

                                                          case if

What is the if statement of a program?

The if part of a program is the input(s) data. We have no control on this data/events.

 What is the then statement of a program?

The intelligence of the program is what the programmer writes about the different then statements, for each if part the designer will write the optimum then part. Each agent can determine what information is in favor or disfavor with his goals (global or local) to make the right decision. This reasoning closely resembles human reasoning. Let us refer again to the chess game as an example.

What difference is there between a good game of electronic chess (programmed using the classical algorithms for game development such as MinMax, Alpha–beta, Scout, etc.) and an excellent player? The difference is the effect of the line of horizon. The computer works in "brute force", working at a prodigious speed to best react to the present situation and the situations that might occur during the next five or six turns. After this line of horizon comes the invisible. Furthermore, the importance of a piece is simply determined by the affectation of a number (weight). For example, it will sacrifice a pawn to save a knight one time. The knight remaining, has once fooled the computer and distracted it by this defensive strategy.

An excellent chess player works by goals and sub-goals. He fixes on an objective strategy from the beginning. He progressively puts the conditions of the win in place of this objective in working out his strategy in goals and sub-goals. His moves are therefore at the service of his strategy and not in "brute" response to the situations he encounters. An algorithm is, incompatible with the notions of strategies and evolution. With a nonalgorithmic technology (Intelligent-Agents), it is possible to overcome the limits of classic technologies in that you can assign goals to your agents without any need for programming, consequently going beyond the algorithmic techniques.

 A Intelligent-Agents Solution

A Intelligent-agents solution consists of a group of intelligent agents, each one with an expertise that communicate among one another, researching the equilibrium of everything -- identical to a human society.

 What is an Intelligent Agent?

An intelligent agent is an entity that is capable of having an effect on itself and its environment. It disposes of a partial representation of this environment. Its behavior is the outcome of its observations, knowledge and interactions.

Thinking Machine?

Until now, the resolution of any computer problem has been to find the best way to move from an initial state to a final state by exploring intermediary states. In fact, the majority of complex programs using classic techniques: object-oriented languages, expert systems, or the classic algorithms are often faced with the problem of combinatorial explosion, intrinsic to the philosophy of the exploration of states. With Intelligent-Agents, the resolution arise from the communication between the agents.

Take for example the Traveling Salesman Problem, where the goal is to find the best order (such as the one which has the shortest distance), in which to visit the cities. Adding a city causes a drastic increase in the number of possible solutions. To illustrate this, here are the numbers of possible solutions for some problem sizes.

 1  => 1

 2  => 2

 3  => 6

 4  => 24

 ...

 10 => 3,628,800

 ...

 20 => 2,432,902,008,176,640,000

 ...

 500 => 1.2201368259111 x 10^1134

With a Intelligent-Agents technology there is no need of exploration of the space of states. The resolution of a problem will emerge as a side effect of the communication between agents.

Essentials Notions

We will list some notions that need to be present within a Intelligent-Agents technology.

Environment

An intelligent agent has a partial vision of the other agents. This vision is tied to the nature of the messages exchanged between the agents. The agent’s environment is its base of observation vis-à-vis the global system; this environment reflects the knowledge that the agent disposes on its milieu. The environment is to an agent what the base of facts is to a knowledge-based system. It represents everything that the agent considers true, generally the information is values taken by agents, attributes or even attributes of agents. The environment can be divided into two zones:

 Public Zone

All of the agents of the systems have the authorization to write in this zone. That translates by the fact that an agent would like to make information available of a general order to certain agents. If an agent discovers an interesting piece of information in the public zone of its environment, it can at any moment transfer this information to its private zone after validated this information.

 Private Zone

The intelligent agent will keep total control of the information in this zone and contains the validated, true and accurate information.

 Inputs/Outputs

The inputs of an intelligent agent are a very simple and powerful way of modeling the streams of information that traverse an agent. Very close to the technique of neural networks, they add the auto-reflexivity to each neuron (this technique does not exist in neural techniques): in regards to the individual and combined force of these streams, the agent may or may not trigger an exit signal. An input is an agent that can trigger several outputs, and an output can be triggered by several inputs. Values can be attached to these outputs according to the different criteria of credibility.

 Organizations

The concept of organization defines a group of intellligent agents.  You can associate only one mailbox to all of the members of the organization, thus, reducing the number of messages. The identifier of the organization plays the role of supervisor for the member agents of the organization. You can benefit from the competencies of all the member agents without having to memorize the group of addresses of all the agents of the organization. 

 Multiplicity of Viewpoints

We are constantly confronted with a multiplicity of viewpoints. For example, in any given company, on any given problem, differing points of view, with contradictions, can arise between the head of the technical department, the director of marketing, and others while they each work for the good of the company. The concept of point of view determines the attitude of the agent vis-à-vis its interlocutors. A person adopt different roles depending on whom he is interacting with.

1. son, if he’s conversing with his father
2. brother, if he’s conversing with his brother or sister
3. friend, if he’s talking with his friends
4. student, if he’s speaking with his teacher
5. etc.

Therefore, an intelligent agent disposes of several interfaces according to the different roles that he must obtain. 

 Logical Simultaneity

Even with a monoprocessor machine, the intelligent agents must act at the same time. In that, the election of an agent at a given moment cannot affect the behavior of another agent until the next cycle. Thus, the intelligent agents conform to reality: planes, cars, trains, etc. all move at the same time.

 Urgent Messages: Present, Past and Future

The messages exchanged between the intelligent agents can be different in nature. They each have an age, and their priorities as well as their semantics depend on the nature of the messages; this is the contrary for object languages wherein the messages (rather than the calls of procedure) are all the same nature.

  Agent Behavior

The programmer can associate (not required) several conditional behaviors (declarative and or procedural) to each one of his intelligent agents.

 Cooperative and Egotistical Agents

The intelligent agents are at the same time egotistical and cooperative. This egoism is found in the notion of survival, inherent to each agent: if an agent doesn't develop itself, it is destroyed as the organization rejects any nonproductive agent.

 Chief Agents

These intelligent agents can receive orders from agents having a special statute, similar to a military hierarchy. These orders must be executed even if they destabilize the receiving agent. .

Approached Solution

It is better to have an acceptable solution than no solution. In object oriented programming or expert systems a solution either exist or does not exist, while a Intelligent-Agent will always provide the better possible solution even if it is just a partial solution.

 Contextual Reasoning

What counts as a significant depends on its presence in a given situation. the word Apple could be a fruit, a computer,Apple of Discord, a city, etc. With a Intelligent-Agents, you can define reasoning strategies based on different context.

 Everything is an agent

In a Object oriented programming an attribute is just a storage for a value. In a Intelligent-Agents system, attribute, message are also an agent.

 Incomplete and Imprecise

The manipulation of imprecise notions and a nuances by using fuzzy logic which results in a more flexible evaluation of information rather than a strict binary response.

 Self-Learning

An intelligent agent behavior is continuously updates it knowledge by the data transmitted to the agent. This evolution is the result of the successes and failures obtained from the anterior actions.

 Intelligent-Agents Technology and Fraud Prevention

Intelligent-Agents technology is close to Human Reasoning. It is difficult to imagine an algorithm capable of adapting to constant changes, it is easier to imagine defining entities with competencies and goals and viewing the resolution as a side effect of negotiation and collaboration. Intelligent-Agent technology offers e-Commerce companies a very unique solution that will enable them to enhance the way they will do business in the future. The speed of development and the reasoning power of this type of solution are linked with improving companies’ profitability. This solution is self-adaptive, by nature, easier to implement and support, lower in cost, very scaleable and less expensive to maintain. Many times, merchants don’t know what to look for or even the questions to be asked. Intelligent-Agents technology is smarter and require less of the merchant and provide a higher degree of security. Such a solution proposes a promising solution to the complexities of fraud.