Cyber-Agents in the form of networked systems of small intelligent autonomous programs are playing an increasing role in managing business and lifestyle services. Their real-life counterparts include human social communities, social insect colonies, the brain, immune systems and the Web itself.
When such virtual agents interact with each other cooperatively they form a collaborative multiagent system, which greatly enhances their level of operational efficiency and problem-solving capacity. Such systems are already routinely applied in a wide range of complex operations including- marketplace trading, financial management, communications, social networks, scientific computing, manufacturing and power distribution.
Because of their potentially high level of collective intelligence, often far exceeding individual expert analyses, they will increasingly be applied to critical decision support functions across all aspects of our social, financial and engineered infrastructure in the future.
At a personal level, virtual agent and avatar programs also learn user's interests preferences and habits and provide proactive personalised assistance to enhance individual and team productivity in the workplace; analogous to personal and executive assistant functions in the real world.
An ideal software agent should not only know its goal and determindly try to achieve it, but as with human agents, it must also be adaptive; capable of learning from experience and responding to unforeseen circumstances when required with a repertoire of knowledge and problem-solving techniques. It should also be relatively autonomous by sensing the current state of its cyber-environment and acting independently to make progress towards its objectives.
Collaborating agents also belong to the wider class of complex adaptive systems- CAS. Such systems are able to spontaneously coordinate their activities in much the same way as crowd-sourcing achieves; cooperatively applying knowledge from a diverse set of databases, algorithms and repositories across the Web. For example broker agents process requests against a domain ontology by finding other agents that can satisfy the required knowledge goals. Resource agents can then be recruited to support the implementation protocols required between user queries and information resources.
Such agent systems are already beginning to play an increasingly important role in enterprise applications such as e-commerce, marketing, scientific computing, intelligent manufacturing, home automation, component-based software construction and smart power grid management.
As multiagent technology moves into the marketplace there is also increasing interest in techniques for modelling collaborative systems as well as methods for constructing them. The most promising approach to multiagent design employs artificial intelligence techniques based on artificial life research. Virtual organisms are programmed to self-organise or adapt autonomously in response to their changing environmental demands, by applying evolutionary principles. Such AI mechanisms incorporate neural networks, evolutionary algorithms and other advanced learning techniques to allow realistic and flexible adaptation.
Basically, Agent-based modelling becomes essential when the difficulty of simulating large numbers of interactions is too complex for traditional analysis. Complex real-world behaviours reflecting millions of decision processes such as traffic congestion, epidemics and financial crashes will increasingly require the application of such techniques. There are already plans to create models with up to 10 billion agents, which could simulate the interactions of a large city population or even the entire planetary biosphere.
Rational agent programming is another related field, used for monitoring and managing complex control systems in engineering and communications, requiring the integration of multiple sub-systems and dynamic controller feedback loops. Automatically monitoring such complex environments is difficult because it is necessary to process large quantities of data acquired by different sensor systems.
Over time, artificial evolution will codify and combine the behaviours of the most effective agents to evolve fitter populations, with each succeeding population better adapted to serve the user's interests and goals. Such agent ecosystems will continually renew themselves as their environment changes, eventually having an enormous social, economic and political impact.
When multiple related multiagent systems link and interact they create powerful cyber-agent ecosystems with emergent properties.
New types of behaviour will start to emerge, not previously envisaged by designers and not capable of being reverse-engineered; interacting counter-intuitively with other critical enterprise operational and administrative processes on the Web, such as cyber-security protocols.
In the near future it is highly likely that such intelligent cyber-agent ecosystems will play a major role in the autonomous management and evolution of the future enterprise.
Blog Link- http://futureenterpriseblog.blogspot.com