Internet is such an open community with a global user base. Unfortunately, trust is a major issue. Web security over decades has become a separate mainstream business on its own. However, to keep up with the pace on updates and new malware is a constant concern not only to users but also for software providers. Inevitably, hackers are able to prune out backdoors into all sorts of software to do harm. And, there just is not enough being done to make it a secure and safe interconnected network of places for users. In the end the integrity of the internet as well as service providers suffers who at times can suffer downtime as a result. There are also privacy issues around the way data is being handled. Perhaps, all this boils down to what is sufficient security and whether it is robust enough. There are also separate issues here in regards to preventative steps as well as actionable software to fix issues once havoc is wreaked through a system. There are a multitude of new alternatives in comparison to the standard ones that most security software is built on. For one, they need to harness better encryption of files. There also needs to be a more distributed security model that facilitates a way to bolster a stronger penetration in an evolvable way. In this respect, natural computation algorithms that provide for global optimization are one aspect of the answer. But, also to utilize multiagents to deductively respond to attacks on a system by making the security even stronger in a manner of adapting and learning per each iteration. Perhaps, even aspects of reinforcement learning would be another way. There is a lot that can be done from utilizing machine learning algorithms. In a way, the security system would first have to detect a penetration, then to respond to the penetration by evolving the system, and then to learn based on patterns of attacks. It would make it extremely difficult for a hacker to detect an evolvable and random change in patterns especially when the system is always a step ahead. Learning based on security patterns is relatively good way to bolster a stronger security model and then to erratically mutate and crossover to add even more variety. So, to optimize the solution would involve aspects of evolutionary computation which would entail adding calculated randomness to the patterns based on some fitness function. Even aspects of utilizing swarm intelligence could work to identify an attack by clustering on as a food source and then to take the appropriate steps to resolve like swarming army of bots. For most artificial intelligence problems and especially for data mining in general, there is usually a set of inputs and a set of desired outputs which can feed into each other on a pipeline. Thus, it can be inferred that most applications of artificial intelligence involve a black box where most of the processing is carried out against a set of inputs and outputs. In a similar manner, for security the inputs can be the penetration attacks on a system, the process can adapt, and adjust to provide for better output of instructions as to what counter steps to take based on a feedback loop. In process, making the system evolve in a form of a game play. For web security, adaptability is probably the best approach as it can also be unique in patterns to each system. Not every system requires the same level of security measures and audits. Web browsers also need to be pushed towards more intelligent means of security as pluggable options which essentially act as clients on a client-server model. Unfortunately, there will always be people looking for backdoors to target ethical users on the internet who may be all too innocuous. Linked data and semantic web could also entail a solution especially for added knowledge representation. Standard systems of today really need to start taking inspiration from artificial intelligence in order to make systems smarter that can be both future proof and make our lives easier as well as more pervasive.
