Many believe that the path to effective artificial intelligence will be realized by mimicking human intelligence. This thinking is limiting understanding of what AI may be capable of achieving, even in the near future. We believe that when AI systems capable of exhibiting true machine intelligence emerge, they will not behave in a human-like fashion.
In an attempt to understand how such machine intelligence might eventually be applied in the cyber security domain, we have established a research effort named Project Blackfin.
Collective intelligence techniques such as swarm intelligence and multi-agent reinforcement learning represent the first steps on a path that may lead towards the evolution of true machine intelligence. Collective intelligence techniques involve the modeling and study of interactions between multiple agents in a system. Interactions between these agents can often lead to the emergence of unexpected behaviours that are not unlike naturally-occurring cooperative phenomena seen in schools of fish or insect colonies.
In our opinion, more research should be directed towards discovering and utilizing emergent machine intelligence in its own unique form, instead of trying to mould it into something that behaves like humans. If our ultimate goal is to create systems that surpass our own intelligence, we should think beyond humans and their capabilities.
Agents in collective intelligence swarms communicate with each other and share knowledge. Federated learning is one example of a commonly-used knowledge-sharing mechanism. As new, better communication and knowledge-sharing mechanisms are established, collective intelligence systems will become more capable and powerful. In the future, when collective intelligence swarms are comprised of truly intelligent agents, their capabilities will extend beyond anything we can currently envision.
In nature, swarms of organisms work together to perform actions that surpass the capabilities of each individual. By interacting with each other, learning, and sharing information, these collections of organisms are able to solve problems in sometimes unexpected ways.
By increasing the complexity of individual agents in a collective intelligence swarm, more intricate behaviours, and thus more complex capabilities emerge. These resulting emergent behaviours can be applied to solve real-world problems such as those we face in the cyber security domain. We believe that collective intelligence techniques will also be relevant in a variety of other fields, such as transport, energy, logistics, and self-driving vehicles.
Project Blackfin is a multi-year research effort aimed at investigating how to apply collective intelligence in the cyber security domain. The research, which is being led by F-Secure’s Artificial Intelligence Centre of Excellence, is a company-wide effort involving F-Secure's engineers, researchers data scientists, and academic partners. The near-term goals of this research are to:
Develop new, more generic methods for detecting adversarial actions.
Create mechanisms capable of tracking attacker actions across multiple endpoints on a network.
Further improve and automate threat intelligence gathering capabilities.
Understand how to implement and improve automated response actions.
Implement mechanisms that are able to perform contextual risk analysis on each endpoint.