Current methodologies won't create the future

Machine intelligence has traditionally been siloed into separate research and development applications. Each application requiring its own sets of algorithms, data and knowledge representations, and specific assumptions. Inevitably, knowledge engineering or modeling creep into the process, eliminating the possibility of a true autonomous intelligence by injecting human intelligence into the solution. Though some successes have been made, such an approach is counter to the goal of developing true Artificial Intelligence. Often, it is guesswork to discover one of many algorithms in a toolset, or a statistical or neural net model that works well on the historic dataset. When one is found that appears to work well, this has never lasted when the solution is applied in the real world. Real world data and the environment it springs from, changes. These static solutions need to be constantly re-worked to keep up.

But, intelligence cannot be manually coded. Instead, an environment must be created that allows for intelligence to emerge from it.

Intelligence is an Artifact

Nature did not set out to create intelligent creatures.  Intelligence is an artifact of configuration (DNA), environment (data and evolution), and processes that drive survival and reproduction.  A race for survival that began with classification and recognition lead to prediction and cogitation.

Rather than focus on any single algorithm or "how" the brain or collections of neurons do what they do, our approach holistically investigates "why" they do what they do. Uncovering the "why" has allowed us to understand the processes from which intelligence emerges. Instead of simulating the mechanics of neural activity, our research replicates the functions of large collections of neurons. This enables it to work - without change - within multiple application and problem domains.

We have created a proprietary system that draws on deep cross-disciplinary insights and discoveries. We have used principles derived from neuroscience, physics, mathematics, computer science and biology to create a framework for machine-generated intelligence. This research has yielded the "Cognitive Processor". The Cognitive Processor is the fundamental cognitive processing unit of intelligence of the COGNITUUM platform.. Cognitive Processors can be linked in any network topology to further process information. Like the brain, some of these topologies can be hierarchical. Others can be more ambiguous. The connections between Cognitive Processors are analogous to the connections mapped by the Human Connectome Project.

We have built the Genie Agents in our COGNITUUM platform to research these connections. But Genie pushes the technology even further...

Autonomy requires Freedom

For true machine intelligence, autonomy of both the agent and the construction of the agent is necessary. Otherwise, a human would always be required in creating the agents.  For true machine intelligence, the machines need the freedom to develop without artificial limits set on them by humans.

That's not to say that we have no control over our Genies. There is an existing precedence for how humans adapt other intelligences. Just as domesticated animals have been bred for centuries to work with us and be our companions, Genies can also be bred to behave as we wish. Desirous traits can be bred into a breed, and undesirable traits can be eliminated.

Our R&D includes provisions for automatically generating the Genie Agents and automatically evolving them within their environments. We use genetic algorithms to both produce and breed new generations of Genies. The result is a powerful platform that can provide solutions to any data-driven problem, in any problem domain.


Genie Technical Overview


The Genie is the convergence of Artificial General Intelligence, Big Data, Deep Learning, and Evolving Algorithms. It is Cognitive Computing that doesn't require modeling the data or knowledge engineering, so it continues learning and adapting in real-time. Its Cognitive Processor works for any problem domain without re-coding.

Genie’s proprietary cognitive computing model uses insights from nature to address extreme complexity. Even using all the world's best minds, there are more problems, each with its own complexities and nuances, than could be manually solved and converted to programs. Nature solves this problem through evolution. Good solutions breed to produce new solutions, with poor ones becoming extinct and better solutions surviving.

A specific custom agent solution is called a “Genie”. The behavior of a Genie is controlled by its “DNA” which evolve via genetic algorithms to better suit the genie's environment. They mutate and breed with other Genies to produce more powerful offspring until a desired behavior is obtained.

Genie’s simple API – compatible with numerous languages and development tools – makes it easy to create a Genie, add its intelligence to virtually any application, and begin reaping the machine intelligence rewards. The application provides the Genie with data, and genie returns knowledge, decisions, recommendations, and actions.

Unlike other AI solutions, Genie supports both text-based and vector-based data, and doesn't require converting one into the other! When a developer's app is ready for classification, pattern recognition, predictions, anomaly detection, or decision support, they simply copy-and-paste the genie's interface URL into their application and use Genie's simple API calls. The mess and complexity of machine intelligence has been distilled to five API calls.

Once created, a Genie is made deployable by putting it in a “bottle”, enabling integration and interaction through its native API or its web interface. Bottles are robust environments with powerful built-in components: backtesting for validation, self-monitoring analytics to characterize and improve data quality, system health monitoring, and more. Future releases will add even more runtime facilities.

Bottles also provide a facility for users to upload existing data to the cloud to test, train, or refine Genie behavior, and a “Live Interface” to interact with the running Genie and query it using its web interface. If desired, its output behavior can be refined or trained with rewards and punishment. Training can be pre-deployment or inherent in post-deployment operation, so that
the Genie learns as it is used.

With Genie’s standardized inputs and outputs, genies can be mixed and mashed with other genies to create more robust and more complex solutions. All genies are compatible with all other genies, allowing crowd-sourced or “swarm” solutions.