Building intelligent machines with Ontologies

In this chapter, we have looked at the role of Ontology in the management of big data assets as knowledge repositories, and understood the need for computational systems to perceive the data as things instead of strings. Although some of the big systems and web search engines use a semantic world view, the adoption of Ontology as a basis for systems is slow. The custodians of data assets (governments and everyone else) need to model knowledge assets in a consistent and standardized manner in order for us to evolve current computational systems into intelligent systems.

Let us consider a use case that leverages Ontology-based knowledge graphs in order to simplify the flight boarding process. We have all experienced a hugely manual and time-consuming process when boarding a flight. From the time we enter the airport to the time we board the flight, we go through a number of security checks and experience document verification. In a connected world where all the knowledge assets are standardized and defined as domain-specific Ontologies, it is possible to develop intelligent agents to make the flight boarding process hassle free and seamless.

Let us define the generic characteristics of an intelligent agent:

A little expansion on the characteristics is as follows:

• Goals: Every intelligent system should have a well defined set of goals. These goals govern the rational decisions taken by the intelligent system and drive actions and hence results. For example, in the case of an intelligent agent that is responsible for the flight boarding process, one of the goals is to restrict access to anyone who does not pass all security checks, even if the person has a valid air ticket. In defining the goals for intelligent agents, one of the prime considerations should be that the AI agent or systems should complement and augment human capabilities.
• Environment: The intelligent agent should operate within the context of the environment. Its decisions and actions cannot be independent of the context. In our example use case, the environment is the airport, the passenger gates, flight schedules, and so on. The agents perceive the environment with various sensors, for example video cameras.
• Data Assets: The intelligent agent needs access to historical data in terms of the domain and the context in which it operates. The data assets can be available locally and globally (internet endpoints). These data assets ideally should be defined as RDF schema structures with standardized representations and protocols. These data assets should be queryable with standard languages and protocols (SPARQL) in order to ensure maximum interoperability.

• Model: This is where the real intelligence of the agent is available as algorithms and learning systems. These models evolve continuously based on the context, historical decisions, actions, and results. As a general rule, the model should perform better (more accurately) over a period of time for similar contextual inputs.
• Effectors: These are the tangible aspects of the agent which facilitate actions. In the example of an airline passenger boarding agent, the effector can be an automated gate opening system which opens a gate once all the passengers are fully validated (having a valid ticket, identity, and no security check failures). The external world perceives the intelligent agent through effectors.
• Actions and Results: Based on the environmental context, the data assets, and the trained models, the intelligent agent makes decisions that trigger actions through the effectors. These actions provide results based on the rationality of the decision and accuracy of the trained model. The results are once again fed into model training in order to improve accuracy over a period of time.

At a high level, the method of the intelligent agent, which facilitates the flight boarding process, can be depicted as follows:

1. When a passenger walks into the airport, a video camera reads the image and matches it to the data assets available to the agent. These data assets are Ontology objects which are loosely coupled and have flexibility of structure and attributes. Some of the inferences are made at the first level of matching to correctly identify the person who has entered the airport. 
2. If the person cannot be identified with the video stream, the first airport gate does not open automatically and requires a fingerprint scan from the passenger. The fingerprint scan is validated against the dataset, which is once again an Ontology object representation of the person entity. If the person is not identified at this stage, they are flagged for further manual security procedures. 
3. Once the person is correctly identified, the agent scans the global active ticket directory in order to ensure that the person has a valid ticket for a flight that departs from the airport in a reasonable time window. The global ticket directory and the flight database is also available as Ontology objects for the agent to refer to in real time.
4. Once ticket validity is ensured, a boarding pass is generated and delivered to the passenger's smartphone, once again by referring to the person Ontology to derive personal details in a secure manner. The real-time instructions for directions to the gate are also sent to the device. 

The agent can seamlessly guide the passenger to the appropriate boarding gate. The system can be built easily once all the heterogeneous data sources are standardized and have Ontological representation, which facilitates maximum interoperability and eliminates a need to code perse knowledge representations. This results in an overall reduction of complexity in the agent software and an increase in efficiency.