Through the last few years, with the media attention and the volume of discussion in public forums, you would think a common definition of “what is AI” wouldn’t be difficult.  It turns out that as many people as you can ask to define artificial intelligence, you’ll get at least that many different responses.  The diversity of AI techniques that continue to evolve offers a wealth of opportunities to firms of every size.  The media has drawn attention to learning-based AI in recent years, and as a result many people consider machine learning to be the birthplace of AI.  This assumption would be incorrect as learning AI simply reflects the most visible field of advancement at present, in part because of advances in computing capacity and availability of data for some visible use cases.  Machine learning and the fields that derive from it in the horizontal including deep learning, supervised and unsupervised learning, reinforcement learning, and transfer learning, as well as hybrid horizontal / vertical applications such as vision and speech, have drawn the bulk of media attention.  While machine learning draws media attention, the commercially successful applications of AI have unified past fields of innovation with advances in learning to create vertical solutions with consumer and business appeal.

The AI Hub uses the following three definitions to describe artificial intelligence and related topics.

Artificial Intelligence

Artificial intelligence is a term that represents a system or set of business processes that can sense its operating environment, reason, learn, and act in response to input and objectives.  An observer of the behaviour would consider the results intelligent if a human performed the action.

Narrow AI

Narrow artificial intelligence is the application of techniques that create the impression of intelligence within an environment with limited variables, conditions and outcomes.  Siri or another voice assistant is an example of a narrow AI, as is a spam detector in email.

The AI Effect

When a problem becomes computationally possible, it becomes ‘less AI’ and more ‘regular practice.’  An autopilot system in an airliner remains an implementation of AI regardless of its origins in the early days of aviation.  An automobile airbag is dependent on Symbolic AI to determine its state, and regardless of the proliferation of websites with cheap flight booking offers, airline and freight logistics remains an active field of research in AI.