Wednesday, July 20, 2016

Using Artificial Intelligence In Business

Comment:  This brief was originally written in Oct 2009 and was updated then posted again on December 21, 2013. Many movies and books have dramatized the use of Artificial Intelligence, AI, delivering social messages of various sorts. However, the use AI methods and techniques are often overlooked. This post will discuss AI basics. I may expand this post to a series on AI since the world has entered a time that involves increasing use of AI as well as biological and quantum computing among other emerging technologies. My estimation is that many technologies will be converging rapidly in the near term. Ray Kurzweil, famed author of the book "The Age of Spiritual Machines", speaks to a technological singularity that he believes will occur in the next 10 years. The technological singularity he discusses is the convergence of principally three technologies that result in machines and/or  humanoids having greater-than-human intelligence. That technology field is call trans-humanism, Image 1.


Image 1: Trans-humanism Symbol
Artificial Intelligence In Business
by 
JT Bogden, PMP

When most people think of AI they think of an all-knowing computer system. However, realistic designs must be formally constrained or have boundary limits set due to processing power, memory, and temporal limitations in traditional computational systems.  Also, decision-making is often multi-layered having some sort of fail-safe or fall back. Hence, there is a simplicity to the design that is wrapped around a Theory of Mind, ToM, which is a humanization of the system. Ultimately, the system interacts and behaves in an intelligent human-centric manner.

An AI engine, Figure 1, is object oriented and receives inputs from its environment or other sources such as a database of the project or operational data. Any AI engine is limited in the ability to make inferences by the information delivered to the inputs from sensory devices and other input equipage. Thus, as part of the design that constraints the data requirements need to be carefully considered.
Figure 1:  AI Engine Model
In coding the elements of the engine a concept of the finite-state machine, FSM, is used as an organizational tool to break the problem set into manageable sub-problems. An FSM has a data structure reflective of the states inherent to the machine, input conditions, and transition functions. Another type of machine used is the Fuzzy-State machine, FuSM's, that handle partial truths. FuSM's do not maintain a defined state but instead compute activation levels and the overall state is determined by the combination of activated states. Skeletal code wraps the object classes of FSM's and FuSM's into a management system unique to the engine. FSM's and FuSM's can be a message, event, data, or inertial driven. State machines yield flexibility and scalability to the intelligence system and are more complex than this brief paragraph.

Another framework is an algorithmic model. In the computational theory of the universe, the natural universe is an irreversible algorithmic computation reflecting times arrow. Algorithmic approaches are useful  for rule and probabilistic based processing having limited sets of outcomes such as the rolling of a dice. There are only 6 possible outcomes using 1 dice. The algorithm is a probabilistic outcome of the roll that one of 6 sides will show. 

The ultimate AI framework is the neural network (a neural net). Natural neural nets are  the structure of animal and human brains. AI reflects the brain activity using input, hidden, and output nodes to process information. Input nodes gather information with some basic processing. Thus, an input node can be some sort of sensory device or even a neugent or agent of some sort. Hidden networked nodes do more advanced processing routing the processed information forming line or pathways of logic. Output nodes usually format the processed information or result in some sort of action. Output nodes can be displays, motors, servers, actuators, voice, or a host of other devices. A node is said to fire when inputs match the states and circumstances for processing. Outputs from several nodes could be the inputs to another node or set of nodes. The complex pathways through a neural net that result in an output are a line of logic.  A valid line of logic is said to be a truth. Whereas, an invalid line of logic is considered to be untrue.  The Neural net  has their place in complex real-time problem solving situations as the nodal network conducts parallel processing while evaluating multiple outcomes simultaneously.

Using these techniques in an operation to compute outcomes and discern knowledge is exceptionally useful in marketplace competition.  Using a complex adaptive architecture the organizational structure can form a neural net processing knowledge and lines of logic if properly designed. In this case, the World Wide Web becomes a medium for processing lines of logic across geospatially dispersed nodes. Some nodes can be human supervised while other nodes are automated. In essence, the entire organization becomes a brain exhibiting life-like qualities.

AI constructs can be employed in projects to monitor for emerging outcomes and assist by algorithmically computing outcomes then correcting for issues before they emerge. Data feeds into the AI engine can monitor schedules and resources as well as change requests for adherence to the project portfolio. AI is particularly useful in identifying emerging issues or partial truths before they become real issues.

AI will not become fully useful to organizations and business until they have evolved and formed structures supportive of AI.  If organizations adopt constructs that reflect the natural environment rather than a human imposed structures only then will they be able to adapt to change and responsibly leverage emerging conditions with greater simplicity.